Authors: Taaha Saleem Bajwa
Abstract: LLMs are predominantly trained on English data, which leads to a significant drop in performance on low-resource languages. Understanding how LLMs handle these languages is crucial for improving their effectiveness. This study focuses on Urdu as a use case for exploring the challenges faced by LLMs in processing low-resource languages. LLMs primarily reason in English when prompted in another language, with the final layers acting as translators to convert the English response into the target language. This study finds that even for low-resource languages, the internal latent response of LLMs in English is quite coherent; however, the translation features are lossy and result in poor translations, leading to reduced performance. By mechanistically removing these translation features and using a separate translation model to translate the internal latent response of LLM, the performance of LLMs improves significantly while also preserving the cultural nuances of the input in low-resource languages.
Authors: Malak Mohamed, Rokaia Emad, Ali Hamdi
Abstract: Social telehealth has revolutionized healthcare by enabling patients to share symptoms and receive medical consultations remotely. Users frequently post symptoms on social media and online health platforms, generating a vast repository of medical data that can be leveraged for disease classification and symptom severity assessment. Large language models (LLMs), such as LLAMA3, GPT-3.5 Turbo, and BERT, process complex medical data to enhance disease classification. This study explores three Arabic medical text preprocessing techniques: text summarization, text refinement, and Named Entity Recognition (NER). Evaluating CAMeL-BERT, AraBERT, and Asafaya-BERT with LoRA, the best performance was achieved using CAMeL-BERT with NER-augmented text (83% type classification, 69% severity assessment). Non-fine-tuned models performed poorly (13%-20% type classification, 40%-49% severity assessment). Integrating LLMs into social telehealth systems enhances diagnostic accuracy and treatment outcomes.
Authors: Qizhen Zhang, Prajjwal Bhargava, Chloe Bi, Chris X. Cai, Jakob Foerster, Jeremy Fu, Punit Singh Koura, Ruan Silva, Sheng Shen, Emily Dinan, Suchin Gururangan, Mike Lewis
Abstract: We present Branch-Train-Stitch (BTS), an efficient and flexible training algorithm for combining independently trained large language model (LLM) experts into a single, capable generalist model. Following Li et al., we start with a single seed language model which is branched into domain-specific (e.g., coding or math) experts with continual pretraining. BTS combines experts into a generalist model using lightweight stitch layers, which are inserted between frozen experts and the seed LLM, and trained on a small datamix of the expert domains. Stitch layers enable the seed LLM to integrate representations from any number of experts during the forward pass, allowing it to generalize to new domains, despite remaining frozen. Because BTS does not alter the constituent LLMs, BTS provides a modular and flexible approach: experts can be easily removed and new experts can be added with only a small amount of training. Compared to alternative model merging approaches, BTS yields the best generalist performance on a variety of downstream tasks, retaining the specialized capabilities of each of the experts.
Authors: Brian J Chan, Jui-Hung Cheng, Mao Xun Huang, Chao-Ting Chen, Hen-Hsen Huang
Abstract: In Transformer-based sequence-to-sequence generation, beam search has proven effective in enhancing the quality of generated sequences compared to greedy decoding. Conventional beam search methods typically adopt either a sequential or batch-based approach. The sequential approach, while memory-efficient, requires multiple decoding passes to construct a complete search tree, leading to significantly slower inference. On the other hand, the batch-based approach enables parallel computation across beams, but at the expense of high memory consumption due to the need to maintain separate key-value (KV) caches for each beam. In this study, we introduce a novel trie (prefix-tree)-based parallel decoding method that addresses the memory inefficiency of batch-based beam search. By sharing a single KV cache among all beams that share the same prefix, the proposed method not only reduces memory consumption dramatically but also enables parallel decoding across all branches. This innovative use of a prefix tree offers an efficient alternative for beam search, achieving significant memory savings while preserving inference speed, making it particularly well-suited for memory-constrained environments or large-scale model deployments.
Authors: Yinghao Li, Vianne Gao, Chao Zhang, MohamadAli Torkamani
Abstract: The training and fine-tuning of large language models (LLMs) often involve diverse textual data from multiple sources, which poses challenges due to conflicting gradient directions, hindering optimization and specialization. These challenges can undermine model generalization across tasks, resulting in reduced downstream performance. Recent research suggests that fine-tuning LLMs on carefully selected, task-specific subsets of data can match or even surpass the performance of using the entire dataset. Building on these insights, we propose the Ensembles of Low-Rank Expert Adapters (ELREA) framework to improve the model's capability to handle diverse tasks. ELREA clusters the training instructions based on their gradient directions, representing different areas of expertise and thereby reducing conflicts during optimization. Expert adapters are then trained on these clusters, utilizing the low-rank adaptation (LoRA) technique to ensure training efficiency and model scalability. During inference, ELREA combines predictions from the most relevant expert adapters based on the input data's gradient similarity to the training clusters, ensuring optimal adapter selection for each task. Experiments show that our method outperforms baseline LoRA adapters trained on the full dataset and other ensemble approaches with similar training and inference complexity across a range of domain-specific tasks.
Authors: Logan Born, M. Willis Monroe, Kathryn Kelley, Anoop Sarkar
Abstract: A numeration system encodes abstract numeric quantities as concrete strings of written characters. The numeration systems used by modern scripts tend to be precise and unambiguous, but this was not so for the ancient and partially-deciphered proto-Elamite (PE) script, where written numerals can have up to four distinct readings depending on the system that is used to read them. We consider the task of disambiguating between these readings in order to determine the values of the numeric quantities recorded in this corpus. We algorithmically extract a list of possible readings for each PE numeral notation, and contribute two disambiguation techniques based on structural properties of the original documents and classifiers learned with the bootstrapping algorithm. We also contribute a test set for evaluating disambiguation techniques, as well as a novel approach to cautious rule selection for bootstrapped classifiers. Our analysis confirms existing intuitions about this script and reveals previously-unknown correlations between tablet content and numeral magnitude. This work is crucial to understanding and deciphering PE, as the corpus is heavily accounting-focused and contains many more numeric tokens than tokens of text.
Authors: Daniel Pluth, Yu Zhou, Vijay K. Gurbani
Abstract: Recent advances in explainable machine learning have highlighted the potential of sparse autoencoders in uncovering mono-semantic features in densely encoded embeddings. While most research has focused on Large Language Model (LLM) embeddings, the applicability of this technique to other domains remains largely unexplored. This study applies sparse autoencoders to speaker embeddings generated from a Titanet model, demonstrating the effectiveness of this technique in extracting mono-semantic features from non-textual embedded data. The results show that the extracted features exhibit characteristics similar to those found in LLM embeddings, including feature splitting and steering. The analysis reveals that the autoencoder can identify and manipulate features such as language and music, which are not evident in the original embedding. The findings suggest that sparse autoencoders can be a valuable tool for understanding and interpreting embedded data in many domains, including audio-based speaker recognition.
Authors: Edward Y. Chang
Abstract: This paper introduces a three-branch checks-and-balances framework for ethical alignment of Large Language Models (LLMs), inspired by governmental systems. It implements three independent yet interacting components: LLMs as the executive branch for knowledge generation, DIKE as the legislative branch establishing ethical guardrails, and ERIS as the judicial branch for contextual interpretation. The adversarial DIKE-ERIS duality enables adaptation to diverse cultural contexts while upholding consistent ethical principles. This architecture addresses limitations of reinforcement learning with human feedback (RLHF) by providing interpretable, adaptable, and culturally-aware ethical reasoning. Through self-supervised learning and adversarial testing, our framework demonstrates how emotional modeling can guide linguistic behaviors toward ethical outcomes while preserving independence across knowledge generation, ethical oversight, and contextual interpretation.
Authors: Binchi Zhang, Zhengzhang Chen, Zaiyi Zheng, Jundong Li, Haifeng Chen
Abstract: Large Language Models (LLMs) excel in natural language processing by encoding extensive human knowledge, but their utility relies on timely updates as knowledge evolves. Updating LLMs involves two key tasks simultaneously: unlearning to remove unwanted knowledge and editing to incorporate new information. Existing methods face two major challenges: ineffective knowledge storage (either too sparse or too dense) and task conflicts between editing and unlearning, as validated through our theoretical and experimental results. To address these issues, we propose LOKA, a conflict-free framework for LLM updating based on a knowledge codebook. During training, updated knowledge is stored in multiple codebook memories. To optimize knowledge storage, a similarity-aware knowledge mapping ensures that related knowledge pieces are clustered and allocated to the same memory. Additionally, LOKA resolves task conflicts by employing task-specific and multi-task memories guided by a conflict score. In the inference stage, LOKA retrieves the most relevant memory from the codebook and plugs it into the original LLM to apply the updated knowledge. A learning-based router controls codebook activation to further improve knowledge utilization. Extensive experiments demonstrate the effectiveness of LOKA in LLM knowledge updating tasks.
Authors: Larin Tonix, Morgana Baskerville, Nathaniel Stourton, Ophelia Tattershall
Abstract: Handling long-range dependencies in neural architectures has remained a persistent challenge due to computational limitations and inefficient contextual retention mechanisms. Tensorial operations have provided a foundation for restructuring model representations, yet conventional architectures have struggled to incorporate such techniques without introducing excessive complexity. A novel approach, Context-Preserving Tensorial Reconfiguration (CPTR), enables dynamic reorganization of weight tensors through structured factorization and adaptive contraction, allowing for enhanced contextual integration without substantial computational overhead. Empirical evaluations demonstrate that CPTR improves coherence retention across extended sequences, leading to measurable reductions in perplexity and improved recall accuracy for long-context tasks. Performance comparisons reveal that CPTR-enhanced models exhibit greater computational efficiency and reduced memory consumption while maintaining competitive language generation fluency and accuracy. Gradient stability metrics further validate the improved training efficiency, revealing more controlled variance in weight updates. Comparative studies across baseline and CPTR-enhanced models confirm that tensorial reconfiguration contributes to more stable and computationally efficient language modeling. The findings support the potential of CPTR in refining contemporary neural architectures for tasks requiring long-range contextual understanding and efficient memory utilization.
Authors: Fei Yu, Yingru Li, Benyou Wang
Abstract: Large language models (LLMs) struggle with multi-step reasoning, where inference-time scaling has emerged as a promising strategy for performance improvement. Verifier-guided search outperforms repeated sampling when sample size is limited by selecting and prioritizing valid reasoning paths. However, we identify a critical limitation: scaling flaws, prevalent across different models (Mistral 7B and DeepSeekMath 7B), benchmarks (GSM8K and MATH), and verifiers (outcome value models and process reward models). As sample size increases, verifier-guided search exhibits diminishing advantages and eventually underperforms repeated sampling. Our analysis attributes this to verifier failures, where imperfect verifiers misrank candidates and erroneously prune all valid paths. These issues are further exacerbated in challenging and out-of-distribution problems, restricting search effectiveness. To mitigate verifier failures, we explore reducing reliance on verifiers and conduct preliminary investigations using two simple methods. Our findings reveal fundamental limitations in verifier-guided search and suggest future directions.
Authors: Huan Ma, Jingdong Chen, Guangyu Wang, Changqing Zhang
Abstract: In recent years, Large Language Models (LLMs) have seen remarkable advancements and have been extensively integrated across various fields. Despite their progress, LLMs are prone to hallucinations, producing responses that may not be dependable if the models lack sufficient grounding knowledge. To mitigate this issue, methods for estimating uncertainty have been adopted, with a focus on critical tokens as indicators of reliability. Nevertheless, probability-based approaches have shown limitations in assessing token-level reliability due to the erosion of evidence strength information acquired during training. In this paper, we introduce Logits-induced Token Uncertainty (LogU), a novel framework designed to estimate token-specific uncertainty in LLMs in real time, without the need for multiple sampling rounds. By leveraging evidence modeling for the implementation of LogU, we utilize the derived uncertainty measures to steer downstream tasks. Our experimental findings highlight the substantial effectiveness and potential of LogU, marking a significant advancement in addressing the challenge of model hallucinations.
Authors: Xiang Liu, Zhenheng Tang, Peijie Dong, Zeyu Li, Bo Li, Xuming Hu, Xiaowen Chu
Abstract: To reduce memory costs in long-context inference with Large Language Models (LLMs), many recent works focus on compressing the key-value (KV) cache of different tokens. However, we identify that the previous KV cache compression methods measure token importance individually, neglecting the dependency between different tokens in the real-world language characterics. In light of this, we introduce ChunkKV, grouping the tokens in a chunk as a basic compressing unit, and retaining the most informative semantic chunks while discarding the less important ones. Furthermore, observing that ChunkKV exhibits higher similarity in the preserved indices across different layers, we propose layer-wise index reuse to further reduce computational overhead. We evaluated ChunkKV on cutting-edge long-context benchmarks including LongBench and Needle-In-A-HayStack, as well as the GSM8K and JailbreakV in-context learning benchmark. Our experiments with instruction tuning and multi-step reasoning (O1 and R1) LLMs, achieve up to 10\% performance improvement under aggressive compression ratios compared to existing methods.
Authors: Alistair Dombrowski, Beatrix Engelhardt, Dimitri Fairbrother, Henry Evidail
Abstract: Token representations influence the efficiency and adaptability of language models, yet conventional tokenization strategies impose rigid segmentation boundaries that do not adjust dynamically to evolving contextual relationships. The introduction of contextual morphogenesis establishes a self-organizing mechanism that restructures token boundaries based on learned contextual dependencies, allowing embeddings to evolve progressively across iterative processing steps. Empirical evaluations demonstrate that dynamically adjusted tokenization contributes to reductions in perplexity while maintaining representational stability, particularly in linguistically complex domains where static segmentation fails to capture nuanced dependencies. Computational trade-offs associated with self-organizing token structures indicate that additional processing overhead remains within feasible limits, provided that optimization strategies account for segmentation update efficiency. Comparative assessments across different linguistic corpora suggest that adaptive tokenization preserves interpretability while improving alignment with contextual cues, reinforcing the potential of morphogenetic segmentation mechanisms to refine predictive accuracy. Stability analyses confirm that evolving token structures maintain consistent segmentation behaviors across varied text distributions, ensuring that representational adaptations remain linguistically coherent. The effectiveness of contextual morphogenesis in refining structural stability and predictive performance highlights its viability as an alternative to traditional tokenization methods. Further analysis of computational efficiency considerations suggests that hybrid strategies integrating both static and dynamic segmentation techniques may offer a balanced approach to optimizing representational flexibility while maintaining inference efficiency.
Authors: Jiaxin Guo, C. L. Philip Chen, Shuzhen Li, Tong Zhang
Abstract: Cold-start active learning (CSAL) selects valuable instances from an unlabeled dataset for manual annotation. It provides high-quality data at a low annotation cost for label-scarce text classification. However, existing CSAL methods overlook weak classes and hard representative examples, resulting in biased learning. To address these issues, this paper proposes a novel dual-diversity enhancing and uncertainty-aware (DEUCE) framework for CSAL. Specifically, DEUCE leverages a pretrained language model (PLM) to efficiently extract textual representations, class predictions, and predictive uncertainty. Then, it constructs a Dual-Neighbor Graph (DNG) to combine information on both textual diversity and class diversity, ensuring a balanced data distribution. It further propagates uncertainty information via density-based clustering to select hard representative instances. DEUCE performs well in selecting class-balanced and hard representative data by dual-diversity and informativeness. Experiments on six NLP datasets demonstrate the superiority and efficiency of DEUCE.
Authors: Nishant Balepur, Alexa Siu, Nedim Lipka, Franck Dernoncourt, Tong Sun, Jordan Boyd-Graber, Puneet Mathur
Abstract: Query-focused summarization (QFS) gives a summary of documents to answer a query. Past QFS work assumes queries have one answer, ignoring debatable ones (Is law school worth it?). We introduce Debatable QFS (DQFS), a task to create summaries that answer debatable queries via documents with opposing perspectives; summaries must comprehensively cover all sources and balance perspectives, favoring no side. These goals elude LLM QFS systems, which: 1) lack structured content plans, failing to guide LLMs to write balanced summaries, and 2) use the same query to retrieve contexts across documents, failing to cover all perspectives specific to each document's content. To overcome this, we design MODS, a multi-LLM framework mirroring human panel discussions. MODS treats documents as individual Speaker LLMs and has a Moderator LLM that picks speakers to respond to tailored queries for planned topics. Speakers use tailored queries to retrieve relevant contexts from their documents and supply perspectives, which are tracked in a rich outline, yielding a content plan to guide the final summary. Experiments on ConflictingQA with controversial web queries and DebateQFS, our new dataset of debate queries from Debatepedia, show MODS beats SOTA by 38-59% in topic paragraph coverage and balance, based on new citation metrics. Users also find MODS's summaries to be readable and more balanced.
Authors: Xin Xu, Qiyun Xu, Tong Xiao, Tianhao Chen, Yuchen Yan, Jiaxin Zhang, Shizhe Diao, Can Yang, Yang Wang
Abstract: Large language models (LLMs) have demonstrated remarkable capabilities in solving complex reasoning tasks, particularly in mathematics. However, the domain of physics reasoning presents unique challenges that have received significantly less attention. Existing benchmarks often fall short in evaluating LLMs' abilities on the breadth and depth of undergraduate-level physics, underscoring the need for a comprehensive evaluation. To fill this gap, we introduce UGPhysics, a large-scale and comprehensive benchmark specifically designed to evaluate UnderGraduate-level Physics (UGPhysics) reasoning with LLMs. UGPhysics includes 5,520 undergraduate-level physics problems in both English and Chinese, covering 13 subjects with seven different answer types and four distinct physics reasoning skills, all rigorously screened for data leakage. Additionally, we develop a Model-Assistant Rule-based Judgment (MARJ) pipeline specifically tailored for assessing answer correctness of physics problems, ensuring accurate evaluation. Our evaluation of 31 leading LLMs shows that the highest overall accuracy, 49.8% (achieved by OpenAI-o1-mini), emphasizes the necessity for models with stronger physics reasoning skills, beyond math abilities. We hope UGPhysics, along with MARJ, will drive future advancements in AI for physics reasoning.
Authors: Jingyuan Yi, Zeqiu Xu, Tianyi Huang, Peiyang Yu
Abstract: The pervasiveness of the dissemination of fake news through social media platforms poses critical risks to the trust of the general public, societal stability, and democratic institutions. This challenge calls for novel methodologies in detection, which can keep pace with the dynamic and multi-modal nature of misinformation. Recent works include powering the detection using large language model advances in multimodal frameworks, methodologies using graphs, and adversarial training in the literature of fake news. Based on the different approaches which can bring success, some key highlights will be underlined: enhanced LLM-improves accuracy through more advanced semantics and cross-modality fusion for robust detections. The review further identifies critical gaps in adaptability to dynamic social media trends, real-time, and cross-platform detection capabilities, as well as the ethical challenges thrown up by the misuse of LLMs. Future directions underline the development of style-agnostic models, cross-lingual detection frameworks, and robust policies with a view to mitigating LLM-driven misinformation. This synthesis thus lays a concrete foundation for those researchers and practitioners committed to reinforcing fake news detection systems with complications that keep on growing in the digital landscape.
Authors: Kosei Kobayashi, Kosuke Matsuzaki, Masaya Taniguchi, Keisuke Sakaguchi, Kentaro Inui, Kentaro Abe
Abstract: The long-range dependencies among the tokens, which originate from hierarchical structures, are a defining hallmark of human language. However, whether similar dependencies exist within the sequential vocalization of non-human animals remains a topic of investigation. Transformer architectures, known for their ability to model long-range dependencies among tokens, provide a powerful tool for investigating this phenomenon. In this study, we employed the Transformer architecture to analyze the songs of Bengalese finch (Lonchura striata domestica), which are characterized by their highly variable and complex syllable sequences. To this end, we developed FinchGPT, a Transformer-based model trained on a textualized corpus of birdsongs, which outperformed other architecture models in this domain. Attention weight analysis revealed that FinchGPT effectively captures long-range dependencies within syllables sequences. Furthermore, reverse engineering approaches demonstrated the impact of computational and biological manipulations on its performance: restricting FinchGPT's attention span and disrupting birdsong syntax through the ablation of specific brain nuclei markedly influenced the model's outputs. Our study highlights the transformative potential of large language models (LLMs) in deciphering the complexities of animal vocalizations, offering a novel framework for exploring the structural properties of non-human communication systems while shedding light on the computational distinctions between biological brains and artificial neural networks.
Authors: Anna Min, Chenxu Hu, Yi Ren, Hang Zhao
Abstract: Current research in speech-to-speech translation (S2ST) primarily concentrates on translation accuracy and speech naturalness, often overlooking key elements like paralinguistic information, which is essential for conveying emotions and attitudes in communication. To address this, our research introduces a novel, carefully curated multilingual dataset from various movie audio tracks. Each dataset pair is precisely matched for paralinguistic information and duration. We enhance this by integrating multiple prosody transfer techniques, aiming for translations that are accurate, natural-sounding, and rich in paralinguistic details. Our experimental results confirm that our model retains more paralinguistic information from the source speech while maintaining high standards of translation accuracy and naturalness.
Authors: Anna Min, Chenxu Hu, Yi Ren, Hang Zhao
Abstract: Though end-to-end speech-to-text translation has been a great success, we argue that the cascaded speech-to-text translation model still has its place, which is usually criticized for the error propagation between automatic speech recognition (ASR) and machine translation (MT) models. In this paper, we explore the benefits of incorporating multiple candidates from ASR and self-supervised speech features into MT. Our analysis reveals that the primary cause of cascading errors stems from the increased divergence between similar samples in the speech domain when mapped to the text domain. By including multiple candidates and self-supervised speech features, our approach allows the machine translation model to choose the right words and ensure precise translation using various speech samples. This strategy minimizes error spread and takes advantage of large ASR and MT datasets, along with pre-trained ASR/MT models, while addressing associated issues.
Authors: Stefano Civelli, Pietro Bernardelle, Gianluca Demartini
Abstract: While pretraining language models with politically diverse content has been shown to improve downstream task fairness, such approaches require significant computational resources often inaccessible to many researchers and organizations. Recent work has established that persona-based prompting can introduce political diversity in model outputs without additional training. However, it remains unclear whether such prompting strategies can achieve results comparable to political pretraining for downstream tasks. We investigate this question using persona-based prompting strategies in multimodal hate-speech detection tasks, specifically focusing on hate speech in memes. Our analysis reveals that when mapping personas onto a political compass and measuring persona agreement, inherent political positioning has surprisingly little correlation with classification decisions. Notably, this lack of correlation persists even when personas are explicitly injected with stronger ideological descriptors. Our findings suggest that while LLMs can exhibit political biases in their responses to direct political questions, these biases may have less impact on practical classification tasks than previously assumed. This raises important questions about the necessity of computationally expensive political pretraining for achieving fair performance in downstream tasks.
Authors: Hasin Jawad Ali, Ajwad Abrar, S. M. Hozaifa Hossain, M. Firoz Mridha
Abstract: In politically sensitive scenarios like wars, social media serves as a platform for polarized discourse and expressions of strong ideological stances. While prior studies have explored ideological stance detection in general contexts, limited attention has been given to conflict-specific settings. This study addresses this gap by analyzing 9,969 Reddit comments related to the Israel-Palestine conflict, collected between October 2023 and August 2024. The comments were categorized into three stance classes: Pro-Israel, Pro-Palestine, and Neutral. Various approaches, including machine learning, pre-trained language models, neural networks, and prompt engineering strategies for open source large language models (LLMs), were employed to classify these stances. Performance was assessed using metrics such as accuracy, precision, recall, and F1-score. Among the tested methods, the Scoring and Reflective Re-read prompt in Mixtral 8x7B demonstrated the highest performance across all metrics. This study provides comparative insights into the effectiveness of different models for detecting ideological stances in highly polarized social media contexts. The dataset used in this research is publicly available for further exploration and validation.
Authors: Turi Abu, Ying Shi, Thomas Fang Zheng, Dong Wang
Abstract: We present a novel Automatic Speech Recognition (ASR) dataset for the Oromo language, a widely spoken language in Ethiopia and neighboring regions. The dataset was collected through a crowd-sourcing initiative, encompassing a diverse range of speakers and phonetic variations. It consists of 100 hours of real-world audio recordings paired with transcriptions, covering read speech in both clean and noisy environments. This dataset addresses the critical need for ASR resources for the Oromo language which is underrepresented. To show its applicability for the ASR task, we conducted experiments using the Conformer model, achieving a Word Error Rate (WER) of 15.32% with hybrid CTC and AED loss and WER of 18.74% with pure CTC loss. Additionally, fine-tuning the Whisper model resulted in a significantly improved WER of 10.82%. These results establish baselines for Oromo ASR, highlighting both the challenges and the potential for improving ASR performance in Oromo. The dataset is publicly available at https://github.com/turinaf/sagalee and we encourage its use for further research and development in Oromo speech processing.
Authors: Yizhe Xiong, Wei Huang, Xin Ye, Hui Chen, Zijia Lin, Haoran Lian, Zhenpeng Su, Jungong Han, Guiguang Ding
Abstract: Post-training is essential for adapting Large Language Models (LLMs) to real-world applications. Deploying post-trained models faces significant challenges due to substantial memory overhead and noticeable inference latency. Existing work has identified significant redundancies in LLMs and proposed efficient architectures, namely intra-layer KV sharing and cross-layer KV sharing. However, intra-layer KV sharing still results in high inference costs, while cross-layer KV sharing leads to significant performance degradation. As a result, both methods remain suboptimal for post-training pre-trained LLMs. In this paper, we identify that the \texttt{Softmax} operation is a primary bottleneck for LLM inference and discover that it is actually highly redundant during post-training. We propose Softmax \textbf{Uni}fication in \textbf{Att}e\textbf{n}tion (\textbf{UniAttn}), a novel post-training method that unifies Softmax activations across transformer blocks to reduce LLM inference costs. Additionally, UniAttn adopts a linear projection to compensate for the errors induced by Softmax unification. Experiments show that UniAttn matches the performance of standard post-training while significantly reducing inference costs, outperforming existing efficient architectures during post-training. Our code will be available at \url{https://github.com/Bostoncake/UniAttn}.
Authors: Taiji Li, Hao Chen, Fei Yu, Yin Zhang
Abstract: Despite the rapid growth of context length of large language models (LLMs) , LLMs still perform poorly in long document summarization. An important reason for this is that relevant information about an event is scattered throughout long documents, and the messy narrative order impairs the accurate understanding and utilization of LLMs for long documents. To address these issues, we propose a novel summary generation framework, called HERA. Specifically, we first segment a long document by its semantic structure and retrieve text segments about the same event, and finally reorder them to form the input context. We evaluate our approach on two long document summarization datasets. The experimental results show that HERA outperforms foundation models in ROUGE, BERTScore and faithfulness metrics, while HERA does not require additional fine-tuning and resources.
Authors: Aishik Mandal, Tanmoy Chakraborty, Iryna Gurevych
Abstract: Mental illness is a widespread and debilitating condition with substantial societal and personal costs. Traditional diagnostic and treatment approaches, such as self-reported questionnaires and psychotherapy sessions, often impose significant burdens on both patients and clinicians, limiting accessibility and efficiency. Recent advances in Artificial Intelligence (AI), particularly in Natural Language Processing and multimodal techniques, hold great potential for recognizing and addressing conditions such as depression, anxiety, bipolar disorder, schizophrenia, and post-traumatic stress disorder. However, privacy concerns, including the risk of sensitive data leakage from datasets and trained models, remain a critical barrier to deploying these AI systems in real-world clinical settings. These challenges are amplified in multimodal methods, where personal identifiers such as voice and facial data can be misused. This paper presents a critical and comprehensive study of the privacy challenges associated with developing and deploying AI models for mental health. We further prescribe potential solutions, including data anonymization, synthetic data generation, and privacy-preserving model training, to strengthen privacy safeguards in practical applications. Additionally, we discuss evaluation frameworks to assess the privacy-utility trade-offs in these approaches. By addressing these challenges, our work aims to advance the development of reliable, privacy-aware AI tools to support clinical decision-making and improve mental health outcomes.
Authors: Yi Liu
Abstract: To address the issue of variability in the output generated by a language model, we present a measure of semantic variability that is statistically consistent under mild assumptions. This measure, denoted as semantic spectral entropy, is a easy to implement algorithm that requires just off the shelf language models. We put very few restrictions on the language models and we have shown in a clear simulation studies that such method can generate accurate metric despite randomness that arise from the language models.
Authors: Md Mehrab Tanjim, Xiang Chen, Victor S. Bursztyn, Uttaran Bhattacharya, Tung Mai, Vaishnavi Muppala, Akash Maharaj, Saayan Mitra, Eunyee Koh, Yunyao Li, Ken Russell
Abstract: Multi-turn conversations with an Enterprise AI Assistant can be challenging due to conversational dependencies in questions, leading to ambiguities and errors. To address this, we propose an NLU-NLG framework for ambiguity detection and resolution through reformulating query automatically and introduce a new task called "Ambiguity-guided Query Rewrite." To detect ambiguities, we develop a taxonomy based on real user conversational logs and draw insights from it to design rules and extract features for a classifier which yields superior performance in detecting ambiguous queries, outperforming LLM-based baselines. Furthermore, coupling the query rewrite module with our ambiguity detecting classifier shows that this end-to-end framework can effectively mitigate ambiguities without risking unnecessary insertions of unwanted phrases for clear queries, leading to an improvement in the overall performance of the AI Assistant. Due to its significance, this has been deployed in the real world application, namely Adobe Experience Platform AI Assistant.
Authors: Anna Seo Gyeong Choi, Jonghyeon Park, Myungwoo Oh
Abstract: Recent advancements in machine learning have significantly improved speech recognition, but recognizing speech from non-fluent or accented speakers remains a challenge. Previous efforts, relying on rule-based pronunciation patterns, have struggled to fully capture non-native errors. We propose two data-driven approaches using speech corpora to automatically detect mispronunciation patterns. By aligning non-native phones with their native counterparts using attention maps, we achieved a 5.7% improvement in speech recognition on native English datasets and a 12.8% improvement for non-native English speakers, particularly Korean speakers. Our method offers practical advancements for robust Automatic Speech Recognition (ASR) systems particularly for situations where prior linguistic knowledge is not applicable.
Authors: Yu Wang, Dmitry Krotov, Yuanzhe Hu, Yifan Gao, Wangchunshu Zhou, Julian McAuley, Dan Gutfreund, Rogerio Feris, Zexue He
Abstract: Equipping large language models (LLMs) with latent-space memory has attracted increasing attention as they can extend the context window of existing language models. However, retaining information from the distant past remains a challenge. For example, MemoryLLM (Wang et al., 2024a), as a representative work with latent-space memory, compresses past information into hidden states across all layers, forming a memory pool of 1B parameters. While effective for sequence lengths up to 16k tokens, it struggles to retain knowledge beyond 20k tokens. In this work, we address this limitation by introducing M+, a memory-augmented model based on MemoryLLM that significantly enhances long-term information retention. M+ integrates a long-term memory mechanism with a co-trained retriever, dynamically retrieving relevant information during text generation. We evaluate M+ on diverse benchmarks, including long-context understanding and knowledge retention tasks. Experimental results show that M+ significantly outperforms MemoryLLM and recent strong baselines, extending knowledge retention from under 20k to over 160k tokens with similar GPU memory overhead.
Authors: Pengfei Yu, Dongming Shen, Silin Meng, Jaewon Lee, Weisu Yin, Andrea Yaoyun Cui, Zhenlin Xu, Yi Zhu, Xingjian Shi, Mu Li, Alex Smola
Abstract: We present RPGBench, the first benchmark designed to evaluate large language models (LLMs) as text-based role-playing game (RPG) engines. RPGBench comprises two core tasks: Game Creation (GC) and Game Simulation (GS). In GC, an LLM must craft a valid and playable RPG world using a structured event-state representation, ensuring logical coherence and proper termination conditions. In GS, the LLM simulates interactive gameplay across multiple rounds while consistently updating states and enforcing game rules. To comprehensively assess performance, RPGBench integrates objective and subjective evaluation methodologies. Objective measures verify adherence to event mechanics and check variable updates without requiring human intervention. Subjective measures, such as content interestingness, action quality, and role-playing capability, are evaluated via an LLM-as-a-judge framework, where a strong LLM grades each candidate's outputs. Empirical results demonstrate that state-of-the-art LLMs can produce engaging stories but often struggle to implement consistent, verifiable game mechanics, particularly in long or complex scenarios. By combining structured, rule-based assessments with LLM-based judgments, RPGBench provides a new standard for evaluating how well LLMs can balance creativity, coherence, and complexity in text-based RPGs, opening avenues for more immersive and controllable interactive storytelling.
Authors: Tianci Liu, Zihan Dong, Linjun Zhang, Haoyu Wang, Jing Gao
Abstract: Large language models (LLMs) have achieved remarkable performance on various natural language tasks. However, they are trained on static corpora and their knowledge can become outdated quickly in the fast-changing world. This motivates the development of knowledge editing (KE) to update specific knowledge in LLMs without changing unrelated others or compromising their pre-trained capabilities. Previous efforts sought to update a small amount of parameters of a LLM and proved effective for making selective updates. Nonetheless, the edited LLM often exhibits degraded ability to reason about the new knowledge. In this work, we identify a key issue: heterogeneous token overfitting (HTO), where the LLM overfits different tokens in the provided knowledge at varying rates. To tackle this, we propose OVERTONE, a token-level smoothing method that mitigates HTO by adaptively refining the target distribution. Theoretically, OVERTONE offers better parameter updates with negligible computation overhead. It also induces an implicit DPO but does not require preference data pairs. Extensive experiments across four editing methods, two LLMs, and diverse scenarios demonstrate the effectiveness and versatility of our method.
Authors: Gabriel Lindenmaier, Sean Papay, Sebastian Pad\'o
Abstract: Transformer-based language models have recently been at the forefront of active research in text generation. However, these models' advances come at the price of prohibitive training costs, with parameter counts in the billions and compute requirements measured in petaflop/s-decades. In this paper, we investigate transformer-based architectures for improving model performance in a low-data regime by selectively replacing attention layers with feed-forward and quasi-recurrent neural network layers. We test these architectures on the standard Enwik8 and Wikitext-103 corpora. Our results show that our reduced architectures outperform existing models with a comparable number of parameters, and obtain comparable performance to larger models while significantly reducing the number of parameters.
Authors: Donglei Yu, Yang Zhao, Jie Zhu, Yangyifan Xu, Yu Zhou, Chengqing Zong
Abstract: Simultaneous Machine Translation (SiMT) generates translations while receiving streaming source inputs. This requires the SiMT model to learn a read/write policy, deciding when to translate and when to wait for more source input. Numerous linguistic studies indicate that audiences in SiMT scenarios have distinct preferences, such as accurate translations, simpler syntax, and no unnecessary latency. Aligning SiMT models with these human preferences is crucial to improve their performances. However, this issue still remains unexplored. Additionally, preference optimization for SiMT task is also challenging. Existing methods focus solely on optimizing the generated responses, ignoring human preferences related to latency and the optimization of read/write policy during the preference optimization phase. To address these challenges, we propose Simultaneous Preference Learning (SimulPL), a preference learning framework tailored for the SiMT task. In the SimulPL framework, we categorize SiMT human preferences into five aspects: \textbf{translation quality preference}, \textbf{monotonicity preference}, \textbf{key point preference}, \textbf{simplicity preference}, and \textbf{latency preference}. By leveraging the first four preferences, we construct human preference prompts to efficiently guide GPT-4/4o in generating preference data for the SiMT task. In the preference optimization phase, SimulPL integrates \textbf{latency preference} into the optimization objective and enables SiMT models to improve the read/write policy, thereby aligning with human preferences more effectively. Experimental results indicate that SimulPL exhibits better alignment with human preferences across all latency levels in Zh$\rightarrow$En, De$\rightarrow$En and En$\rightarrow$Zh SiMT tasks. Our data and code will be available at \url{https://github.com/EurekaForNLP/SimulPL}.
Authors: Borui Xu, Yao Chen, Zeyi Wen, Weiguo Liu, Bingsheng He
Abstract: The increasing demand for efficient summarization tools in resource-constrained environments highlights the need for effective solutions. While large language models (LLMs) deliver superior summarization quality, their high computational resource requirements limit practical use applications. In contrast, small language models (SLMs) present a more accessible alternative, capable of real-time summarization on edge devices. However, their summarization capabilities and comparative performance against LLMs remain underexplored. This paper addresses this gap by presenting a comprehensive evaluation of 19 SLMs for news summarization across 2,000 news samples, focusing on relevance, coherence, factual consistency, and summary length. Our findings reveal significant variations in SLM performance, with top-performing models such as Phi3-Mini and Llama3.2-3B-Ins achieving results comparable to those of 70B LLMs while generating more concise summaries. Notably, SLMs are better suited for simple prompts, as overly complex prompts may lead to a decline in summary quality. Additionally, our analysis indicates that instruction tuning does not consistently enhance the news summarization capabilities of SLMs. This research not only contributes to the understanding of SLMs but also provides practical insights for researchers seeking efficient summarization solutions that balance performance and resource use.
Authors: Wenzhe Li, Yong Lin, Mengzhou Xia, Chi Jin
Abstract: Ensembling outputs from diverse sources is a straightforward yet effective approach to boost performance. Mixture-of-Agents (MoA) is one such popular ensemble method that aggregates outputs from multiple different Large Language Models (LLMs). This paper raises the question in the context of language models: is mixing different LLMs truly beneficial? We propose Self-MoA -- an ensemble method that aggregates outputs from only the single top-performing LLM. Our extensive experiments reveal that, surprisingly, Self-MoA outperforms standard MoA that mixes different LLMs in a large number of scenarios: Self-MoA achieves $6.6\%$ improvement over MoA on the AlpacaEval 2.0 benchmark, and an average of $3.8\%$ improvement across various benchmarks, including MMLU, CRUX, and MATH. Applying Self-MoA to one of the top-ranking models in AlpacaEval 2.0 directly achieves the new state-of-the-art performance on the leaderboard. To understand the effectiveness of Self-MoA, we systematically investigate the trade-off between diversity and quality of outputs under various MoA settings. We confirm that the MoA performance is rather sensitive to the quality, and mixing different LLMs often lowers the average quality of the models. To complement the study, we identify the scenarios where mixing different LLMs could be helpful. This paper further introduces a sequential version of Self-MoA, that is capable of aggregating a large number of LLM outputs on-the-fly over multiple rounds, and is as effective as aggregating all outputs at once.
Authors: Minghang Deng, Ashwin Ramachandran, Canwen Xu, Lanxiang Hu, Zhewei Yao, Anupam Datta, Hao Zhang
Abstract: Text-to-SQL systems have unlocked easier access to critical data insights by enabling natural language queries over structured databases. However, deploying such systems in enterprise environments remains challenging due to factors such as large, complex schemas (> 3000 columns), diverse SQL dialects (e.g., BigQuery, Snowflake) and sophisticated query requirements (e.g., transformation, analytics). Current state-of-the-art performance on the Spider 2.0 dataset -- a benchmark built to mimic such complex environments -- remains limited at 20%. Key limitations include inadequate instruction-following, poor long-context comprehension, weak self-refinement, and insufficient dialect-specific knowledge. To address these gaps, we propose ReFoRCE (Self-Refinement Agent with Format Restriction and Column Exploration) which introduces (1) table compression to mitigate long-context limitations (2) format restriction to ensure accurate answer format, and (3) iterative column exploration for enhanced schema understanding. Additionally, it employs self-refinement pipeline consisting of (1) parallelized workflows with voting mechanisms and (2) a Common Table Expression (CTE) based refinement approach to handle unresolved cases. ReFoRCE achieves state-of-the-art results scoring 26.69 on the Spider 2.0-Snow and scoring 24.50 on the Spider 2.0-Lite tasks.
Authors: Atsumoto Ohashi, Ryuichiro Higashinaka
Abstract: Post-processing networks (PPNs) are components that modify the outputs of arbitrary modules in task-oriented dialogue systems and are optimized using reinforcement learning (RL) to improve the overall task completion capability of the system. However, previous PPN-based approaches have been limited to handling only a subset of modules within a system, which poses a significant limitation in improving the system performance. In this study, we propose a joint optimization method for post-processing the outputs of all modules using universal post-processing networks (UniPPNs), which are language-model-based networks that can modify the outputs of arbitrary modules in a system as a sequence-transformation task. Moreover, our RL algorithm, which employs a module-level Markov decision process, enables fine-grained value and advantage estimation for each module, thereby stabilizing joint learning for post-processing the outputs of all modules. Through both simulation-based and human evaluation experiments using the MultiWOZ dataset, we demonstrated that UniPPN outperforms conventional PPNs in the task completion capability of task-oriented dialogue systems.
Authors: Michael Mangrum, Jonathan Pemberton, Benedict Wetherby, Philip Montague
Abstract: Computational efficiency has remained a critical consideration in scaling high-capacity language models, with inference latency and resource consumption presenting significant constraints on real-time applications. The study has introduced a structured latency perturbation mechanism that modifies computational pathways through recursive state induction, enabling dynamic suppression of redundant activations while preserving generative fidelity. A formal mathematical framework has been established to describe recursive perturbations, ensuring that modifications remain adaptive rather than statically imposed. Experiments have demonstrated that applying recursive state adjustments reduces inference latency across varying sequence lengths, with longer text generations benefiting from cumulative efficiency improvements. Comparative evaluations against structured pruning and quantization have indicated that latency gains can be achieved without compromising token retention or memory utilization. The analysis of computational overhead has suggested that selectively suppressing redundant activations contributes to improved power efficiency, particularly in scenarios requiring extended text generation. An assessment of linguistic stability has shown that token-level consistency remains largely intact under controlled perturbation thresholds, reinforcing the viability of structural latency modifications as an alternative to weight-centric optimization techniques. The results have supported the hypothesis that recursive state induction offers an effective method for reducing computational complexity without requiring architectural modifications or external augmentation.
Authors: Liangyu Xu, Xuemiao Zhang, Feiyu Duan, Sirui Wang, Jingang Wang, Xunliang Cai
Abstract: Selecting high-quality data can significantly improve the pre-training efficiency of large language models (LLMs). Existing methods often rely on heuristic techniques and single quality signals, limiting their ability to comprehensively evaluate data quality. In this work, we propose FIRE, a flexible and scalable framework for integrating multiple data quality raters, which allows for a comprehensive assessment of data quality across various dimensions. FIRE aligns multiple quality signals into a unified space, and integrates diverse data quality raters to provide a comprehensive quality signal for each data point. Further, we introduce a progressive data selection scheme based on FIRE that iteratively refines the selection of high-quality data points, balancing computational complexity with the refinement of orthogonality. Experiments on the SlimPajama dataset reveal that FIRE consistently outperforms other selection methods and significantly enhances the pre-trained model across a wide range of downstream tasks, with a 2.9\% average performance boost and reducing the FLOPs necessary to achieve a certain performance level by more than half.
Authors: Ling Xing, Alex Jinpeng Wang, Rui Yan, Jinhui Tang
Abstract: Large Language Models (LLMs) have revolutionized natural language processing, excelling in handling longer sequences. However, the inefficiency and redundancy in processing extended in-context tokens remain a challenge. Many attempts to address this rely on compressing tokens with smaller text encoders, yet we question whether text encoders are truly indispensable. Our journey leads to an unexpected discovery-a much smaller vision encoder, applied directly to sequences of text tokens, can rival text encoders on text tasks. When pre-trained on large amounts of data and transferred to multiple mid-sized or small text understanding benchmarks, VIST leads to comparable results with 16% fewer FLOPs and 50% less memory usage. We further uncover significant token redundancy and devise a frequency-based masking strategy to guide the focus of the visual encoder toward the most critical tokens. Interestingly, we observe the trained visual encoder performs like a summarizer, selectively ignoring less important words such as prepositions and conjunctions. This approach delivers remarkable results, outperforming traditional text encoder-based methods by 5.7% on average over benchmarks like TriviaQA, NQ, PopQA, TREF, SST2, and SST5, setting a new standard for token efficiency in LLMs.
Authors: Zheng-Lin Lin, Yu-Fei Shih, Shu-Kai Hsieh
Abstract: This paper investigates the utilization of Large Language Models (LLMs) for solving complex linguistic puzzles, a domain requiring advanced reasoning and adept translation capabilities akin to human cognitive processes. We explore specific prompting techniques designed to enhance ability of LLMs to reason and elucidate their decision-making pathways, with a focus on Input-Output Prompting (IO), Chain-of-Thought Prompting (CoT), and Solo Performance Prompting (SPP). Utilizing datasets from the Puzzling Machine Competition and various Linguistics Olympiads, we employ a comprehensive set of metrics to assess the performance of GPT-4 0603, a prominent LLM, across these prompting methods. Our findings illuminate the potential of LLMs in linguistic reasoning and complex translation tasks, highlighting their capabilities and identifying limitations in the context of linguistic puzzles. This research contributes significantly to the broader field of Natural Language Processing (NLP) by providing insights into the optimization of LLM applications for improved reasoning and translation accuracy, thereby enriching the ongoing dialogue in NLP advancements.
Authors: Julian Perry, Frank Sanders, Carter Scott
Abstract: In this paper, we presents a novel method for improving text-to-image generation by combining Large Language Models (LLMs) with diffusion models, a hybrid approach aimed at achieving both higher quality and efficiency in image synthesis from text descriptions. Our approach introduces a new dynamic KL-weighting strategy to optimize the diffusion process, along with incorporating semantic understanding from pre-trained LLMs to guide the generation process. The proposed method significantly improves both the visual quality and alignment of generated images with text descriptions, addressing challenges such as computational inefficiency, instability in training, and robustness to textual variability. We evaluate our method on the COCO dataset and demonstrate its superior performance over traditional GAN-based models, both quantitatively and qualitatively. Extensive experiments, including ablation studies and human evaluations, confirm that our method outperforms existing approaches in terms of image realism, relevance to the input text, and overall aesthetic quality. Our approach also shows promise in scalability to other multimodal tasks, making it a versatile solution for a wide range of generative applications.
Authors: Robert Long, Eric Gonzalez, Harrison Fuller
Abstract: The advancement of large language models (LLMs) has opened new frontiers in natural language processing, particularly in specialized domains like healthcare. In this paper, we propose the Incremental Curriculum-Based Fine-Tuning (ICFT) framework to enhance the generative capabilities of medical large language models (MLLMs). ICFT combines curriculum-based learning, dual-stage memory coordination, and parameter-efficient fine-tuning to enable a progressive transition from general linguistic knowledge to strong domain-specific expertise. Experimental results across diverse medical NLP tasks, including question answering, preference classification, and response generation, demonstrate that ICFT consistently outperforms state-of-the-art baselines, achieving improvements in both accuracy and efficiency. Further analysis reveals the framework's ability to generalize to unseen data, reduce errors, and deliver diverse, contextually relevant medical responses. These findings establish ICFT as a robust and scalable solution for adapting LLMs to the medical domain, offering practical benefits for real-world healthcare applications.
Authors: Hadi Mohammadi, Ayoub Bagheri, Anastasia Giachanou, Daniel L. Oberski
Abstract: Natural Language Processing (NLP) has become a cornerstone in many critical sectors, including healthcare, finance, and customer relationship management. This is especially true with the development and use of advanced models such as GPT-based architectures and BERT, which are widely used in decision-making processes. However, the black-box nature of these advanced NLP models has created an urgent need for transparency and explainability. This review explores explainable NLP (XNLP) with a focus on its practical deployment and real-world applications, examining its implementation and the challenges faced in domain-specific contexts. The paper underscores the importance of explainability in NLP and provides a comprehensive perspective on how XNLP can be designed to meet the unique demands of various sectors, from healthcare's need for clear insights to finance's emphasis on fraud detection and risk assessment. Additionally, this review aims to bridge the knowledge gap in XNLP literature by offering a domain-specific exploration and discussing underrepresented areas such as real-world applicability, metric evaluation, and the role of human interaction in model assessment. The paper concludes by suggesting future research directions that could enhance the understanding and broader application of XNLP.
Authors: Jamshid Mozafari, Bhawna Piryani, Abdelrahman Abdallah, Adam Jatowt
Abstract: Large Language Models (LLMs) are transforming how people find information, and many users turn nowadays to chatbots to obtain answers to their questions. Despite the instant access to abundant information that LLMs offer, it is still important to promote critical thinking and problem-solving skills. Automatic hint generation is a new task that aims to support humans in answering questions by themselves by creating hints that guide users toward answers without directly revealing them. In this context, hint evaluation focuses on measuring the quality of hints, helping to improve the hint generation approaches. However, resources for hint research are currently spanning different formats and datasets, while the evaluation tools are missing or incompatible, making it hard for researchers to compare and test their models. To overcome these challenges, we introduce HintEval, a Python library that makes it easy to access diverse datasets and provides multiple approaches to generate and evaluate hints. HintEval aggregates the scattered resources into a single toolkit that supports a range of research goals and enables a clear, multi-faceted, and reliable evaluation. The proposed library also includes detailed online documentation, helping users quickly explore its features and get started. By reducing barriers to entry and encouraging consistent evaluation practices, HintEval offers a major step forward for facilitating hint generation and analysis research within the NLP/IR community.
Authors: Christoffer Loeffler, Andrea Mart\'inez Freile, Tom\'as Rey Pizarro
Abstract: This study addresses the growing concern of information asymmetry in consumer contracts, exacerbated by the proliferation of online services with complex Terms of Service that are rarely even read. Even though research on automatic analysis methods is conducted, the problem is aggravated by the general focus on English-language Machine Learning approaches and on major jurisdictions, such as the European Union. We introduce a new methodology and a substantial dataset addressing this gap. We propose a novel annotation scheme with four categories and a total of 20 classes, and apply it on 50 online Terms of Service used in Chile. Our evaluation of transformer-based models highlights how factors like language- and/or domain-specific pre-training, few-shot sample size, and model architecture affect the detection and classification of potentially abusive clauses. Results show a large variability in performance for the different tasks and models, with the highest macro-F1 scores for the detection task ranging from 79% to 89% and micro-F1 scores up to 96%, while macro-F1 scores for the classification task range from 60% to 70% and micro-F1 scores from 64% to 80%. Notably, this is the first Spanish-language multi-label classification dataset for legal clauses, applying Chilean law and offering a comprehensive evaluation of Spanish-language models in the legal domain. Our work lays the ground for future research in method development for rarely considered legal analysis and potentially leads to practical applications to support consumers in Chile and Latin America as a whole.
Authors: Ehsaneddin Asgari, Yassine El Kheir, Mohammad Ali Sadraei Javaheri
Abstract: Tokenization is fundamental to Natural Language Processing (NLP), directly impacting model efficiency and linguistic fidelity. While Byte Pair Encoding (BPE) is widely used in Large Language Models (LLMs), it often disregards morpheme boundaries, leading to suboptimal segmentation, particularly in morphologically rich languages. We introduce MorphBPE, a morphology-aware extension of BPE that integrates linguistic structure into subword tokenization while preserving statistical efficiency. Additionally, we propose two morphology-based evaluation metrics: (i) Morphological Consistency F1-Score, which quantifies the consistency between morpheme sharing and token sharing, contributing to LLM training convergence, and (ii) Morphological Edit Distance, which measures alignment between morphemes and tokens concerning interpretability. Experiments on English, Russian, Hungarian, and Arabic across 300M and 1B parameter LLMs demonstrate that MorphBPE consistently reduces cross-entropy loss, accelerates convergence, and improves morphological alignment scores. Fully compatible with existing LLM pipelines, MorphBPE requires minimal modifications for integration. The MorphBPE codebase and tokenizer playground will be available at: https://github.com/llm-lab-org/MorphBPE and https://tokenizer.llm-lab.org
URLs: https://github.com/llm-lab-org/MorphBPE, https://tokenizer.llm-lab.org
Authors: Taewoo Kang, Kjerstin Thorson, Tai-Quan Peng, Dan Hiaeshutter-Rice, Sanguk Lee, Stuart Soroka
Abstract: This study attempts to advancing content analysis methodology from consensus-oriented to coordination-oriented practices, thereby embracing diverse coding outputs and exploring the dynamics among differential perspectives. As an exploratory investigation of this approach, we evaluate six GPT-4o configurations to analyze sentiment in Fox News and MSNBC transcripts on Biden and Trump during the 2020 U.S. presidential campaign, examining patterns across these models. By assessing each model's alignment with ideological perspectives, we explore how partisan selective processing could be identified in LLM-Assisted Content Analysis (LACA). Findings reveal that partisan persona LLMs exhibit stronger ideological biases when processing politically congruent content. Additionally, intercoder reliability is higher among same-partisan personas compared to cross-partisan pairs. This approach enhances the nuanced understanding of LLM outputs and advances the integrity of AI-driven social science research, enabling simulations of real-world implications.
Authors: Alice Heiman
Abstract: A common language with standardized definitions is crucial for effective climate discussions. However, concerns exist about LLMs misrepresenting climate terms. We compared 300 official IPCC glossary definitions with those generated by GPT-4o-mini, Llama3.1 8B, and Mistral 7B, analyzing adherence, robustness, and readability using SBERT sentence embeddings. The LLMs scored an average adherence of $0.57-0.59 \pm 0.15$, and their definitions proved harder to read than the originals. Model-generated definitions vary mainly among words with multiple or ambiguous definitions, showing the potential to highlight terms that need standardization. The results show how LLMs could support environmental discourse while emphasizing the need to align model outputs with established terminology for clarity and consistency.
Authors: Stephen Zhang, Mustafa Khan, Vardan Papyan
Abstract: Two prominent features of large language models (LLMs) is the presence of large-norm (outlier) features and the tendency for tokens to attend very strongly to a select few tokens. Despite often having no semantic relevance, these select tokens, called attention sinks, along with the large outlier features, have proven important for model performance, compression, and streaming. Consequently, investigating the roles of these phenomena within models and exploring how they might manifest in the model parameters has become an area of active interest. Through an empirical investigation, we demonstrate that attention sinks utilize outlier features to: catch a sequence of tokens, tag the captured tokens by applying a common perturbation, and then release the tokens back into the residual stream, where the tagged tokens are eventually retrieved. We prove that simple tasks, like averaging, necessitate the 'catch, tag, release' mechanism hence explaining why it would arise organically in modern LLMs. Our experiments also show that the creation of attention sinks can be completely captured in the model parameters using low-rank matrices, which has important implications for model compression and substantiates the success of recent approaches that incorporate a low-rank term to offset performance degradation.
Authors: Cliff Wong, Sam Preston, Qianchu Liu, Zelalem Gero, Jass Bagga, Sheng Zhang, Shrey Jain, Theodore Zhao, Yu Gu, Yanbo Xu, Sid Kiblawi, Roshanthi Weerasinghe, Rom Leidner, Kristina Young, Brian Piening, Carlo Bifulco, Tristan Naumann, Mu Wei, Hoifung Poon
Abstract: The vast majority of real-world patient information resides in unstructured clinical text, and the process of medical abstraction seeks to extract and normalize structured information from this unstructured input. However, traditional medical abstraction methods can require significant manual efforts that can include crafting rules or annotating training labels, limiting scalability. In this paper, we propose UniMedAbstractor (UMA), a zero-shot medical abstraction framework leveraging Large Language Models (LLMs) through a modular and customizable prompt template. We refer to our approach as universal abstraction as it can quickly scale to new attributes through its universal prompt template without curating attribute-specific training labels or rules. We evaluate UMA for oncology applications, focusing on fifteen key attributes representing the cancer patient journey, from short-context attributes (e.g., performance status, treatment) to complex long-context attributes requiring longitudinal reasoning (e.g., tumor site, histology, TNM staging). Experiments on real-world data show UMA's strong performance and generalizability. Compared to supervised and heuristic baselines, UMA with GPT-4o achieves on average an absolute 2-point F1/accuracy improvement for both short-context and long-context attribute abstraction. For pathologic T staging, UMA even outperforms the supervised model by 20 points in accuracy.
Authors: Wentao Shi, Zichun Yu, Fuli Feng, Xiangnan He, Chenyan Xiong
Abstract: Monte Carlo Tree Search (MCTS) based methods provide promising approaches for generating synthetic data to enhance the self-training of Large Language Model (LLM) based multi-agent systems (MAS). These methods leverage Q-values to estimate individual agent contributions. However, relying solely on Q-values to identify informative data may misalign with the data synthesis objective, as the focus should be on selecting data that best enhances model training. To address this discrepancy, we propose Data Influence-oriented Tree Search (DITS), a novel framework that incorporates influence scores to guide both tree search and data selection. By leveraging influence scores, we effectively identify the most impactful data for system improvement, thereby enhancing model performance. Furthermore, we derive influence score estimation methods tailored for non-differentiable metrics, significantly reducing computational overhead by utilizing inference computations. Extensive experiments on eight multi-agent datasets demonstrate the robustness and effectiveness of the proposed methods. Notably, our findings reveal that allocating more inference resources to estimate influence scores, rather than Q-values, during data synthesis can more effectively and efficiently enhance model training.
Authors: Xiangci Li, Zhiyu Chen, Jason Ingyu Choi, Nikhita Vedula, Besnik Fetahu, Oleg Rokhlenko, Shervin Malmasi
Abstract: The goal of conversational product search (CPS) is to develop an intelligent, chat-based shopping assistant that can directly interact with customers to understand shopping intents, ask clarification questions, and find relevant products. However, training such assistants is hindered mainly due to the lack of reliable and large-scale datasets. Prior human-annotated CPS datasets are extremely small in size and lack integration with real-world product search systems. We propose a novel approach, TRACER, which leverages large language models (LLMs) to generate realistic and natural conversations for different shopping domains. TRACER's novelty lies in grounding the generation to dialogue plans, which are product search trajectories predicted from a decision tree model, that guarantees relevant product discovery in the shortest number of search conditions. We also release the first target-oriented CPS dataset Wizard of Shopping (WoS), containing highly natural and coherent conversations (3.6k) from three shopping domains. Finally, we demonstrate the quality and effectiveness of WoS via human evaluations and downstream tasks.
Authors: Litu Ou, Mirella Lapata
Abstract: Hierarchical Merging is a technique commonly used to summarize very long texts ($>$100K tokens) by breaking down the input into smaller sections, summarizing those sections individually, and then merging or combining those summaries into a final coherent summary. Although it helps address the limitations of large language models (LLMs) with fixed input length constraints, the recursive merging process can amplify LLM hallucinations, increasing the risk of factual inaccuracies. In this paper, we seek to mitigate hallucinations by enriching hierarchical merging with context from the source document. Specifically, we propose different approaches to contextual augmentation ranging from \emph{replacing} intermediate summaries with relevant input context, to \emph{refining} them while using the context as supporting evidence, and \emph{aligning} them implicitly (via citations) to the input. Experimental results on datasets representing legal and narrative domains show that contextual augmentation consistently outperforms zero-shot and hierarchical merging baselines for the Llama 3.1 model family. Our analysis further reveals that refinement methods tend to perform best when paired with extractive summarization for identifying relevant input.
Authors: Paul Albert, Frederic Z. Zhang, Hemanth Saratchandran, Cristian Rodriguez-Opazo, Anton van den Hengel, Ehsan Abbasnejad
Abstract: Low-Rank Adaptation (LoRA) and its variants have shown impressive results in reducing the number of trainable parameters and memory requirements of large transformer networks while maintaining fine-tuning performance. However, the low-rank nature of the weight update inherently limits the representation power of fine-tuned models, potentially compromising performance on complex tasks. This raises a critical question: when a performance gap between LoRA and standard fine-tuning is observed, is it due to the reduced number of trainable parameters or the rank deficiency? This paper aims to answer this question by introducing RandLoRA, a parameter-efficient method that performs full-rank updates using a learned linear combinations of low-rank, non-trainable random matrices. Our method limits the number of trainable parameters by restricting optimization to diagonal scaling matrices applied to the fixed random matrices. This allows us to effectively overcome the low-rank limitations while maintaining parameter and memory efficiency during training. Through extensive experimentation across vision, language, and vision-language benchmarks, we systematically evaluate the limitations of LoRA and existing random basis methods. Our findings reveal that full-rank updates are beneficial across vision and language tasks individually, and even more so for vision-language tasks, where RandLoRA significantly reduces -- and sometimes eliminates -- the performance gap between standard fine-tuning and LoRA, demonstrating its efficacy.
Authors: Kanika Goswami, Puneet Mathur, Ryan Rossi, Franck Dernoncourt
Abstract: Scientific data visualization is pivotal for transforming raw data into comprehensible visual representations, enabling pattern recognition, forecasting, and the presentation of data-driven insights. However, novice users often face difficulties due to the complexity of selecting appropriate tools and mastering visualization techniques. Large Language Models (LLMs) have recently demonstrated potential in assisting code generation, though they struggle with accuracy and require iterative debugging. In this paper, we propose PlotGen, a novel multi-agent framework aimed at automating the creation of precise scientific visualizations. PlotGen orchestrates multiple LLM-based agents, including a Query Planning Agent that breaks down complex user requests into executable steps, a Code Generation Agent that converts pseudocode into executable Python code, and three retrieval feedback agents - a Numeric Feedback Agent, a Lexical Feedback Agent, and a Visual Feedback Agent - that leverage multimodal LLMs to iteratively refine the data accuracy, textual labels, and visual correctness of generated plots via self-reflection. Extensive experiments show that PlotGen outperforms strong baselines, achieving a 4-6 percent improvement on the MatPlotBench dataset, leading to enhanced user trust in LLM-generated visualizations and improved novice productivity due to a reduction in debugging time needed for plot errors.
Authors: Kanika Goswami, Puneet Mathur, Ryan Rossi, Franck Dernoncourt
Abstract: Large Language Models (LLMs) can perform chart question-answering tasks but often generate unverified hallucinated responses. Existing answer attribution methods struggle to ground responses in source charts due to limited visual-semantic context, complex visual-text alignment requirements, and difficulties in bounding box prediction across complex layouts. We present ChartCitor, a multi-agent framework that provides fine-grained bounding box citations by identifying supporting evidence within chart images. The system orchestrates LLM agents to perform chart-to-table extraction, answer reformulation, table augmentation, evidence retrieval through pre-filtering and re-ranking, and table-to-chart mapping. ChartCitor outperforms existing baselines across different chart types. Qualitative user studies show that ChartCitor helps increase user trust in Generative AI by providing enhanced explainability for LLM-assisted chart QA and enables professionals to be more productive.
Authors: Ben Zhou, Sarthak Jain, Yi Zhang, Qiang Ning, Shuai Wang, Yassine Benajiba, Dan Roth
Abstract: Large language models have been shown to suffer from reasoning inconsistency issues. That is, they fail more in situations unfamiliar to the training data, even though exact or very similar reasoning paths exist in more common cases that they can successfully solve. Such observations motivate us to propose methods that encourage models to understand the high-level and abstract reasoning processes during training instead of only the final answer. This way, models can transfer the exact solution to similar cases, regardless of their relevance to the pre-training data distribution. In this work, we propose SAL, a self-supervised analogical learning framework. SAL mimics the human analogy process and trains models to explicitly transfer high-quality symbolic solutions from cases that they know how to solve to other rare cases in which they tend to fail more. We show that the resulting models after SAL learning outperform base language models on a wide range of reasoning benchmarks, such as StrategyQA, GSM8K, and HotpotQA, by 2% to 20%. At the same time, we show that our model is more generalizable and controllable through analytical studies.
Authors: Yuhang Zhou, Giannis Karamanolakis, Victor Soto, Anna Rumshisky, Mayank Kulkarni, Furong Huang, Wei Ai, Jianhua Lu
Abstract: The recent success of specialized Large Language Models (LLMs) in domains such as mathematical reasoning and coding has led to growing interest in methods for merging these expert LLMs into a unified Mixture-of-Experts (MoE) model, with the goal of enhancing performance in each domain while retaining effectiveness on general tasks. However, the effective merging of expert models remains an open challenge, especially for models with highly divergent weight parameters or different architectures. State-of-the-art MoE merging methods only work with homogeneous model architectures and rely on simple unweighted averaging to merge expert layers, which does not address parameter interference and requires extensive fine-tuning of the merged MoE to restore performance. To address these limitations, this paper introduces new MoE merging techniques, including strategies to mitigate parameter interference, routing heuristics to reduce the need for MoE fine-tuning, and a novel method for merging experts with different architectures. Extensive experiments across multiple domains demonstrate the effectiveness of our proposed methods, reducing fine-tuning costs, improving performance over state-of-the-art methods, and expanding the applicability of MoE merging.
Authors: Roy Xie, Junlin Wang, Paul Rosu, Chunyuan Deng, Bolun Sun, Zihao Lin, Bhuwan Dhingra
Abstract: Large language models (LLMs) process entire input contexts indiscriminately, which is inefficient in cases where the information required to answer a query is localized within the context. We present dynamic context cutoff, a human-inspired method enabling LLMs to self-terminate processing upon acquiring sufficient task-relevant information. Through analysis of model internals, we discover that specific attention heads inherently encode "sufficiency signals" - detectable through lightweight classifiers - that predict when critical information has been processed. This reveals a new efficiency paradigm: models' internal understanding naturally dictates processing needs rather than external compression heuristics. Comprehensive experiments across six QA datasets (up to 40K tokens) with three model families (LLaMA/Qwen/Mistral, 1B0-70B) demonstrate 1.33x average token reduction while improving accuracy by 1.3%. Furthermore, our method demonstrates better performance with the same rate of token reduction compared to other context efficiency methods. Additionally, we observe an emergent scaling phenomenon: while smaller models require require probing for sufficiency detection, larger models exhibit intrinsic self-assessment capabilities through prompting.
Authors: Zequan Liu, Yi Zhao, Ming Tan, Wei Zhu, Aaron Xuxiang Tian
Abstract: In the realm of parameter-efficient fine-tuning (PEFT) methods, while options like LoRA are available, there is a persistent demand in the industry for a PEFT approach that excels in both efficiency and performance within the context of single-backbone multi-tenant applications. This paper introduces a new and straightforward PEFT technique, termed \underline{P}rompt \underline{A}ware \underline{R}epresentation \underline{A}djustment (PARA). The core of our proposal is to integrate a lightweight vector generator within each Transformer layer. This generator produces vectors that are responsive to input prompts, thereby adjusting the hidden representations accordingly. Our extensive experimentation across diverse tasks has yielded promising results. Firstly, the PARA method has been shown to surpass current PEFT benchmarks in terms of performance, despite having a similar number of adjustable parameters. Secondly, it has proven to be more efficient than LoRA in the single-backbone multi-tenant scenario, highlighting its significant potential for industrial adoption.
Authors: Seungri Yoon, Woosang Jeon, Sanghyeok Choi, Taehyeong Kim, Tae In Ahn
Abstract: The development of biological data analysis tools and large language models (LLMs) has opened up new possibilities for utilizing AI in plant science research, with the potential to contribute significantly to knowledge integration and research gap identification. Nonetheless, current LLMs struggle to handle complex biological data and theoretical models in photosynthesis research and often fail to provide accurate scientific contexts. Therefore, this study proposed a photosynthesis research assistant (PRAG) based on OpenAI's GPT-4o with retrieval-augmented generation (RAG) techniques and prompt optimization. Vector databases and an automated feedback loop were used in the prompt optimization process to enhance the accuracy and relevance of the responses to photosynthesis-related queries. PRAG showed an average improvement of 8.7% across five metrics related to scientific writing, with a 25.4% increase in source transparency. Additionally, its scientific depth and domain coverage were comparable to those of photosynthesis research papers. A knowledge graph was used to structure PRAG's responses with papers within and outside the database, which allowed PRAG to match key entities with 63% and 39.5% of the database and test papers, respectively. PRAG can be applied for photosynthesis research and broader plant science domains, paving the way for more in-depth data analysis and predictive capabilities.
Authors: Jiali Chen, Xusen Hei, Yuqi Xue, Zihan Wu, Jiayuan Xie, Yi Cai
Abstract: Chinese literary classics hold significant cultural and educational value, offering deep insights into morality, history, and human nature. These works often include classical Chinese and complex narratives, making them difficult for children to read. To bridge this gap, we introduce a child-friendly literary adaptation (CLA) task to adapt the Chinese literary classic into engaging and accessible text for children. However, recent large language models (LLMs) overlook children's reading preferences (\ie, vivid character portrayals, concise narrative structures, and appropriate readability), which poses challenges in CLA. In this paper, we propose a method called InstructChild, which augments the LLM with these preferences for adaptation. Specifically, we first obtain the characters' personalities and narrative structure as additional information for fine-grained instruction tuning. Then, we devise a readability metric as the reward to align the LLM with the children's reading level. Finally, a lookahead decoding strategy is applied to improve the readability of the generated text during inference. To support the evaluation of CLA task, we construct the Classic4Children dataset, which comprises both the original and child-friendly versions of the Four Great Classical Novels of Chinese literature. Experimental results show that our InstructChild significantly improves automatic and human evaluation performance.
Authors: Farid Ariai, Maryam Tayefeh Mahmoudi, Ali Moeini
Abstract: In the era of pervasive internet use and the dominance of social networks, researchers face significant challenges in Persian text mining including the scarcity of adequate datasets in Persian and the inefficiency of existing language models. This paper specifically tackles these challenges, aiming to amplify the efficiency of language models tailored to the Persian language. Focusing on enhancing the effectiveness of sentiment analysis, our approach employs an aspect-based methodology utilizing the ParsBERT model, augmented with a relevant lexicon. The study centers on sentiment analysis of user opinions extracted from the Persian website 'Digikala.' The experimental results not only highlight the proposed method's superior semantic capabilities but also showcase its efficiency gains with an accuracy of 88.2% and an F1 score of 61.7. The importance of enhancing language models in this context lies in their pivotal role in extracting nuanced sentiments from user-generated content, ultimately advancing the field of sentiment analysis in Persian text mining by increasing efficiency and accuracy.
Authors: Vaishnavi Shrivastava, Ananya Kumar, Percy Liang
Abstract: Language models (LMs) should provide reliable confidence estimates to help users detect mistakes in their outputs and defer to human experts when necessary. Asking a language model to assess its confidence ("Score your confidence from 0-1.") is a natural way of evaluating its uncertainty. However, models struggle to provide absolute assessments of confidence (i.e. judging confidence in answering a question independent of other questions) and the coarse-grained scores they produce are not useful for evaluating the correctness of their answers. We propose relative confidence estimation, where we match up questions against each other and ask the model to make relative judgments of confidence ("Which question are you more confident in answering correctly?"). Treating each question as a "player" in a series of matchups against other questions and the model's preferences as match outcomes, we can use rank aggregation methods like Elo rating and Bradley-Terry to translate the model's confidence preferences into confidence scores. We evaluate relative confidence estimation against absolute confidence estimation and self-consistency confidence methods on five state-of-the-art LMs -- GPT-4, GPT-4o, Gemini 1.5 Pro, Claude 3.5 Sonnet, and Llama 3.1 405B -- across 14 challenging STEM, social science, and commonsense reasoning question answering tasks. Our results demonstrate that relative confidence estimation consistently provides more reliable confidence scores than absolute confidence estimation, with average gains of 3.5% in selective classification AUC over direct absolute confidence estimation methods and 1.7% over self-consistency approaches across all models and datasets.
Authors: Yu-Ling Hsu, Hsuan Su, Shang-Tse Chen
Abstract: Large language models (LLMs) have seen rapid development in recent years, revolutionizing various applications and significantly enhancing convenience and productivity. However, alongside their impressive capabilities, ethical concerns and new types of attacks, such as jailbreaking, have emerged. While most prompting techniques focus on optimizing adversarial inputs for individual cases, resulting in higher computational costs when dealing with large datasets. Less research has addressed the more general setting of training a universal attacker that can transfer to unseen tasks. In this paper, we introduce JUMP, a prompt-based method designed to jailbreak LLMs using universal multi-prompts. We also adapt our approach for defense, which we term DUMP. Experimental results demonstrate that our method for optimizing universal multi-prompts outperforms existing techniques.
Authors: Wen Lai, Alexander Fraser, Ivan Titov
Abstract: Parameter-efficient fine-tuning (PEFT) methods, such as LoRA, are commonly used to adapt LLMs. However, the effectiveness of standard PEFT methods is limited in low-resource scenarios with only a few hundred examples. Recent advances in interpretability research have inspired the emergence of activation editing techniques, which modify the activations of specific model components. These methods, due to their extremely small parameter counts, show promise for small datasets. However, their performance is highly dependent on identifying the correct modules to edit and often lacks stability across different datasets. In this paper, we propose Joint Localization and Activation Editing (JoLA), a method that jointly learns (1) which heads in the Transformer to edit (2) whether the intervention should be additive, multiplicative, or both and (3) the intervention parameters themselves - the vectors applied as additive offsets or multiplicative scalings to the head output. Through evaluations on three benchmarks spanning commonsense reasoning, natural language understanding, and natural language generation, we demonstrate that JoLA consistently outperforms existing methods.
Authors: Seokjin Oh, Keonwoong Noh, Woohwan Jung
Abstract: Despite the significant advances in neural machine translation, performance remains subpar for low-resource language pairs. Ensembling multiple systems is a widely adopted technique to enhance performance, often accomplished by combining probability distributions. However, the previous approaches face the challenge of high computational costs for training multiple models. Furthermore, for black-box models, averaging token-level probabilities at each decoding step is not feasible. To address the problems of multi-model ensemble methods, we present a pivot-based single model ensemble. The proposed strategy consists of two steps: pivot-based candidate generation and post-hoc aggregation. In the first step, we generate candidates through pivot translation. This can be achieved with only a single model and facilitates knowledge transfer from high-resource pivot languages, resulting in candidates that are not only diverse but also more accurate. Next, in the aggregation step, we select k high-quality candidates from the generated candidates and merge them to generate a final translation that outperforms the existing candidates. Our experimental results show that our method produces translations of superior quality by leveraging candidates from pivot translation to capture the subtle nuances of the source sentence.
Authors: Jonathan Tonglet, Gabriel Thiem, Iryna Gurevych
Abstract: Images taken out of their context are the most prevalent form of multimodal misinformation. Debunking them requires (1) providing the true context of the image and (2) checking the veracity of the image's caption. However, existing automated fact-checking methods fail to tackle both objectives explicitly. In this work, we introduce COVE, a new method that predicts first the true COntext of the image and then uses it to predict the VEracity of the caption. COVE beats the SOTA context prediction model on all context items, often by more than five percentage points. It is competitive with the best veracity prediction models on synthetic data and outperforms them on real-world data, showing that it is beneficial to combine the two tasks sequentially. Finally, we conduct a human study that reveals that the predicted context is a reusable and interpretable artifact to verify new out-of-context captions for the same image. Our code and data are made available.
Authors: Jenna Kanerva, Cassandra Ledins, Siiri K\"apyaho, Filip Ginter
Abstract: Optical Character Recognition (OCR) systems often introduce errors when transcribing historical documents, leaving room for post-correction to improve text quality. This study evaluates the use of open-weight LLMs for OCR error correction in historical English and Finnish datasets. We explore various strategies, including parameter optimization, quantization, segment length effects, and text continuation methods. Our results demonstrate that while modern LLMs show promise in reducing character error rates (CER) in English, a practically useful performance for Finnish was not reached. Our findings highlight the potential and limitations of LLMs in scaling OCR post-correction for large historical corpora.
Authors: Sam Kirkham, Patrycja Strycharczuk, Rob Davies, Danielle Welburn
Abstract: Short-term phonetic accommodation is a fundamental driver behind accent change, but how does real-time input from another speaker's voice shape the speech planning representations of an interlocutor? We advance a computational model of change in phonetic representations during phonetic accommodation, grounded in dynamic neural field equations for movement planning and memory dynamics. We test the model's ability to capture empirical patterns from an experimental study where speakers shadowed a model talker with a different accent from their own. The experimental data shows vowel-specific degrees of convergence during shadowing, followed by return to baseline (or minor divergence) post-shadowing. The model can reproduce these phenomena by modulating the magnitude of inhibitory memory dynamics, which may reflect resistance to accommodation due to phonological and/or sociolinguistic pressures. We discuss the implications of these results for the relation between short-term phonetic accommodation and longer-term patterns of sound change.
Authors: Hichem Ammar Khodja, Fr\'ed\'eric B\'echet, Quentin Brabant, Alexis Nasr, Gw\'enol\'e Lecorv\'e
Abstract: This paper explores the temporal robustness of language models (LMs) in handling factual knowledge. While LMs can often complete simple factual statements, their ability to manage temporal facts (those valid only within specific timeframes) remains uncertain. We design a controlled experiment to test the robustness of temporal factual knowledge inside LMs, which we use to evaluate several pretrained and instruction-tuned models using prompts on popular Wikidata facts, assessing their performance across different temporal granularities (Day, Month, and Year). Our findings indicate that even very large state-of-the-art models, such as Llama-3.1-70B, vastly lack robust knowledge of temporal facts. In addition, they are incapable of generalizing their knowledge from one granularity to another. These results highlight the inherent limitations of using LMs as temporal knowledge bases. The source code and data to reproduce our experiments will be released.
Authors: Chengfeng Zhou, Ji Wang, Juanjuan Qin, Yining Wang, Ling Sun, Weiwei Dai
Abstract: Large language models (LLMs) have shown significant promise across various medical applications, with ophthalmology being a notable area of focus. Many ophthalmic tasks have shown substantial improvement through the integration of LLMs. However, before these models can be widely adopted in clinical practice, evaluating their capabilities and identifying their limitations is crucial. To address this research gap and support the real-world application of LLMs, we introduce the OphthBench, a specialized benchmark designed to assess LLM performance within the context of Chinese ophthalmic practices. This benchmark systematically divides a typical ophthalmic clinical workflow into five key scenarios: Education, Triage, Diagnosis, Treatment, and Prognosis. For each scenario, we developed multiple tasks featuring diverse question types, resulting in a comprehensive benchmark comprising 9 tasks and 591 questions. This comprehensive framework allows for a thorough assessment of LLMs' capabilities and provides insights into their practical application in Chinese ophthalmology. Using this benchmark, we conducted extensive experiments and analyzed the results from 39 popular LLMs. Our evaluation highlights the current gap between LLM development and its practical utility in clinical settings, providing a clear direction for future advancements. By bridging this gap, we aim to unlock the potential of LLMs and advance their development in ophthalmology.
Authors: Sameer Pimparkhede, Pushpak Bhattacharyya
Abstract: Intent classification is crucial for conversational agents (chatbots), and deep learning models perform well in this area. However, little research has been done on the explainability of intent classification due to the absence of suitable benchmark data. Human annotation of explanation signals in text samples is time-consuming and costly. However, from inspection of data on intent classification, we see that, more often than not, the main verb denotes the action, and the direct object indicates the domain of conversation, serving as explanation signals for intent. This observation enables us to hypothesize that the main predicate in the text utterances, along with the arguments of the main predicate, can serve as explanation signals. Leveraging this, we introduce a new technique to automatically augment text samples from intent classification datasets with word-level explanations. We mark main predicates (primarily verbs) and their arguments (dependency relations) as explanation signals in benchmark intent classification datasets ATIS and SNIPS, creating a unique 21k-instance dataset for explainability. Further, we experiment with deep learning and language models. We observe that models that work well for classification do not perform well in explainability metrics like plausibility and faithfulness. We also observe that guiding models to focus on explanation signals from our dataset during training improves the plausibility Token F1 score by 3-4%, improving the model's reasoning.
Authors: Ahmed Masry, Juan A. Rodriguez, Tianyu Zhang, Suyuchen Wang, Chao Wang, Aarash Feizi, Akshay Kalkunte Suresh, Abhay Puri, Xiangru Jian, Pierre-Andr\'e No\"el, Sathwik Tejaswi Madhusudhan, Marco Pedersoli, Bang Liu, Nicolas Chapados, Yoshua Bengio, Enamul Hoque, Christopher Pal, Issam H. Laradji, David Vazquez, Perouz Taslakian, Spandana Gella, Sai Rajeswar
Abstract: Aligning visual features with language embeddings is a key challenge in vision-language models (VLMs). The performance of such models hinges on having a good connector that maps visual features generated by a vision encoder to a shared embedding space with the LLM while preserving semantic similarity. Existing connectors, such as multilayer perceptrons (MLPs), often produce out-of-distribution or noisy inputs, leading to misalignment between the modalities. In this work, we propose a novel vision-text alignment method, AlignVLM, that maps visual features to a weighted average of LLM text embeddings. Our approach leverages the linguistic priors encoded by the LLM to ensure that visual features are mapped to regions of the space that the LLM can effectively interpret. AlignVLM is particularly effective for document understanding tasks, where scanned document images must be accurately mapped to their textual content. Our extensive experiments show that AlignVLM achieves state-of-the-art performance compared to prior alignment methods. We provide further analysis demonstrating improved vision-text feature alignment and robustness to noise.
Authors: Giorgos Filandrianos, Angeliki Dimitriou, Maria Lymperaiou, Konstantinos Thomas, Giorgos Stamou
Abstract: The advent of Large Language Models (LLMs) has revolutionized product recommendation systems, yet their susceptibility to adversarial manipulation poses critical challenges, particularly in real-world commercial applications. Our approach is the first one to tap into human psychological principles, seamlessly modifying product descriptions, making these adversarial manipulations hard to detect. In this work, we investigate cognitive biases as black-box adversarial strategies, drawing parallels between their effects on LLMs and human purchasing behavior. Through extensive experiments on LLMs of varying scales, we reveal significant vulnerabilities in their use as recommenders, providing critical insights into safeguarding these systems.
Authors: Yuyang Gong, Zhuo Chen, Miaokun Chen, Fengchang Yu, Wei Lu, Xiaofeng Wang, Xiaozhong Liu, Jiawei Liu
Abstract: Retrieval-Augmented Generation (RAG) systems based on Large Language Models (LLMs) have become essential for tasks such as question answering and content generation. However, their increasing impact on public opinion and information dissemination has made them a critical focus for security research due to inherent vulnerabilities. Previous studies have predominantly addressed attacks targeting factual or single-query manipulations. In this paper, we address a more practical scenario: topic-oriented adversarial opinion manipulation attacks on RAG models, where LLMs are required to reason and synthesize multiple perspectives, rendering them particularly susceptible to systematic knowledge poisoning. Specifically, we propose Topic-FlipRAG, a two-stage manipulation attack pipeline that strategically crafts adversarial perturbations to influence opinions across related queries. This approach combines traditional adversarial ranking attack techniques and leverages the extensive internal relevant knowledge and reasoning capabilities of LLMs to execute semantic-level perturbations. Experiments show that the proposed attacks effectively shift the opinion of the model's outputs on specific topics, significantly impacting user information perception. Current mitigation methods cannot effectively defend against such attacks, highlighting the necessity for enhanced safeguards for RAG systems, and offering crucial insights for LLM security research.
Authors: Vinay Setty, Adam James Becker
Abstract: Podcasts are a popular medium on the web, featuring diverse and multilingual content that often includes unverified claims. Fact-checking podcasts is a challenging task, requiring transcription, annotation, and claim verification, all while preserving the contextual details of spoken content. Our tool offers a novel approach to tackle these challenges by enabling real-time annotation of podcasts during playback. This unique capability allows users to listen to the podcast and annotate key elements, such as check-worthy claims, claim spans, and contextual errors, simultaneously. By integrating advanced transcription models like OpenAI's Whisper and leveraging crowdsourced annotations, we create high-quality datasets to fine-tune multilingual transformer models such as XLM-RoBERTa for tasks like claim detection and stance classification. Furthermore, we release the annotated podcast transcripts and sample annotations with preliminary experiments.
Authors: Utkarsh Tiwari, Aviral Gupta, Michael Hahn
Abstract: Transformer architectures are the backbone of most modern language models, but understanding the inner workings of these models still largely remains an open problem. One way that research in the past has tackled this problem is by isolating the learning capabilities of these architectures by training them over well-understood classes of formal languages. We extend this literature by analyzing models trained over counter languages, which can be modeled using counter variables. We train transformer models on 4 counter languages, and equivalently formulate these languages using stacks, whose depths can be understood as the counter values. We then probe their internal representations for stack depths at each input token to show that these models when trained as next token predictors learn stack-like representations. This brings us closer to understanding the algorithmic details of how transformers learn languages and helps in circuit discovery.
Authors: David Rodriguez, William Seymour, Jose M. Del Alamo, Jose Such
Abstract: Large Language Models (LLMs) have gained unprecedented prominence, achieving widespread adoption across diverse domains and integrating deeply into society. The capability to fine-tune general-purpose LLMs, such as Generative Pre-trained Transformers (GPT), for specific tasks has facilitated the emergence of numerous Custom GPTs. These tailored models are increasingly made available through dedicated marketplaces, such as OpenAI's GPT Store. However, their black-box nature introduces significant safety and compliance risks. In this work, we present a scalable framework for the automated evaluation of Custom GPTs against OpenAI's usage policies, which define the permissible behaviors of these systems. Our framework integrates three core components: (1) automated discovery and data collection of models from the GPT store, (2) a red-teaming prompt generator tailored to specific policy categories and the characteristics of each target GPT, and (3) an LLM-as-a-judge technique to analyze each prompt-response pair for potential policy violations. We validate our framework with a manually annotated ground truth, and evaluate it through a large-scale study with 782 Custom GPTs across three categories: Romantic, Cybersecurity, and Academic GPTs. Our manual annotation process achieved an F1 score of 0.975 in identifying policy violations, confirming the reliability of the framework's assessments. The results reveal that 58.7% of the analyzed models exhibit indications of non-compliance, exposing weaknesses in the GPT store's review and approval processes. Furthermore, our findings indicate that a model's popularity does not correlate with compliance, and non-compliance issues largely stem from behaviors inherited from base models rather than user-driven customizations. We believe this approach is extendable to other chatbot platforms and policy domains, improving LLM-based systems safety.
Authors: Jinwei Hu, Zhenglin Huang, Xiangyu Yin, Wenjie Ruan, Guangliang Cheng, Yi Dong, Xiaowei Huang
Abstract: Large language models have been widely applied, but can inadvertently encode sensitive or harmful information, raising significant safety concerns. Machine unlearning has emerged to alleviate this concern; however, existing training-time unlearning approaches, relying on coarse-grained loss combinations, have limitations in precisely separating knowledge and balancing removal effectiveness with model utility. In contrast, we propose Fine-grained Activation manipuLation by Contrastive Orthogonal uNalignment (FALCON), a novel representation-guided unlearning approach that leverages information-theoretic guidance for efficient parameter selection, employs contrastive mechanisms to enhance representation separation, and projects conflict gradients onto orthogonal subspaces to resolve conflicts between forgetting and retention objectives. Extensive experiments demonstrate that FALCON achieves superior unlearning effectiveness while maintaining model utility, exhibiting robust resistance against knowledge recovery attempts.
Authors: Verna Dankers, Vikas Raunak
Abstract: In this work, we explore how instance-level memorization in the teacher Neural Machine Translation (NMT) model gets inherited by the student model in sequence-level knowledge distillation (SeqKD). We find that despite not directly seeing the original training data, students memorize more than baseline models (models of the same size, trained on the original data) -- 3.4% for exact matches and 57% for extractive memorization -- and show increased hallucination rates. Further, under this SeqKD setting, we also characterize how students behave on specific training data subgroups, such as subgroups with low quality and specific counterfactual memorization (CM) scores, and find that students exhibit amplified denoising on low-quality subgroups. Finally, we propose a modification to SeqKD named Adaptive-SeqKD, which intervenes in SeqKD to reduce memorization and hallucinations. Overall, we recommend caution when applying SeqKD: students inherit both their teachers' superior performance and their fault modes, thereby requiring active monitoring.
Authors: Gazi Tanbhir, Md. Farhan Shahriyar, Khandker Shahed, Abdullah Md Raihan Chy, Md Al Adnan
Abstract: Smishing is a social engineering attack using SMS containing malicious content to deceive individuals into disclosing sensitive information or transferring money to cybercriminals. Smishing attacks have surged by 328%, posing a major threat to mobile users, with losses exceeding \$54.2 million in 2019. Despite its growing prevalence, the issue remains significantly under-addressed. This paper presents a novel hybrid machine learning model for detecting Bangla smishing texts, combining Bidirectional Encoder Representations from Transformers (BERT) with Convolutional Neural Networks (CNNs) for enhanced character-level analysis. Our model addresses multi-class classification by distinguishing between Normal, Promotional, and Smishing SMS. Unlike traditional binary classification methods, our approach integrates BERT's contextual embeddings with CNN's character-level features, improving detection accuracy. Enhanced by an attention mechanism, the model effectively prioritizes crucial text segments. Our model achieves 98.47% accuracy, outperforming traditional classifiers, with high precision and recall in Smishing detection, and strong performance across all categories.
Authors: Zongxi Li, Yang Li, Haoran Xie, S. Joe Qin
Abstract: Large language models (LLMs) are prone to hallucinations in question-answering (QA) tasks when faced with ambiguous questions. Users often assume that LLMs share their cognitive alignment, a mutual understanding of context, intent, and implicit details, leading them to omit critical information in the queries. However, LLMs generate responses based on assumptions that can misalign with user intent, which may be perceived as hallucinations if they misalign with the user's intent. Therefore, identifying those implicit assumptions is crucial to resolve ambiguities in QA. Prior work, such as AmbigQA, reduces ambiguity in queries via human-annotated clarifications, which is not feasible in real application. Meanwhile, ASQA compiles AmbigQA's short answers into long-form responses but inherits human biases and fails capture explicit logical distinctions that differentiates the answers. We introduce Conditional Ambiguous Question-Answering (CondAmbigQA), a benchmark with 200 ambiguous queries and condition-aware evaluation metrics. Our study pioneers the concept of ``conditions'' in ambiguous QA tasks, where conditions stand for contextual constraints or assumptions that resolve ambiguities. The retrieval-based annotation strategy uses retrieved Wikipedia fragments to identify possible interpretations for a given query as its conditions and annotate the answers through those conditions. Such a strategy minimizes human bias introduced by different knowledge levels among annotators. By fixing retrieval results, CondAmbigQA evaluates how RAG systems leverage conditions to resolve ambiguities. Experiments show that models considering conditions before answering improve performance by $20\%$, with an additional $5\%$ gain when conditions are explicitly provided. These results underscore the value of conditional reasoning in QA, offering researchers tools to rigorously evaluate ambiguity resolution.
Authors: Raja Marjieh, Veniamin Veselovsky, Thomas L. Griffiths, Ilia Sucholutsky
Abstract: Numbers are a basic part of how humans represent and describe the world around them. As a consequence, learning effective representations of numbers is critical for the success of large language models as they become more integrated into everyday decisions. However, these models face a challenge: depending on context, the same sequence of digit tokens, e.g., 911, can be treated as a number or as a string. What kind of representations arise from this duality, and what are its downstream implications? Using a similarity-based prompting technique from cognitive science, we show that LLMs learn representational spaces that blend string-like and numerical representations. In particular, we show that elicited similarity judgments from these models over integer pairs can be captured by a combination of Levenshtein edit distance and numerical Log-Linear distance, suggesting an entangled representation. In a series of experiments we show how this entanglement is reflected in the latent embeddings, how it can be reduced but not entirely eliminated by context, and how it can propagate into a realistic decision scenario. These results shed light on a representational tension in transformer models that must learn what a number is from text input.
Authors: Mingyu Jin, Kai Mei, Wujiang Xu, Mingjie Sun, Ruixiang Tang, Mengnan Du, Zirui Liu, Yongfeng Zhang
Abstract: Large language models (LLMs) have achieved remarkable success in contextual knowledge understanding. In this paper, we show that these concentrated massive values consistently emerge in specific regions of attention queries (Q) and keys (K) while not having such patterns in values (V) in various modern transformer-based LLMs (Q, K, and V mean the representations output by the query, key, and value layers respectively). Through extensive experiments, we further demonstrate that these massive values play a critical role in interpreting contextual knowledge (knowledge obtained from the current context window) rather than in retrieving parametric knowledge stored within the model's parameters. Our further investigation of quantization strategies reveals that ignoring these massive values leads to a pronounced drop in performance on tasks requiring rich contextual understanding, aligning with our analysis. Finally, we trace the emergence of concentrated massive values and find that such concentration is caused by Rotary Positional Encoding (RoPE), which has appeared since the first layers. These findings shed new light on how Q and K operate in LLMs and offer practical insights for model design and optimization. The Code is Available at https://github.com/MingyuJ666/Rope_with_LLM.
Authors: Deqian Kong, Minglu Zhao, Dehong Xu, Bo Pang, Shu Wang, Edouardo Honig, Zhangzhang Si, Chuan Li, Jianwen Xie, Sirui Xie, Ying Nian Wu
Abstract: We propose a novel family of language models, Latent-Thought Language Models (LTMs), which incorporate explicit latent thought vectors that follow an explicit prior model in latent space. These latent thought vectors guide the autoregressive generation of ground tokens through a Transformer decoder. Training employs a dual-rate optimization process within the classical variational Bayes framework: fast learning of local variational parameters for the posterior distribution of latent vectors, and slow learning of global decoder parameters. Empirical studies reveal that LTMs possess additional scaling dimensions beyond traditional LLMs, yielding a structured design space. Higher sample efficiency can be achieved by increasing training compute per token, with further gains possible by trading model size for more inference steps. Designed based on these scaling properties, LTMs demonstrate superior sample and parameter efficiency compared to conventional autoregressive models and discrete diffusion models. They significantly outperform these counterparts in validation perplexity and zero-shot language modeling. Additionally, LTMs exhibit emergent few-shot in-context reasoning capabilities that scale with model and latent size, and achieve competitive performance in conditional and unconditional text generation.
Authors: Benjamin A. Spiegel, Lucas Gelfond, George Konidaris
Abstract: Abstract symbolic writing systems are semiotic codes that are ubiquitous in modern society but are otherwise absent in the animal kingdom. Anthropological evidence suggests that the earliest forms of some writing systems originally consisted of iconic pictographs, which signify their referent via visual resemblance. While previous studies have examined the emergence and, separately, the evolution of pictographic writing systems through a computational lens, most employ non-naturalistic methodologies that make it difficult to draw clear analogies to human and animal cognition. We develop a multi-agent reinforcement learning testbed for emergent communication called a Signification Game, and formulate a model of inferential communication that enables agents to leverage visual theory of mind to communicate actions using pictographs. Our model, which is situated within a broader formalism for animal communication, sheds light on the cognitive and cultural processes that led to the development of early writing systems.
Authors: Peng Lu, Ivan Kobyzev, Mehdi Rezagholizadeh, Boxing Chen, Philippe Langlais
Abstract: Recent advancements in Large Language Models (LLMs) have set themselves apart with their exceptional performance in complex language modelling tasks. However, these models are also known for their significant computational and storage requirements, primarily due to the quadratic computation complexity of softmax attention. To mitigate this issue, linear attention has been designed to reduce the quadratic space-time complexity that is inherent in standard transformers. In this work, we embarked on a comprehensive exploration of three key components that substantially impact the performance of the Gated Linear Attention module: feature maps, normalization, and the gating mechanism. We developed a feature mapping function to address some crucial issues that previous suggestions overlooked. Then we offered further rationale for the integration of normalization layers to stabilize the training process. Moreover, we explored the saturation phenomenon of the gating mechanism and augmented it with a refining module. We conducted extensive experiments and showed our architecture outperforms previous Gated Linear Attention mechanisms in extensive tasks including training from scratch and post-linearization with continual pre-training.
Authors: Tiago Timponi Torrent, Mark Turner, Nicol\'as Hinrichs, Frederico Belcavello, Igor Louren\c{c}o, Arthur Lorenzi Almeida, Marcelo Viridiano, Ely Edison Matos
Abstract: This paper presents a methodology combining multimodal semantic analysis with an eye-tracking experimental protocol to investigate the cognitive effort involved in understanding the communication of future scenarios. To demonstrate the methodology, we conduct a pilot study examining how visual fixation patterns vary during the evaluation of valence and counterfactuality in fictional ad pieces describing futuristic scenarios, using a portable eye tracker. Participants eye movements are recorded while evaluating the stimuli and describing them to a conversation partner. Gaze patterns are analyzed alongside semantic representations of the stimuli and participants descriptions, constructed from a frame semantic annotation of both linguistic and visual modalities. Preliminary results show that far-future and pessimistic scenarios are associated with longer fixations and more erratic saccades, supporting the hypothesis that fractures in the base spaces underlying the interpretation of future scenarios increase cognitive load for comprehenders.
Authors: Yue Huang, Yanbo Wang, Zixiang Xu, Chujie Gao, Siyuan Wu, Jiayi Ye, Xiuying Chen, Pin-Yu Chen, Xiangliang Zhang
Abstract: Recent advances in Large Language Models (LLMs) have revolutionized generative systems, achieving excellent performance across diverse domains. Although these models perform well in controlled environments, their real-world applications frequently encounter inputs containing both essential and irrelevant details. Our investigation has revealed a critical vulnerability in LLMs, which we term Contextual Distraction Vulnerability (CDV). This phenomenon arises when models fail to maintain consistent performance on questions modified with semantically coherent but irrelevant context. To systematically investigate this vulnerability, we propose an efficient tree-based search methodology to automatically generate CDV examples. Our approach successfully generates CDV examples across four datasets, causing an average performance degradation of approximately 45% in state-of-the-art LLMs. To address this critical issue, we explore various mitigation strategies and find that post-targeted training approaches can effectively enhance model robustness against contextual distractions. Our findings highlight the fundamental nature of CDV as an ability-level challenge rather than a knowledge-level issue since models demonstrate the necessary knowledge by answering correctly in the absence of distractions. This calls the community's attention to address CDV during model development to ensure reliability. The code is available at https://github.com/wyf23187/LLM_CDV.
Authors: Tatsuki Kuribayashi, Yohei Oseki, Souhaib Ben Taieb, Kentaro Inui, Timothy Baldwin
Abstract: Recent cognitive modeling studies have reported that larger language models (LMs) exhibit a poorer fit to human reading behavior, leading to claims of their cognitive implausibility. In this paper, we revisit this argument through the lens of mechanistic interpretability and argue that prior conclusions were skewed by an exclusive focus on the final layers of LMs. Our analysis reveals that next-word probabilities derived from internal layers of larger LMs align with human sentence processing data as well as, or better than, those from smaller LMs. This alignment holds consistently across behavioral (self-paced reading times, gaze durations, MAZE task processing times) and neurophysiological (N400 brain potentials) measures, challenging earlier mixed results and suggesting that the cognitive plausibility of larger LMs has been underestimated. Furthermore, we first identify an intriguing relationship between LM layers and human measures: earlier layers correspond more closely with fast gaze durations, while later layers better align with relatively slower signals such as N400 potentials and MAZE processing times. Our work opens new avenues for interdisciplinary research at the intersection of mechanistic interpretability and cognitive modeling.
Authors: Muhammad Zain Raza, Jiawei Xu, Terence Lim, Lily Boddy, Carlos M. Mery, Andrew Well, Ying Ding
Abstract: Thematic Analysis (TA) is a fundamental method in healthcare research for analyzing transcript data, but it is resource-intensive and difficult to scale for large, complex datasets. This study investigates the potential of large language models (LLMs) to augment the inductive TA process in high-stakes healthcare settings. Focusing on interview transcripts from parents of children with Anomalous Aortic Origin of a Coronary Artery (AAOCA), a rare congenital heart disease, we propose an LLM-Enhanced Thematic Analysis (LLM-TA) pipeline. Our pipeline integrates an affordable state-of-the-art LLM (GPT-4o mini), LangChain, and prompt engineering with chunking techniques to analyze nine detailed transcripts following the inductive TA framework. We evaluate the LLM-generated themes against human-generated results using thematic similarity metrics, LLM-assisted assessments, and expert reviews. Results demonstrate that our pipeline outperforms existing LLM-assisted TA methods significantly. While the pipeline alone has not yet reached human-level quality in inductive TA, it shows great potential to improve scalability, efficiency, and accuracy while reducing analyst workload when working collaboratively with domain experts. We provide practical recommendations for incorporating LLMs into high-stakes TA workflows and emphasize the importance of close collaboration with domain experts to address challenges related to real-world applicability and dataset complexity. https://github.com/jiaweixu98/LLM-TA
Authors: Akshat Gupta, Phudish Prateepamornkul, Maochuan Lu, Ahmed Alaa, Thomas Hartvigsen, Gopala Anumanchipalli
Abstract: Prior work in parameter-modifying knowledge editing has shown that large-scale sequential editing leads to significant model degradation. In this paper, we study the reasons behind this and scale sequential knowledge editing to 10,000 sequential edits, while maintaining the downstream performance of the original model. We first show that locate-then-edit knowledge editing methods lead to overfitting on the edited facts. We also show that continuous knowledge editing using these methods leads to disproportionate growth in the norm of the edited matrix. We then provide a crucial insight into the inner workings of locate-then-edit methods. We show that norm-growth is a hidden trick employed by these methods that gives larger importance to the output activations produced from the edited layers. With this "importance hacking", the edited layers provide a much larger contributions to the model's output. To mitigate these issues, we present ENCORE - Early stopping and Norm-Constrained Robust knowledge Editing. ENCORE controls for overfitting and the disproportionate norm-growth to enable long-term sequential editing, where we are able to perform up to 10,000 sequential edits without loss of downstream performance. ENCORE is also 61% faster than MEMIT and 64% faster than AlphaEdit on Llama3-8B.
Authors: Da Yu, Edith Cohen, Badih Ghazi, Yangsibo Huang, Pritish Kamath, Ravi Kumar, Daogao Liu, Chiyuan Zhang
Abstract: We propose SCONE ($\textbf{S}$calable, $\textbf{C}$ontextualized, $\textbf{O}$ffloaded, $\textbf{N}$-gram $\textbf{E}$mbedding), a method for extending input embedding layers to enhance language model performance as layer size scales. To avoid increased decoding costs, SCONE retains the original vocabulary while introducing embeddings for a set of frequent $n$-grams. These embeddings provide contextualized representation for each input token and are learned with a separate model during training. During inference, they are precomputed and stored in off-accelerator memory with minimal impact on inference speed. SCONE enables two new scaling strategies: increasing the number of cached $n$-gram embeddings and scaling the model used to learn them, all while maintaining fixed inference-time FLOPS. We show that scaling both aspects allows SCONE to outperform a 1.9B parameter baseline across diverse corpora, while using only half the inference-time FLOPS.
Authors: Arianna Salazar-Miranda, Emily Talen
Abstract: Cities are at the forefront of addressing global sustainability challenges, particularly those exacerbated by climate change. Traditional zoning codes, which often segregate land uses, have been linked to increased vehicular dependence, urban sprawl, and social disconnection, undermining broader social and environmental sustainability objectives. This study investigates the adoption and impact of form-based codes (FBCs), which aim to promote sustainable, compact, and mixed-use urban forms as a solution to these issues. Using Natural Language Processing (NLP) techniques, we analyzed zoning documents from over 2000 U.S. census-designated places to identify linguistic patterns indicative of FBC principles. Our findings reveal widespread adoption of FBCs across the country, with notable variations within regions. FBCs are associated with higher floor-to-area ratios, narrower and more consistent street setbacks, and smaller plots. We also find that places with FBCs have improved walkability, shorter commutes, and a higher share of multi-family housing. Our findings highlight the utility of NLP for evaluating zoning codes and underscore the potential benefits of form-based zoning reforms for enhancing urban sustainability.
Authors: Kamer Ali Yuksel, Hassan Sawaf
Abstract: Financial metrics like the Sharpe ratio are pivotal in evaluating investment performance by balancing risk and return. However, traditional metrics often struggle with robustness and generalization, particularly in dynamic and volatile market conditions. This paper introduces AlphaSharpe, a novel framework leveraging large language models (LLMs) to iteratively evolve and optimize financial metrics to discover enhanced risk-return metrics that outperform traditional approaches in robustness and correlation with future performance metrics by employing iterative crossover, mutation, and evaluation. Key contributions of this work include: (1) a novel use of LLMs to generate and refine financial metrics with implicit domain-specific knowledge, (2) a scoring mechanism to ensure that evolved metrics generalize effectively to unseen data, and (3) an empirical demonstration of 3x predictive power for future risk-returns, and 2x portfolio performance. Experimental results in a real-world dataset highlight the superiority of discovered metrics, making them highly relevant to portfolio managers and financial decision-makers. This framework not only addresses the limitations of existing metrics but also showcases the potential of LLMs in advancing financial analytics, paving the way for informed and robust investment strategies.
Authors: Tom Wallace, Naser Ezzati-Jivan, Beatrice Ombuki-Berman
Abstract: Advancements in Natural Language Processing are heavily reliant on the Transformer architecture, whose improvements come at substantial resource costs due to ever-growing model sizes. This study explores optimization techniques, including Quantization, Knowledge Distillation, and Pruning, focusing on energy and computational efficiency while retaining performance. Among standalone methods, 4-bit Quantization significantly reduces energy use with minimal accuracy loss. Hybrid approaches, like NVIDIA's Minitron approach combining KD and Structured Pruning, further demonstrate promising trade-offs between size reduction and accuracy retention. A novel optimization equation is introduced, offering a flexible framework for comparing various methods. Through the investigation of these compression methods, we provide valuable insights for developing more sustainable and efficient LLMs, shining a light on the often-ignored concern of energy efficiency.
Authors: Colin Sisate, Alistair Goldfinch, Vincent Waterstone, Sebastian Kingsley, Mariana Blackthorn
Abstract: Contextually Entangled Gradient Mapping (CEGM) introduces a new approach to gradient optimization, redefining the relationship between contextual embeddings and gradient updates to enhance semantic coherence and reasoning capabilities in neural architectures. By treating gradients as dynamic carriers of contextual dependencies rather than isolated numerical entities, the proposed methodology bridges critical gaps in existing optimization strategies. The integration of entangled gradient dynamics into a loss regularization framework demonstrated significant improvements in tasks involving long-form reasoning, contextual retention, and adaptability to unseen domains. Experimental evaluations showed that the CEGM-enhanced model consistently outperformed baseline approaches, achieving higher accuracy in token-level predictions and greater resilience to noisy inputs. Practical implementations involved modifications to training pipelines, introducing entanglement layers and dynamic coefficient adjustments that seamlessly align with existing architectures. Results further highlighted reductions in semantic drift during sequential transformations and improvements in embedding coherence across paraphrased sentences, showing the robustness and versatility of the proposed methodology. The findings demonstrate the broader implications of gradient entanglement for both theoretical advancements and practical applications in optimization strategies.
Authors: Kamil\.e Luko\v{s}i\=ut\.e, Adam Swanda
Abstract: Large language models (LLMs) are demonstrating increasing prowess in cybersecurity applications, creating creating inherent risks alongside their potential for strengthening defenses. In this position paper, we argue that current efforts to evaluate risks posed by these capabilities are misaligned with the goal of understanding real-world impact. Evaluating LLM cybersecurity risk requires more than just measuring model capabilities -- it demands a comprehensive risk assessment that incorporates analysis of threat actor adoption behavior and potential for impact. We propose a risk assessment framework for LLM cyber capabilities and apply it to a case study of language models used as cybersecurity assistants. Our evaluation of frontier models reveals high compliance rates but moderate accuracy on realistic cyber assistance tasks. However, our framework suggests that this particular use case presents only moderate risk due to limited operational advantages and impact potential. Based on these findings, we recommend several improvements to align research priorities with real-world impact assessment, including closer academia-industry collaboration, more realistic modeling of attacker behavior, and inclusion of economic metrics in evaluations. This work represents an important step toward more effective assessment and mitigation of LLM-enabled cybersecurity risks.
Authors: Ahmed Heakl, Sara Ghaboura, Omkar Thawkar, Fahad Shahbaz Khan, Hisham Cholakkal, Rao Muhammad Anwer, Salman Khan
Abstract: Amid the swift progress of large language models (LLMs) and their evolution into large multimodal models (LMMs), significant strides have been made in high-resource languages such as English and Chinese. While Arabic LLMs have seen notable progress, Arabic LMMs remain largely unexplored, often narrowly focusing on a few specific aspects of the language and visual understanding. To bridge this gap, we introduce AIN-the Arabic Inclusive Multimodal Model-designed to excel across diverse domains. AIN is an English-Arabic bilingual LMM designed to excel in English and Arabic, leveraging carefully constructed 3.6 million high-quality Arabic-English multimodal data samples. AIN demonstrates state-of-the-art Arabic performance, while also possessing strong English-language visual capabilities. On the recent CAMEL-Bench benchmark comprising 38 sub-domains including, multi-image understanding, complex visual perception, handwritten document understanding, video understanding, medical imaging, plant diseases, and remote sensing-based land use understanding, our AIN demonstrates strong performance with the 7B model outperforming GPT-4o by an absolute gain of 3.4% averaged over eight domains and 38 sub-domains. AIN's superior capabilities position it as a significant step toward empowering Arabic speakers with advanced multimodal generative AI tools across diverse applications.
Authors: Abdurrahim Yilmaz, Furkan Yuceyalcin, Ece Gokyayla, Donghee Choi, Ozan Erdem Ali Anil Demircali, Rahmetullah Varol, Ufuk Gorkem Kirabali, Gulsum Gencoglan, Joram M. Posma, Burak Temelkuran
Abstract: A major barrier to developing vision large language models (LLMs) in dermatology is the lack of large image--text pairs dataset. We introduce DermaSynth, a dataset comprising of 92,020 synthetic image--text pairs curated from 45,205 images (13,568 clinical and 35,561 dermatoscopic) for dermatology-related clinical tasks. Leveraging state-of-the-art LLMs, using Gemini 2.0, we used clinically related prompts and self-instruct method to generate diverse and rich synthetic texts. Metadata of the datasets were incorporated into the input prompts by targeting to reduce potential hallucinations. The resulting dataset builds upon open access dermatological image repositories (DERM12345, BCN20000, PAD-UFES-20, SCIN, and HIBA) that have permissive CC-BY-4.0 licenses. We also fine-tuned a preliminary Llama-3.2-11B-Vision-Instruct model, DermatoLlama 1.0, on 5,000 samples. We anticipate this dataset to support and accelerate AI research in dermatology. Data and code underlying this work are accessible at https://github.com/abdurrahimyilmaz/DermaSynth.
Authors: Luyang Zhang, Cathy Jiao, Beibei Li, Chenyan Xiong
Abstract: Training data is a pivotal resource for building large language models (LLMs), but unfair pricing in data markets poses a serious challenge for both data buyers (e.g., LLM builders) and sellers (e.g., human annotators), which discourages market participation, reducing data quantity and quality. In this paper, we propose a fairshare pricing framework that sets training data prices using data valuation methods to quantify their contribution to LLMs. In our framework, buyers make purchasing decisions using data valuation and sellers set prices to maximize their profits based on the anticipated buyer purchases. We theoretically show that pricing derived from our framework is tightly linked to data valuation and buyers' budget, optimal for both buyers and sellers. Through market simulations using current LLMs and datasets (math problems, medical diagnosis, and physical reasoning), we show that our framework is fairshare for buyers by ensuring their purchased data is reflective of model training value, leading to higher LLM task performances per-dollar spent on data, and fairshare for sellers by ensuring they sell their data at optimal prices. Our framework lays the foundation for future research on equitable and sustainable data markets for large-scale AI.
Authors: Shengyang Sun, Yian Zhang, Alexander Bukharin, David Mosallanezhad, Jiaqi Zeng, Soumye Singhal, Gerald Shen, Adi Renduchintala, Tugrul Konuk, Yi Dong, Zhilin Wang, Dmitry Chichkov, Olivier Delalleau, Oleksii Kuchaiev
Abstract: The rapid development of large language model (LLM) alignment algorithms has resulted in a complex and fragmented landscape, with limited clarity on the effectiveness of different methods and their inter-connections. This paper introduces Reward-Aware Preference Optimization (RPO), a mathematical framework that unifies popular preference optimization techniques in LLM alignment, including DPO, IPO, SimPO, and REINFORCE (LOO), among others. RPO provides a structured approach to disentangle and systematically study the impact of various design choices, such as the optimization objective, the number of responses per prompt, and the use of implicit versus explicit reward models, on LLM preference optimization. We additionally propose a new experimental setup that enables the clean and direct ablation of such design choices. Through an extensive series of ablation studies within the RPO framework, we gain insights into the critical factors shaping model alignment, offering practical guidance on the most effective strategies for improving LLM alignment.
Authors: Keegan Harris, Aleksandrs Slivkins
Abstract: We evaluate the ability of the current generation of large language models (LLMs) to help a decision-making agent facing an exploration-exploitation tradeoff. We use LLMs to explore and exploit in silos in various (contextual) bandit tasks. We find that while the current LLMs often struggle to exploit, in-context mitigations may be used to substantially improve performance for small-scale tasks. However even then, LLMs perform worse than a simple linear regression. On the other hand, we find that LLMs do help at exploring large action spaces with inherent semantics, by suggesting suitable candidates to explore.
Authors: Shiqi He, Insu Jang, Mosharaf Chowdhury
Abstract: Incorporating multiple modalities into large language models (LLMs) is a powerful way to enhance their understanding of non-textual data, enabling them to perform multimodal tasks. Vision language models (VLMs) form the fastest growing category of multimodal models because of their many practical use cases, including in healthcare, robotics, and accessibility. Unfortunately, even though different VLMs in the literature demonstrate impressive visual capabilities in different benchmarks, they are handcrafted by human experts; there is no automated framework to create task-specific multimodal models. We introduce Mordal, an automated multimodal model search framework that efficiently finds the best VLM for a user-defined task without manual intervention. Mordal achieves this both by reducing the number of candidates to consider during the search process and by minimizing the time required to evaluate each remaining candidate. Our evaluation shows that Mordal can find the best VLM for a given problem using up to $8.9\times$--$11.6\times$ lower GPU hours than grid search. In the process of our evaluation, we have also discovered new VLMs that outperform their state-of-the-art counterparts.
Authors: Hongyi Liu, Rajarshi Saha, Zhen Jia, Youngsuk Park, Jiaji Huang, Shoham Sabach, Yu-Xiang Wang, George Karypis
Abstract: Large Language Models (LLMs) have demonstrated exceptional performance in natural language processing tasks, yet their massive size makes serving them inefficient and costly. Semi-structured pruning has emerged as an effective method for model acceleration, but existing approaches are suboptimal because they focus on local, layer-wise optimizations using heuristic rules, failing to leverage global feedback. We present ProxSparse, a learning-based framework for mask selection enabled by regularized optimization. ProxSparse transforms the rigid, non-differentiable mask selection process into a smoother optimization procedure, allowing gradual mask exploration with flexibility. ProxSparse does not involve additional weight updates once the mask is determined. Our extensive evaluations on 7 widely used models show that ProxSparse consistently outperforms previously proposed semi-structured mask selection methods with significant improvement, demonstrating the effectiveness of our learned approach towards semi-structured pruning.
Authors: Ali Naseh, Yuefeng Peng, Anshuman Suri, Harsh Chaudhari, Alina Oprea, Amir Houmansadr
Abstract: Retrieval-Augmented Generation (RAG) enables Large Language Models (LLMs) to generate grounded responses by leveraging external knowledge databases without altering model parameters. Although the absence of weight tuning prevents leakage via model parameters, it introduces the risk of inference adversaries exploiting retrieved documents in the model's context. Existing methods for membership inference and data extraction often rely on jailbreaking or carefully crafted unnatural queries, which can be easily detected or thwarted with query rewriting techniques common in RAG systems. In this work, we present Interrogation Attack (IA), a membership inference technique targeting documents in the RAG datastore. By crafting natural-text queries that are answerable only with the target document's presence, our approach demonstrates successful inference with just 30 queries while remaining stealthy; straightforward detectors identify adversarial prompts from existing methods up to ~76x more frequently than those generated by our attack. We observe a 2x improvement in TPR@1%FPR over prior inference attacks across diverse RAG configurations, all while costing less than $0.02 per document inference.
Authors: Alaa Nfissi, Wassim Bouachir, Nizar Bouguila, Brian Mishara
Abstract: In the field of human-computer interaction and psychological assessment, speech emotion recognition (SER) plays an important role in deciphering emotional states from speech signals. Despite advancements, challenges persist due to system complexity, feature distinctiveness issues, and noise interference. This paper introduces a new end-to-end (E2E) deep learning multi-resolution framework for SER, addressing these limitations by extracting meaningful representations directly from raw waveform speech signals. By leveraging the properties of the fast discrete wavelet transform (FDWT), including the cascade algorithm, conjugate quadrature filter, and coefficient denoising, our approach introduces a learnable model for both wavelet bases and denoising through deep learning techniques. The framework incorporates an activation function for learnable asymmetric hard thresholding of wavelet coefficients. Our approach exploits the capabilities of wavelets for effective localization in both time and frequency domains. We then combine one-dimensional dilated convolutional neural networks (1D dilated CNN) with a spatial attention layer and bidirectional gated recurrent units (Bi-GRU) with a temporal attention layer to efficiently capture the nuanced spatial and temporal characteristics of emotional features. By handling variable-length speech without segmentation and eliminating the need for pre or post-processing, the proposed model outperformed state-of-the-art methods on IEMOCAP and EMO-DB datasets. The source code of this paper is shared on the Github repository: https://github.com/alaaNfissi/SigWavNet-Learning-Multiresolution-Signal-Wavelet-Network-for-Speech-Emotion-Recognition.
Authors: Hadi Hosseini, Samarth Khanna
Abstract: The growing interest in employing large language models (LLMs) for decision-making in social and economic contexts has raised questions about their potential to function as agents in these domains. A significant number of societal problems involve the distribution of resources, where fairness, along with economic efficiency, play a critical role in the desirability of outcomes. In this paper, we examine whether LLM responses adhere to fundamental fairness concepts such as equitability, envy-freeness, and Rawlsian maximin, and investigate their alignment with human preferences. We evaluate the performance of several LLMs, providing a comparative benchmark of their ability to reflect these measures. Our results demonstrate a lack of alignment between current LLM responses and human distributional preferences. Moreover, LLMs are unable to utilize money as a transferable resource to mitigate inequality. Nonetheless, we demonstrate a stark contrast when (some) LLMs are tasked with selecting from a predefined menu of options rather than generating one. In addition, we analyze the robustness of LLM responses to variations in semantic factors (e.g. intentions or personas) or non-semantic prompting changes (e.g. templates or orderings). Finally, we highlight potential strategies aimed at enhancing the alignment of LLM behavior with well-established fairness concepts.
Authors: Di Chai, Pengbo Li, Feiyuan Zhang, Yilun Jin, Han Tian, Junxue Zhang, Kai Chen
Abstract: Token filtering has been proposed to enhance utility of large language models (LLMs) by eliminating inconsequential tokens during training. While using fewer tokens should reduce computational workloads, existing studies have not succeeded in achieving higher efficiency. This is primarily due to the insufficient sparsity caused by filtering tokens only in the output layers, as well as inefficient sparse GEMM (General Matrix Multiplication), even when having sufficient sparsity. This paper presents Collider, a system unleashing the full efficiency of token filtering in LLM training. At its core, Collider filters activations of inconsequential tokens across all layers to maintain sparsity. Additionally, it features an automatic workflow that transforms sparse GEMM into dimension-reduced dense GEMM for optimized efficiency. Evaluations on three LLMs-TinyLlama-1.1B, Qwen2.5-1.5B, and Phi1.5-1.4B-demonstrate that Collider reduces backpropagation time by up to 35.1% and end-to-end training time by up to 22.0% when filtering 40% of tokens. Utility assessments of training TinyLlama on 15B tokens indicate that Collider sustains the utility advancements of token filtering by relatively improving model utility by 16.3% comparing to regular training, and reduces training time from 4.7 days to 3.5 days using 8 GPUs. Collider is designed for easy integration into existing LLM training frameworks, allowing systems already using token filtering to accelerate training with just one line of code.
Authors: Debdeep Sanyal, Murari Mandal
Abstract: Information removal or suppression in large language models (LLMs) is a desired functionality, useful in AI regulation, legal compliance, safety, and privacy. LLM unlearning methods aim to remove information on demand from LLMs. Current LLM unlearning methods struggle to balance the unlearning efficacy and utility due to the competing nature of these objectives. Keeping the unlearning process computationally feasible without assuming access to the model weights is an overlooked area. We present the first agentic LLM unlearning (ALU) method, a multi-agent, retrain-free, model-agnostic approach to LLM unlearning that achieves effective unlearning while preserving the utility. Our ALU framework unlearns by involving multiple LLM agents, each designed for a specific step in the unlearning process, without the need to update model weights for any of the agents in the framework. Users can easily request any set of unlearning instances in any sequence, and ALU seamlessly adapts in real time. This is facilitated without requiring any changes in the underlying LLM model. Through extensive experiments on established benchmarks (TOFU, WMDP, WPU) and jailbreaking techniques (many shot, target masking, other languages), we demonstrate that ALU consistently stands out as the most robust LLM unlearning framework among current state-of-the-art methods while incurring a low constant-time cost. We further highlight ALU's superior performance compared to existing methods when evaluated at scale. Specifically, ALU is assessed on up to 1000 unlearning targets, exceeding the evaluation scope of all previously proposed LLM unlearning methods.
Authors: Clovis Varangot-Reille, Christophe Bouvard, Antoine Gourru, Mathieu Ciancone, Marion Schaeffer, Fran\c{c}ois Jacquenet
Abstract: Large Language Models (LLM)-based systems, i.e. interconnected elements that include an LLM as a central component (e.g., conversational agents), are typically monolithic static architectures that rely on a single LLM for all user queries. However, they often require different preprocessing strategies, levels of reasoning, or knowledge. Generalist LLMs (e.g. GPT-4) trained on very large multi-topic corpora can perform well in a variety of tasks. They require significant financial, energy, and hardware resources that may not be justified for basic tasks. This implies potentially investing in unnecessary costs for a given query. To overcome this problem, a routing mechanism routes user queries to the most suitable components, such as smaller LLMs or experts in specific topics. This approach may improve response quality while minimising costs. Routing can be expanded to other components of the conversational agent architecture, such as the selection of optimal embedding strategies. This paper explores key considerations for integrating routing into LLM-based systems, focusing on resource management, cost definition, and strategy selection. Our main contributions include a formalisation of the problem, a novel taxonomy of existing approaches emphasising relevance and resource efficiency, and a comparative analysis of these strategies in relation to industry practices. Finally, we identify critical challenges and directions for future research.
Authors: George Fatouros, Kostas Metaxas, John Soldatos, Manos Karathanassis
Abstract: MarketSenseAI is a novel framework for holistic stock analysis which leverages Large Language Models (LLMs) to process financial news, historical prices, company fundamentals and the macroeconomic environment to support decision making in stock analysis and selection. In this paper, we present the latest advancements on MarketSenseAI, driven by rapid technological expansion in LLMs. Through a novel architecture combining Retrieval-Augmented Generation and LLM agents, the framework processes SEC filings and earnings calls, while enriching macroeconomic analysis through systematic processing of diverse institutional reports. We demonstrate a significant improvement in fundamental analysis accuracy over the previous version. Empirical evaluation on S\&P 100 stocks over two years (2023-2024) shows MarketSenseAI achieving cumulative returns of 125.9% compared to the index return of 73.5%, while maintaining comparable risk profiles. Further validation on S\&P 500 stocks during 2024 demonstrates the framework's scalability, delivering a 33.8% higher Sortino ratio than the market. This work marks a significant advancement in applying LLM technology to financial analysis, offering insights into the robustness of LLM-driven investment strategies.
Authors: Yingxuan Yang, Bo Huang, Siyuan Qi, Chao Feng, Haoyi Hu, Yuxuan Zhu, Jinbo Hu, Haoran Zhao, Ziyi He, Xiao Liu, Zongyu Wang, Lin Qiu, Xuezhi Cao, Xunliang Cai, Yong Yu, Weinan Zhang
Abstract: Large Language Model (LLM) agents frameworks often employ modular architectures, incorporating components such as planning, reasoning, action execution, and reflection to tackle complex tasks. However, quantifying the contribution of each module to overall system performance remains a significant challenge, impeding optimization and interpretability. To address this, we introduce CapaBench (Capability-level Assessment Benchmark), an evaluation framework grounded in cooperative game theory's Shapley Value, which systematically measures the marginal impact of individual modules and their interactions within an agent's architecture. By replacing default modules with test variants across all possible combinations, CapaBench provides a principle method for attributing performance contributions. Key contributions include: (1) We are the first to propose a Shapley Value-based methodology for quantifying the contributions of capabilities in LLM agents; (2) Modules with high Shapley Values consistently lead to predictable performance gains when combined, enabling targeted optimization; and (3) We build a multi-round dataset of over 1,000 entries spanning diverse domains and practical task scenarios, enabling comprehensive evaluation of agent capabilities. CapaBench bridges the gap between component-level evaluation and holistic system assessment, providing actionable insights for optimizing modular LLM agents and advancing their deployment in complex, real-world scenarios.
Authors: Zhi Zhou, Tan Yuhao, Zenan Li, Yuan Yao, Lan-Zhe Guo, Xiaoxing Ma, Yu-Feng Li
Abstract: Recent advancements in large language models (LLMs) have demonstrated remarkable reasoning capabilities. However, single-shot inference often yields unreliable results for complex reasoning tasks, leading researchers to explore multiple reasoning paths through methods such as perplexity and self-consistency. In this paper, we present the first theoretical error decomposition analysis of these techniques, breaking down their error into estimation error and model error. Our analysis reveals a fundamental trade-off: perplexity methods suffer from substantial model error due to the absence of a proper consistency function, while self-consistency exhibits high estimation error due to a slow error convergence rate. To overcome these limitations, we propose Reasoning-Pruning Perplexity Consistency (RPC). This approach combines Perplexity Consistency, which seamlessly integrates LLM perplexity with self-consistency, and Reasoning Pruning, which eliminates low-probability reasoning paths to effectively prevent the degeneration of estimation error reduction. Theoretical analysis demonstrates that RPC not only accelerates the convergence rate of estimation error to an exponential level but also holds strong potential for further reducing model error. Extensive empirical evaluations on seven benchmark datasets confirm that RPC can significantly improve reasoning performance, sample efficiency, and confidence reliability.
Authors: Songhao Wu, Ang Lv, Xiao Feng, Yufei Zhang, Xun Zhang, Guojun Yin, Wei Lin, Rui Yan
Abstract: The KV cache in large language models is a dominant factor in memory usage, limiting their broader applicability. Quantizing the cache to lower bit widths is an effective way to reduce computational costs; however, previous methods struggle with quantizing key vectors due to outliers, resulting in excessive overhead. We propose a novel quantization approach called PolarQuant, which efficiently addresses the outlier challenge. We observe that outliers typically appear in only one of two dimensions, which are rotated together by a specific angle when rotary position embeddings are applied. When represented as two-dimensional vectors, these dimensions exhibit well-structured patterns, with radii and angles smoothly distributed in polar coordinates. This alleviates the challenge of outliers on per-channel quantization, making them well-suited for quantization. Thus, PolarQuant divides key vectors into groups of two-dimensional sub-vectors, encoding them as the corresponding quantized radius and the polar angle, rather than quantizing original key vectors directly. PolarQuant achieves the superior efficiency in KV cache quantization and accelerates the decoding process by turning the query-key inner product into a table lookup, all while maintaining the downstream performance of full-precision models.
Authors: Zachary Huemann, Samuel Church, Joshua D. Warner, Daniel Tran, Xin Tie, Alan B McMillan, Junjie Hu, Steve Y. Cho, Meghan Lubner, Tyler J. Bradshaw
Abstract: Vision-language models can connect the text description of an object to its specific location in an image through visual grounding. This has potential applications in enhanced radiology reporting. However, these models require large annotated image-text datasets, which are lacking for PET/CT. We developed an automated pipeline to generate weak labels linking PET/CT report descriptions to their image locations and used it to train a 3D vision-language visual grounding model. Our pipeline finds positive findings in PET/CT reports by identifying mentions of SUVmax and axial slice numbers. From 25,578 PET/CT exams, we extracted 11,356 sentence-label pairs. Using this data, we trained ConTEXTual Net 3D, which integrates text embeddings from a large language model with a 3D nnU-Net via token-level cross-attention. The model's performance was compared against LLMSeg, a 2.5D version of ConTEXTual Net, and two nuclear medicine physicians. The weak-labeling pipeline accurately identified lesion locations in 98% of cases (246/251), with 7.5% requiring boundary adjustments. ConTEXTual Net 3D achieved an F1 score of 0.80, outperforming LLMSeg (F1=0.22) and the 2.5D model (F1=0.53), though it underperformed both physicians (F1=0.94 and 0.91). The model achieved better performance on FDG (F1=0.78) and DCFPyL (F1=0.75) exams, while performance dropped on DOTATE (F1=0.58) and Fluciclovine (F1=0.66). The model performed consistently across lesion sizes but showed reduced accuracy on lesions with low uptake. Our novel weak labeling pipeline accurately produced an annotated dataset of PET/CT image-text pairs, facilitating the development of 3D visual grounding models. ConTEXTual Net 3D significantly outperformed other models but fell short of the performance of nuclear medicine physicians. Our study suggests that even larger datasets may be needed to close this performance gap.
Authors: Changdae Oh, Zhen Fang, Shawn Im, Xuefeng Du, Yixuan Li
Abstract: Multimodal large language models (MLLMs) have shown promising capabilities but struggle under distribution shifts, where evaluation data differ from instruction tuning distributions. Although previous works have provided empirical evaluations, we argue that establishing a formal framework that can characterize and quantify the risk of MLLMs is necessary to ensure the safe and reliable application of MLLMs in the real world. By taking an information-theoretic perspective, we propose the first theoretical framework that enables the quantification of the maximum risk of MLLMs under distribution shifts. Central to our framework is the introduction of Effective Mutual Information (EMI), a principled metric that quantifies the relevance between input queries and model responses. We derive an upper bound for the EMI difference between in-distribution (ID) and out-of-distribution (OOD) data, connecting it to visual and textual distributional discrepancies. Extensive experiments on real benchmark datasets, spanning 61 shift scenarios empirically validate our theoretical insights.
Authors: Stuart Armstrong, Matija Franklin, Connor Stevens, Rebecca Gorman
Abstract: Recent work showed Best-of-N (BoN) jailbreaking using repeated use of random augmentations (such as capitalization, punctuation, etc) is effective against all major large language models (LLMs). We have found that $100\%$ of the BoN paper's successful jailbreaks (confidence interval $[99.65\%, 100.00\%]$) and $99.8\%$ of successful jailbreaks in our replication (confidence interval $[99.28\%, 99.98\%]$) were blocked with our Defense Against The Dark Prompts (DATDP) method. The DATDP algorithm works by repeatedly utilizing an evaluation LLM to evaluate a prompt for dangerous or manipulative behaviors--unlike some other approaches, DATDP also explicitly looks for jailbreaking attempts--until a robust safety rating is generated. This success persisted even when utilizing smaller LLMs to power the evaluation (Claude and LLaMa-3-8B-instruct proved almost equally capable). These results show that, though language models are sensitive to seemingly innocuous changes to inputs, they seem also capable of successfully evaluating the dangers of these inputs. Versions of DATDP can therefore be added cheaply to generative AI systems to produce an immediate significant increase in safety.
Authors: Jie Zhang, Kuan-Chieh Wang, Bo-Wei Chiu, Min-Te Sun
Abstract: Recent advances in deep learning have established Transformer architectures as the predominant modeling paradigm. Central to the success of Transformers is the self-attention mechanism, which scores the similarity between query and key matrices to modulate a value matrix. This operation bears striking similarities to digraph convolution, prompting an investigation into whether digraph convolution could serve as an alternative to self-attention. In this study, we formalize this concept by introducing a synthetic unitary digraph convolution based on the digraph Fourier transform. The resulting model, which we term Converter, effectively converts a Transformer into a Directed Graph Neural Network (DGNN) form. We have tested Converter on Long-Range Arena benchmark, long document classification, and DNA sequence-based taxonomy classification. Our experimental results demonstrate that Converter achieves superior performance while maintaining computational efficiency and architectural simplicity, which establishes it as a lightweight yet powerful Transformer variant.
Authors: Yi Jiang, Oubo Ma, Yong Yang, Tong Zhang, Shouling Ji
Abstract: Human language encompasses a wide range of intricate and diverse implicit features, which attackers can exploit to launch adversarial or backdoor attacks, compromising DNN models for NLP tasks. Existing model-oriented defenses often require substantial computational resources as model size increases, whereas sample-oriented defenses typically focus on specific attack vectors or schemes, rendering them vulnerable to adaptive attacks. We observe that the root cause of both adversarial and backdoor attacks lies in the encoding process of DNN models, where subtle textual features, negligible for human comprehension, are erroneously assigned significant weight by less robust or trojaned models. Based on it we propose a unified and adaptive defense framework that is effective against both adversarial and backdoor attacks. Our approach leverages reformulation modules to address potential malicious features in textual inputs while preserving the original semantic integrity. Extensive experiments demonstrate that our framework outperforms existing sample-oriented defense baselines across a diverse range of malicious textual features.
Authors: Yimu Wang, Evelien Riddell, Adrian Chow, Sean Sedwards, Krzysztof Czarnecki
Abstract: Existing vision-language model (VLM)-based methods for out-of-distribution (OOD) detection typically rely on similarity scores between input images and in-distribution (ID) text prototypes. However, the modality gap between image and text often results in high false positive rates, as OOD samples can exhibit high similarity to ID text prototypes. To mitigate the impact of this modality gap, we propose incorporating ID image prototypes along with ID text prototypes. We present theoretical analysis and empirical evidence indicating that this approach enhances VLM-based OOD detection performance without any additional training. To further reduce the gap between image and text, we introduce a novel few-shot tuning framework, SUPREME, comprising biased prompts generation (BPG) and image-text consistency (ITC) modules. BPG enhances image-text fusion and improves generalization by conditioning ID text prototypes on the Gaussian-based estimated image domain bias; ITC reduces the modality gap by minimizing intra- and inter-modal distances. Moreover, inspired by our theoretical and empirical findings, we introduce a novel OOD score $S_{\textit{GMP}}$, leveraging uni- and cross-modal similarities. Finally, we present extensive experiments to demonstrate that SUPREME consistently outperforms existing VLM-based OOD detection methods.
Authors: Hyeong Kyu Choi, Maxim Khanov, Hongxin Wei, Yixuan Li
Abstract: Dataset contamination, where evaluation datasets overlap with pre-training corpora, inflates performance metrics and undermines the reliability of model evaluations. Quantifying dataset contamination thus becomes essential to ensure that performance evaluations genuinely reflect a model's ability to generalize to unseen data, rather than relying on memorized examples. To address this problem, we propose Kernel Divergence Score (KDS), a novel method that quantifies dataset contamination by computing the divergence between the kernel similarity matrix of sample embeddings, before and after fine-tuning on the benchmark dataset. Leveraging the insight that fine-tuning affects unseen samples more significantly than seen ones, KDS provides a reliable measure of contamination. Through extensive experiments on controlled contamination scenarios, KDS demonstrates a near-perfect correlation with contamination levels and outperforms existing baselines. Additionally, we perform comprehensive ablation studies to analyze the impact of key design choices, providing deeper insights into the components and effectiveness of KDS. These ablations highlight the importance of leveraging fine-grained kernel-based information and confirm the reliability of the proposed framework across diverse datasets and settings.
Authors: Qika Lin, Zhen Peng, Kaize Shi, Kai He, Yiming Xu, Erik Cambria, Mengling Feng
Abstract: Recent years have witnessed rapid advances in graph representation learning, with the continuous embedding approach emerging as the dominant paradigm. However, such methods encounter issues regarding parameter efficiency, interpretability, and robustness. Thus, Quantized Graph Representation (QGR) learning has recently gained increasing interest, which represents the graph structure with discrete codes instead of conventional continuous embeddings. Given its analogous representation form to natural language, QGR also possesses the capability to seamlessly integrate graph structures with large language models (LLMs). As this emerging paradigm is still in its infancy yet holds significant promise, we undertake this thorough survey to promote its rapid future prosperity. We first present the background of the general quantization methods and their merits. Moreover, we provide an in-depth demonstration of current QGR studies from the perspectives of quantized strategies, training objectives, distinctive designs, knowledge graph quantization, and applications. We further explore the strategies for code dependence learning and integration with LLMs. At last, we give discussions and conclude future directions, aiming to provide a comprehensive picture of QGR and inspire future research.
Authors: Haozhe Wang, Long Li, Chao Qu, Fengming Zhu, Weidi Xu, Wei Chu, Fangzhen Lin
Abstract: Recent research on tool integration for math Large Language Models (LLMs) aims to combine complementary strengths of chain-of-thought (CoT) reasoning and code execution. However, we discover a critical limitation: current tool-integrated math LLMs rely on externally dictated instructions to decide whether to use CoT or code, lacking the autonomy to choose the most appropriate method independently. This prompts us to study \emph{Autonomous Code integration} for math LLMs, which enables models to \emph{independently} develop their own methodology-selection strategy in the absence of reliable supervision. To address this challenge, we propose an innovative Expectation-Maximization (EM) formulation that refines the model's decision-making through the exploration of its capabilities. This framework alternates between (a) computing a reference strategy that improves the model's belief over its capabilities through self-exploration, and (b) updating the model based on the refined belief. We further enhance this framework with an efficient implementation, incorporating a novel data synthesis strategy and off-policy reinforcement learning. Extensive experiments demonstrate that our approach, using only a public query set, significantly boosts the performance of existing math LLMs, raising accuracy by nearly 20\% to 65.28\% on the challenging MATH benchmark, while reducing code executions by up to 65\% .
Authors: Ivica Nikolic, Teodora Baluta, Prateek Saxena
Abstract: Large language models are increasingly customized through fine-tuning and other adaptations, creating challenges in enforcing licensing terms and managing downstream impacts. Tracking model origins is crucial both for protecting intellectual property and for identifying derived models when biases or vulnerabilities are discovered in foundation models. We address this challenge by developing a framework for testing model provenance: Whether one model is derived from another. Our approach is based on the key observation that real-world model derivations preserve significant similarities in model outputs that can be detected through statistical analysis. Using only black-box access to models, we employ multiple hypothesis testing to compare model similarities against a baseline established by unrelated models. On two comprehensive real-world benchmarks spanning models from 30M to 4B parameters and comprising over 600 models, our tester achieves 90-95% precision and 80-90% recall in identifying derived models. These results demonstrate the viability of systematic provenance verification in production environments even when only API access is available.
Authors: Divya Jyoti Bajpai, Manjesh Kumar Hanawal
Abstract: Early Exit (EE) techniques have emerged as a means to reduce inference latency in Deep Neural Networks (DNNs). The latency improvement and accuracy in these techniques crucially depend on the criteria used to make exit decisions. We propose a new decision criterion where exit classifiers are treated as experts BEEM and aggregate their confidence scores. The confidence scores are aggregated only if neighbouring experts are consistent in prediction as the samples pass through them, thus capturing their ensemble effect. A sample exits when the aggregated confidence value exceeds a threshold. The threshold is set using the error rates of the intermediate exits aiming to surpass the performance of conventional DNN inference. Experimental results on the COCO dataset for Image captioning and GLUE datasets for various language tasks demonstrate that our method enhances the performance of state-of-the-art EE methods, achieving improvements in speed-up by a factor 1.5x to 2.1x. When compared to the final layer, its accuracy is comparable in harder Image Captioning and improves in the easier language tasks. The source code for this work is publicly available at https://github.com/Div290/BEEM1/tree/main
Authors: Giovanni Pio Delvecchio, Huy Hong Nguyen, Isao Echizen
Abstract: The widespread prevalence of misinformation poses significant societal concerns. Out-of-context misinformation, where authentic images are paired with false text, is particularly deceptive and easily misleads audiences. Most existing detection methods primarily evaluate image-text consistency but often lack sufficient explanations, which are essential for effectively debunking misinformation. We present a model that detects multimodal misinformation through cross-modality consistency checks, requiring minimal training time. Additionally, we propose a lightweight model that achieves competitive performance using only one-third of the parameters. We also introduce a dual-purpose zero-shot learning task for generating contextualized warnings, enabling automated debunking and enhancing user comprehension. Qualitative and human evaluations of the generated warnings highlight both the potential and limitations of our approach.
Authors: Jianfeng Cai, Jinhua Zhu, Ruopei Sun, Yue Wang, Li Li, Wengang Zhou, Houqiang Li
Abstract: Reinforcement Learning from Human Feedback (RLHF) has achieved considerable success in aligning large language models (LLMs) by modeling human preferences with a learnable reward model and employing a reinforcement learning algorithm to maximize the reward model's scores. However, these reward models are susceptible to exploitation through various superficial confounding factors, with length bias emerging as a particularly significant concern. Moreover, while the pronounced impact of length bias on preference modeling suggests that LLMs possess an inherent sensitivity to length perception, our preliminary investigations reveal that fine-tuned LLMs consistently struggle to adhere to explicit length instructions. To address these two limitations, we propose a novel framework wherein the reward model explicitly differentiates between human semantic preferences and response length requirements. Specifically, we introduce a Response-conditioned Bradley-Terry (Rc-BT) model that enhances the reward model's capability in length bias mitigating and length instruction following, through training on our augmented dataset. Furthermore, we propose the Rc-DPO algorithm to leverage the Rc-BT model for direct policy optimization (DPO) of LLMs, simultaneously mitigating length bias and promoting adherence to length instructions. Extensive evaluations demonstrate that our approach substantially improves both preference modeling and length instruction compliance, with its effectiveness validated across various foundational models and preference datasets.
Authors: Subhash Kantamneni, Max Tegmark
Abstract: Mathematical reasoning is an increasingly important indicator of large language model (LLM) capabilities, yet we lack understanding of how LLMs process even simple mathematical tasks. To address this, we reverse engineer how three mid-sized LLMs compute addition. We first discover that numbers are represented in these LLMs as a generalized helix, which is strongly causally implicated for the tasks of addition and subtraction, and is also causally relevant for integer division, multiplication, and modular arithmetic. We then propose that LLMs compute addition by manipulating this generalized helix using the "Clock" algorithm: to solve $a+b$, the helices for $a$ and $b$ are manipulated to produce the $a+b$ answer helix which is then read out to model logits. We model influential MLP outputs, attention head outputs, and even individual neuron preactivations with these helices and verify our understanding with causal interventions. By demonstrating that LLMs represent numbers on a helix and manipulate this helix to perform addition, we present the first representation-level explanation of an LLM's mathematical capability.
Authors: Teng Xiao, Yige Yuan, Zhengyu Chen, Mingxiao Li, Shangsong Liang, Zhaochun Ren, Vasant G Honavar
Abstract: Existing preference optimization objectives for language model alignment require additional hyperparameters that must be extensively tuned to achieve optimal performance, increasing both the complexity and time required for fine-tuning large language models. In this paper, we propose a simple yet effective hyperparameter-free preference optimization algorithm for alignment. We observe that promising performance can be achieved simply by optimizing inverse perplexity, which is calculated as the inverse of the exponentiated average log-likelihood of the chosen and rejected responses in the preference dataset. The resulting simple learning objective, SimPER, is easy to implement and eliminates the need for expensive hyperparameter tuning and a reference model, making it both computationally and memory efficient. Extensive experiments on widely used real-world benchmarks, including MT-Bench, AlpacaEval 2, and 10 key benchmarks of the Open LLM Leaderboard with 5 base models, demonstrate that SimPER consistently and significantly outperforms existing approaches-even without any hyperparameters or a reference model . For example, despite its simplicity, SimPER outperforms state-of-the-art methods by up to 5.7 points on AlpacaEval 2 and achieves the highest average ranking across 10 benchmarks on the Open LLM Leaderboard. The source code for SimPER is publicly available at: https://github.com/tengxiao1/SimPER.
Authors: Chao Wang, Xuancheng Zhou, Weiwei Fu, Yang Zhou
Abstract: Large Visual Language Models (LVLMs) integrate visual and linguistic modalities, exhibiting exceptional performance across various multimodal tasks. Nevertheless, LVLMs remain vulnerable to the issue of object hallucinations. Previous efforts to mitigate this issue focus on supervised fine-tuning (SFT) or incorporating external knowledge, both of which entail significant costs related to training and the acquisition of external data. To address these challenges, we propose a novel model-agnostic approach termed Internal Fact-based Contrastive Decoding (IFCD), designed to mitigate and suppress hallucinations during the inference process of LVLMs by exploiting the LVLMs' own hallucinations. IFCD is grounded in experimental observations that alterations to the LVLMs' internal representations tend to amplify hallucinations caused by language bias. By contrasting disturbed distribution, IFCD calibrates the LVLMs' output and effectively removes the hallucinatory logits from the final predictions. Experimental results validate that IFCD significantly alleviates both object-level and attribute-level hallucinations while achieving an average 9% accuracy improvement on POPE and 8% accuracy improvement on MME object hallucinations subset compared with direct decoding, respectively.
Authors: Dongwon Jo, Jiwon Song, Yulhwa Kim, Jae-Joon Kim
Abstract: While large language models (LLMs) excel at handling long-context sequences, they require substantial key-value (KV) caches to store contextual information, which can heavily burden computational efficiency and memory usage. Previous efforts to compress these KV caches primarily focused on reducing memory demands but were limited in enhancing latency. To address this issue, we introduce FastKV, a KV cache compression method designed to enhance latency for long-context sequences. To enhance processing speeds while maintaining accuracy, FastKV adopts a novel Token-Selective Propagation (TSP) approach that retains the full context information in the initial layers of LLMs and selectively propagates only a portion of this information in deeper layers even in the prefill stage. Additionally, FastKV incorporates grouped-query attention (GQA)-aware KV cache compression to exploit the advantages of GQA in both memory and computational efficiency. Our experimental results show that FastKV achieves 2.00$\times$ and 1.40$\times$ improvements in time-to-first-token (TTFT) and throughput, respectively, compared to HeadKV, the state-of-the-art KV cache compression method. Moreover, FastKV successfully maintains accuracy on long-context benchmarks at levels comparable to the baselines. Our code is available at https://github.com/dongwonjo/FastKV.
Authors: Vernon Y. H. Toh, Yew Ken Chia, Deepanway Ghosal, Soujanya Poria
Abstract: The releases of OpenAI's o1 and o3 mark a significant paradigm shift in Large Language Models towards advanced reasoning capabilities. Notably, o3 outperformed humans in novel problem-solving and skill acquisition on the Abstraction and Reasoning Corpus for Artificial General Intelligence (ARC-AGI). However, this benchmark is limited to symbolic patterns, whereas humans often perceive and reason about multimodal scenarios involving both vision and language data. Thus, there is an urgent need to investigate advanced reasoning capabilities in multimodal tasks. To this end, we track the evolution of the GPT-[n] and o-[n] series models on challenging multimodal puzzles, requiring fine-grained visual perception with abstract or algorithmic reasoning. The superior performance of o1 comes at nearly 750 times the computational cost of GPT-4o, raising concerns about its efficiency. Our results reveal a clear upward trend in reasoning capabilities across model iterations, with notable performance jumps across GPT-series models and subsequently to o1. Nonetheless, we observe that the o1 model still struggles with simple multimodal puzzles requiring abstract reasoning. Furthermore, its performance in algorithmic puzzles remains poor. We plan to continuously track new models in the series and update our results in this paper accordingly. All resources used in this evaluation are openly available https://github.com/declare-lab/LLM-PuzzleTest.
Authors: Jiali Cheng, Hadi Amiri
Abstract: Tool-augmented large language models (LLMs) are often trained on datasets of query-response pairs, which embed the ability to use tools or APIs directly into the parametric knowledge of LLMs. Tool-augmented LLMs need the ability to forget learned tools due to security vulnerabilities, privacy regulations, or tool deprecations. However, ``tool unlearning'' has not been investigated in unlearning literature. We introduce this novel task, which requires addressing distinct challenges compared to traditional unlearning: knowledge removal rather than forgetting individual samples, the high cost of optimizing LLMs, and the need for principled evaluation metrics. To bridge these gaps, we propose ToolDelete, the first approach for unlearning tools from tool-augmented LLMs. It implements three key properties to address the above challenges for effective tool unlearning and introduces a new membership inference attack (MIA) model for effective evaluation. Extensive experiments on multiple tool learning datasets and tool-augmented LLMs show that ToolDelete effectively unlearns randomly selected tools, while preserving the LLM's knowledge on non-deleted tools and maintaining performance on general tasks.
Authors: Bill Yuchen Lin, Ronan Le Bras, Kyle Richardson, Ashish Sabharwal, Radha Poovendran, Peter Clark, Yejin Choi
Abstract: We investigate the logical reasoning capabilities of large language models (LLMs) and their scalability in complex non-monotonic reasoning. To this end, we introduce ZebraLogic, a comprehensive evaluation framework for assessing LLM reasoning performance on logic grid puzzles derived from constraint satisfaction problems (CSPs). ZebraLogic enables the generation of puzzles with controllable and quantifiable complexity, facilitating a systematic study of the scaling limits of models such as Llama, o1 models, and DeepSeek-R1. By encompassing a broad range of search space complexities and diverse logical constraints, ZebraLogic provides a structured environment to evaluate reasoning under increasing difficulty. Our results reveal a significant decline in accuracy as problem complexity grows -- a phenomenon we term the curse of complexity. This limitation persists even with larger models and increased inference-time computation, suggesting inherent constraints in current LLM reasoning capabilities. Additionally, we explore strategies to enhance logical reasoning, including Best-of-N sampling, backtracking mechanisms, and self-verification prompts. Our findings offer critical insights into the scalability of LLM reasoning, highlight fundamental limitations, and outline potential directions for improvement.
Authors: Linhao Luo, Zicheng Zhao, Gholamreza Haffari, Dinh Phung, Chen Gong, Shirui Pan
Abstract: Retrieval-augmented generation (RAG) has proven effective in integrating knowledge into large language models (LLMs). However, conventional RAGs struggle to capture complex relationships between pieces of knowledge, limiting their performance in intricate reasoning that requires integrating knowledge from multiple sources. Recently, graph-enhanced retrieval augmented generation (GraphRAG) builds graph structure to explicitly model these relationships, enabling more effective and efficient retrievers. Nevertheless, its performance is still hindered by the noise and incompleteness within the graph structure. To address this, we introduce GFM-RAG, a novel graph foundation model (GFM) for retrieval augmented generation. GFM-RAG is powered by an innovative graph neural network that reasons over graph structure to capture complex query-knowledge relationships. The GFM with 8M parameters undergoes a two-stage training process on large-scale datasets, comprising 60 knowledge graphs with over 14M triples and 700k documents. This results in impressive performance and generalizability for GFM-RAG, making it the first graph foundation model applicable to unseen datasets for retrieval without any fine-tuning required. Extensive experiments on three multi-hop QA datasets and seven domain-specific RAG datasets demonstrate that GFM-RAG achieves state-of-the-art performance while maintaining efficiency and alignment with neural scaling laws, highlighting its potential for further improvement.
Authors: Guanlin Li, Kangjie Chen, Shangwei Guo, Jie Zhang, Han Qiu, Chao Zhang, Guoyin Wang, Tianwei Zhang, Jiwei Li
Abstract: Large language models (LLMs) have emerged as powerful tools for addressing a wide range of general inquiries and tasks. Despite this, fine-tuning aligned LLMs on smaller, domain-specific datasets, critical to adapting them to specialized tasks, can inadvertently degrade their safety alignment, even when the datasets are benign. This phenomenon makes models more susceptible to providing inappropriate responses. In this study, we systematically examine the factors contributing to safety alignment degradation in benign fine-tuning scenarios. Our analysis identifies three critical factors affecting aligned LLMs: answer structure, identity calibration, and role-play. Additionally, we evaluate the reliability of state-of-the-art reward models (RMs), which are often used to guide alignment processes. Our findings reveal that these RMs frequently fail to accurately reflect human preferences regarding safety, underscoring their limitations in practical applications. By uncovering these challenges, our work highlights the complexities of maintaining safety alignment during fine-tuning and offers guidance to help developers balance utility and safety in LLMs. Datasets and fine-tuning code used in our experiments can be found in https://github.com/GuanlinLee/llm_instruction_tuning.
Authors: Xinyan Guan, Jiali Zeng, Fandong Meng, Chunlei Xin, Yaojie Lu, Hongyu Lin, Xianpei Han, Le Sun, Jie Zhou
Abstract: Large Language Models (LLMs) have shown remarkable potential in reasoning while they still suffer from severe factual hallucinations due to timeliness, accuracy, and coverage of parametric knowledge. Meanwhile, integrating reasoning with retrieval-augmented generation (RAG) remains challenging due to ineffective task decomposition and redundant retrieval, which can introduce noise and degrade response quality. In this paper, we propose DeepRAG, a framework that models retrieval-augmented reasoning as a Markov Decision Process (MDP), enabling strategic and adaptive retrieval. By iteratively decomposing queries, DeepRAG dynamically determines whether to retrieve external knowledge or rely on parametric reasoning at each step. Experiments show that DeepRAG improves retrieval efficiency while improving answer accuracy by 21.99%, demonstrating its effectiveness in optimizing retrieval-augmented reasoning.
Authors: Hao Li, Di Huang, Ziyu Wang, Amir M. Rahmani
Abstract: Memorization in Large Language Models (LLMs) poses privacy and security risks, as models may unintentionally reproduce sensitive or copyrighted data. Existing analyses focus on average-case scenarios, often neglecting the highly skewed distribution of memorization. This paper examines memorization in LLM supervised fine-tuning (SFT), exploring its relationships with training duration, dataset size, and inter-sample similarity. By analyzing memorization probabilities over sequence lengths, we link this skewness to the token generation process, offering insights for estimating memorization and comparing it to established metrics. Through theoretical analysis and empirical evaluation, we provide a comprehensive understanding of memorization behaviors and propose strategies to detect and mitigate risks, contributing to more privacy-preserving LLMs.
Authors: Xiaotong Ji, Shyam Sundhar Ramesh, Matthieu Zimmer, Ilija Bogunovic, Jun Wang, Haitham Bou Ammar
Abstract: Even highly capable large language models (LLMs) can produce biased or unsafe responses, and alignment techniques, such as RLHF, aimed at mitigating this issue, are expensive and prone to overfitting as they retrain the LLM. This paper introduces a novel inference-time alignment approach that ensures LLMs generate safe responses almost surely, i.e., with a probability approaching one. We achieve this by framing the safe generation of inference-time responses as a constrained Markov decision process within the LLM's latent space. Crucially, we augment a safety state that tracks the evolution of safety constraints and enables us to demonstrate formal safety guarantees upon solving the MDP in the latent space. Building on this foundation, we propose InferenceGuard, a practical implementation that safely aligns LLMs without modifying the model weights. Empirically, we demonstrate InferenceGuard effectively balances safety and task performance, outperforming existing inference-time alignment methods in generating safe and aligned responses.
Authors: Xiang Lisa Li, Neil Chowdhury, Daniel D. Johnson, Tatsunori Hashimoto, Percy Liang, Sarah Schwettmann, Jacob Steinhardt
Abstract: Language models exhibit complex, diverse behaviors when prompted with free-form text, making it difficult to characterize the space of possible outputs. We study the problem of behavior elicitation, where the goal is to search for prompts that induce specific target behaviors (e.g., hallucinations or harmful responses) from a target language model. To navigate the exponentially large space of possible prompts, we train investigator models to map randomly-chosen target behaviors to a diverse distribution of outputs that elicit them, similar to amortized Bayesian inference. We do this through supervised fine-tuning, reinforcement learning via DPO, and a novel Frank-Wolfe training objective to iteratively discover diverse prompting strategies. Our investigator models surface a variety of effective and human-interpretable prompts leading to jailbreaks, hallucinations, and open-ended aberrant behaviors, obtaining a 100% attack success rate on a subset of AdvBench (Harmful Behaviors) and an 85% hallucination rate.
Authors: Ismail Khalfaoui-Hassani, Stefan Kesselheim
Abstract: This paper investigates scalable neural networks with learnable activation functions based on orthogonal function bases and tropical polynomials, targeting ImageNet-1K classification and next token prediction on OpenWebText. Traditional activations, such as ReLU, are static. In contrast, learnable activations enable the network to adapt dynamically during training. However, stability issues, such as vanishing or exploding gradients, arise with improper variance management in deeper networks. To remedy this, we propose an initialization scheme that single-handedly preserves unitary variance in transformers and convolutional networks, ensuring stable gradient flow even in deep architectures. Extensive experiments demonstrate that networks with Hermite, Fourier, and Tropical-based learnable activations significantly improve over GPT-2 and ConvNeXt networks in terms of accuracy and perplexity in train and test, highlighting the viability of learnable activations in large-scale tasks. The activation functions developed here are the subject of a library coded entirely in pure PyTorch: torchortho, available at https://github.com/K-H-Ismail/torchortho.
Authors: Hanlin Wu, Xiaohui Rao, Zhenguang G. Cai
Abstract: Listeners adapt language comprehension based on their mental representations of speakers, but how these representations are dynamically updated remains unclear. We investigated whether listeners probabilistically adapt their comprehension based on the likelihood of speakers producing stereotype-incongruent utterances. Our findings reveal two potential mechanisms: a speaker-general mechanism that adjusts overall expectations about speaker-content relationships, and a speaker-specific mechanism that updates individual speaker models. In two EEG experiments, participants heard speakers make stereotype-congruent or incongruent utterances, with incongruency base rate manipulated between blocks. In Experiment 1, speaker incongruency modulated both high-beta (21-30 Hz) and theta (4-6 Hz) oscillations: incongruent utterances decreased oscillatory power in low base rate condition but increased it in high base rate condition. The theta effect varied with listeners' openness trait: less open participants showed theta increases to speaker-incongruencies, suggesting maintenance of speaker-specific information, while more open participants showed theta decreases, indicating flexible model updating. In Experiment 2, we dissociated base rate from the target speaker by manipulating the overall base rate using an alternative non-target speaker. Only the high-beta effect persisted, showing power decrease for speaker-incongruencies in low base rate condition but no effect in high base rate condition. The high-beta oscillations might reflect the speaker-general adjustment, while theta oscillations may index the speaker-specific model updating. These findings provide evidence for how language processing is shaped by social cognition in real time.
Authors: Jinzhi Liao, Zenghua Liao, Xiang Zhao
Abstract: The enhance of accuracy in reasoning results of LLMs arouses the community's interests, wherein pioneering studies investigate post-hoc strategies to rectify potential mistakes. Despite extensive efforts, they are all stuck in a state of resource competition demanding significant time and computing expenses. The cause of the situation lies in the failure of identifying the fundamental feature of the solutions in this line, coined as the self-denial of LLMs. In other words, LLMs should confidently determine the potential existence of mistakes and carefully execute the targeted correction. As the whole procedure conducts within LLMs, supporting and persuasive references are hard to acquire, while the absence of specific steps towards refining hidden mistakes persists even when errors are acknowledged. In response to the challenges, we present PSSD, which refers to and implements the human psyche structure such that three distinct and interconnected roles contribute to human reasoning. Specifically, PSSD leverages the recent multi-agent paradigm, and is further enhanced with three innovatively conceived roles: (1) the intuition-based id role that provides initial attempts based on benign LLMs; (2) the rule-driven superego role that summarizes rules to regulate the above attempts, and returns specific key points as guidance; and (3) the script-centric ego role that absorbs all procedural information to generate executable script for the final answer prediction. Extensive experiments demonstrate that the proposed design not only better enhance reasoning capabilities, but also seamlessly integrate with current models, leading to superior performance.
Authors: Abhinav Pratap, Amit Pathak
Abstract: In an era dominated by datafication, the reduction of human experiences to quantifiable metrics raises profound philosophical and ethical questions. This paper explores these issues through the lens of Meursault, the protagonist of Albert Camus' The Stranger, whose emotionally detached existence epitomizes the existential concept of absurdity. Using natural language processing (NLP) techniques including emotion detection (BERT), sentiment analysis (VADER), and named entity recognition (spaCy)-this study quantifies key events and behaviors in Meursault's life. Our analysis reveals the inherent limitations of applying algorithmic models to complex human experiences, particularly those rooted in existential alienation and moral ambiguity. By examining how modern AI tools misinterpret Meursault's actions and emotions, this research underscores the broader ethical dilemmas of reducing nuanced human narratives to data points, challenging the foundational assumptions of our data-driven society. The findings presented in this paper serve as a critique of the increasing reliance on data-driven narratives and advocate for incorporating humanistic values in artificial intelligence.
Authors: Oussama Zekri, Nicolas Boull\'e
Abstract: Discrete diffusion models have recently gained significant attention due to their ability to process complex discrete structures for language modeling. However, fine-tuning these models with policy gradient methods, as is commonly done in Reinforcement Learning from Human Feedback (RLHF), remains a challenging task. We propose an efficient, broadly applicable, and theoretically justified policy gradient algorithm, called Score Entropy Policy Optimization (SEPO), for fine-tuning discrete diffusion models over non-differentiable rewards. Our numerical experiments across several discrete generative tasks demonstrate the scalability and efficiency of our method. Our code is available at https://github.com/ozekri/SEPO
Authors: Gaole He, Gianluca Demartini, Ujwal Gadiraju
Abstract: Since the explosion in popularity of ChatGPT, large language models (LLMs) have continued to impact our everyday lives. Equipped with external tools that are designed for a specific purpose (e.g., for flight booking or an alarm clock), LLM agents exercise an increasing capability to assist humans in their daily work. Although LLM agents have shown a promising blueprint as daily assistants, there is a limited understanding of how they can provide daily assistance based on planning and sequential decision making capabilities. We draw inspiration from recent work that has highlighted the value of 'LLM-modulo' setups in conjunction with humans-in-the-loop for planning tasks. We conducted an empirical study (N = 248) of LLM agents as daily assistants in six commonly occurring tasks with different levels of risk typically associated with them (e.g., flight ticket booking and credit card payments). To ensure user agency and control over the LLM agent, we adopted LLM agents in a plan-then-execute manner, wherein the agents conducted step-wise planning and step-by-step execution in a simulation environment. We analyzed how user involvement at each stage affects their trust and collaborative team performance. Our findings demonstrate that LLM agents can be a double-edged sword -- (1) they can work well when a high-quality plan and necessary user involvement in execution are available, and (2) users can easily mistrust the LLM agents with plans that seem plausible. We synthesized key insights for using LLM agents as daily assistants to calibrate user trust and achieve better overall task outcomes. Our work has important implications for the future design of daily assistants and human-AI collaboration with LLM agents.
Authors: Zhiteng Li, Mingyuan Xia, Jingyuan Zhang, Zheng Hui, Linghe Kong, Yulun Zhang, Xiaokang Yang
Abstract: Large language models (LLMs) have achieved remarkable success in natural language processing (NLP) tasks, yet their substantial memory requirements present significant challenges for deployment on resource-constrained devices. Singular Value Decomposition (SVD) has emerged as a promising compression technique for LLMs, offering considerable reductions in memory overhead. However, existing SVD-based methods often struggle to effectively mitigate the errors introduced by SVD truncation, leading to a noticeable performance gap when compared to the original models. Furthermore, applying a uniform compression ratio across all transformer layers fails to account for the varying importance of different layers. To address these challenges, we propose AdaSVD, an adaptive SVD-based LLM compression approach. Specifically, AdaSVD introduces adaComp, which adaptively compensates for SVD truncation errors by alternately updating the singular matrices U and V^T. Additionally, AdaSVD introduces adaCR, which adaptively assigns layer-specific compression ratios based on the relative importance of each layer. Extensive experiments across multiple LLM families and evaluation metrics demonstrate that AdaSVD consistently outperforms state-of-the-art (SOTA) SVD-based methods, achieving superior performance with significantly reduced memory requirements. The code and models will be available at https://github.com/ZHITENGLI/AdaSVD.
Authors: Jonathan Drechsel, Steffen Herbold
Abstract: AI systems frequently exhibit and amplify social biases, including gender bias, leading to harmful consequences in critical areas. This study introduces a novel encoder-decoder approach that leverages model gradients to learn a single monosemantic feature neuron encoding gender information. We show that our method can be used to debias transformer-based language models, while maintaining other capabilities. We demonstrate the effectiveness of our approach across multiple encoder-only based models and highlight its potential for broader applications.
Authors: Jack H. Culbert, Yoed N. Kenett, Philipp Mayr
Abstract: In this research-in-progress paper, we apply a computational measure correlating with originality from creativity science: Divergent Semantic Integration (DSI), to a selection of 99,557 scientific abstracts and titles selected from the Web of Science. We observe statistically significant differences in DSI between subject and field of research, and a slight rise in DSI over time. We model the base 10 logarithm of the citation count after 5 years with DSI and find a statistically significant positive correlation in all fields of research with an adjusted $R^2$ of 0.13.
Authors: Ganqu Cui, Lifan Yuan, Zefan Wang, Hanbin Wang, Wendi Li, Bingxiang He, Yuchen Fan, Tianyu Yu, Qixin Xu, Weize Chen, Jiarui Yuan, Huayu Chen, Kaiyan Zhang, Xingtai Lv, Shuo Wang, Yuan Yao, Xu Han, Hao Peng, Yu Cheng, Zhiyuan Liu, Maosong Sun, Bowen Zhou, Ning Ding
Abstract: Dense process rewards have proven a more effective alternative to the sparse outcome-level rewards in the inference-time scaling of large language models (LLMs), particularly in tasks requiring complex multi-step reasoning. While dense rewards also offer an appealing choice for the reinforcement learning (RL) of LLMs since their fine-grained rewards have the potential to address some inherent issues of outcome rewards, such as training efficiency and credit assignment, this potential remains largely unrealized. This can be primarily attributed to the challenges of training process reward models (PRMs) online, where collecting high-quality process labels is prohibitively expensive, making them particularly vulnerable to reward hacking. To address these challenges, we propose PRIME (Process Reinforcement through IMplicit rEwards), which enables online PRM updates using only policy rollouts and outcome labels through implict process rewards. PRIME combines well with various advantage functions and forgoes the dedicated reward model training phrase that existing approaches require, substantially reducing the development overhead. We demonstrate PRIME's effectiveness on competitional math and coding. Starting from Qwen2.5-Math-7B-Base, PRIME achieves a 15.1% average improvement across several key reasoning benchmarks over the SFT model. Notably, our resulting model, Eurus-2-7B-PRIME, surpasses Qwen2.5-Math-7B-Instruct on seven reasoning benchmarks with 10% of its training data.
Authors: Jingzhe Shi, Qinwei Ma, Hongyi Liu, Hang Zhao, Jeng-Neng Hwang, Serge Belongie, Lei Li
Abstract: Long Context Language Models have drawn great attention in the past few years. There has been work discussing the impact of long context on Language Model performance: some find that long irrelevant context could harm performance, while some experimentally summarize loss reduction by relevant long context as Scaling Laws. This calls for a more thorough understanding on how long context impact Language Modeling. In this work, we (1) propose a clean and effective theoretical framework on explaining the impact of context length to Language Modeling, from an Intrinsic Space perspective; and (2) conduct experiments on natural language and synthetic data, validating our proposed theoretical assumptions and deductions. Our theoretical framework can provide practical insights such as establishing that training dataset size dictates an optimal context length and bounds context length scaling for certain case. We hope our work may inspire new long context Language Models, as well as future work studying Physics for Language Models. Code for our experiments is available at this url: https://github.com/JingzheShi/NLPCtlScalingAndBounds.
Authors: Kelsey Allen, Ishita Dasgupta, Eliza Kosoy, Andrew K. Lampinen
Abstract: Inductive biases are what allow learners to make guesses in the absence of conclusive evidence. These biases have often been studied in cognitive science using concepts or categories -- e.g. by testing how humans generalize a new category from a few examples that leave the category boundary ambiguous. We use these approaches to study generalization in foundation models during in-context learning. Modern foundation models can condition on both vision and text, and differences in how they interpret and learn from these different modalities is an emerging area of study. Here, we study how their generalizations vary by the modality in which stimuli are presented, and the way the stimuli are described in text. We study these biases with three different experimental paradigms, across three different vision-language models. We find that the models generally show some bias towards generalizing according to shape over color. This shape bias tends to be amplified when the examples are presented visually. By contrast, when examples are presented in text, the ordering of adjectives affects generalization. However, the extent of these effects vary across models and paradigms. These results help to reveal how vision-language models represent different types of inputs in context, and may have practical implications for the use of vision-language models.
Authors: Dawei Li, Renliang Sun, Yue Huang, Ming Zhong, Bohan Jiang, Jiawei Han, Xiangliang Zhang, Wei Wang, Huan Liu
Abstract: Large Language Models (LLMs) as judges and LLM-based data synthesis have emerged as two fundamental LLM-driven data annotation methods in model development. While their combination significantly enhances the efficiency of model training and evaluation, little attention has been given to the potential contamination brought by this new model development paradigm. In this work, we expose preference leakage, a contamination problem in LLM-as-a-judge caused by the relatedness between the synthetic data generators and LLM-based evaluators. To study this issue, we first define three common relatednesses between data generator LLM and judge LLM: being the same model, having an inheritance relationship, and belonging to the same model family. Through extensive experiments, we empirically confirm the bias of judges towards their related student models caused by preference leakage across multiple LLM baselines and benchmarks. Further analysis suggests that preference leakage is a pervasive issue that is harder to detect compared to previously identified biases in LLM-as-a-judge scenarios. All of these findings imply that preference leakage is a widespread and challenging problem in the area of LLM-as-a-judge. We release all codes and data at: https://github.com/David-Li0406/Preference-Leakage.
Authors: Duy-Cat Can, Linh D. Dang, Quang-Huy Tang, Dang Minh Ly, Huong Ha, Guillaume Blanc, Oliver Y. Ch\'en, Binh T. Nguyen
Abstract: Objective: Assessing Alzheimer's disease (AD) using high-dimensional radiology images is clinically important but challenging. Although Artificial Intelligence (AI) has advanced AD diagnosis, it remains unclear how to design AI models embracing predictability and explainability. Here, we propose VisTA, a multimodal language-vision model assisted by contrastive learning, to optimize disease prediction and evidence-based, interpretable explanations for clinical decision-making. Methods: We developed VisTA (Vision-Text Alignment Model) for AD diagnosis. Architecturally, we built VisTA from BiomedCLIP and fine-tuned it using contrastive learning to align images with verified abnormalities and their descriptions. To train VisTA, we used a constructed reference dataset containing images, abnormality types, and descriptions verified by medical experts. VisTA produces four outputs: predicted abnormality type, similarity to reference cases, evidence-driven explanation, and final AD diagnoses. To illustrate VisTA's efficacy, we reported accuracy metrics for abnormality retrieval and dementia prediction. To demonstrate VisTA's explainability, we compared its explanations with human experts' explanations. Results: Compared to 15 million images used for baseline pretraining, VisTA only used 170 samples for fine-tuning and obtained significant improvement in abnormality retrieval and dementia prediction. For abnormality retrieval, VisTA reached 74% accuracy and an AUC of 0.87 (26% and 0.74, respectively, from baseline models). For dementia prediction, VisTA achieved 88% accuracy and an AUC of 0.82 (30% and 0.57, respectively, from baseline models). The generated explanations agreed strongly with human experts' and provided insights into the diagnostic process. Taken together, VisTA optimize prediction, clinical reasoning, and explanation.
Authors: Archiki Prasad, Elias Stengel-Eskin, Justin Chih-Yao Chen, Zaid Khan, Mohit Bansal
Abstract: Unit tests (UTs) play an instrumental role in assessing code correctness as well as providing feedback to a large language model (LLM) as it iteratively debugs faulty code, motivating automated test generation. However, we uncover a trade-off between generating unit test inputs that reveal errors when given a faulty code and correctly predicting the unit test output without access to the gold solution. To address this trade-off, we propose UTGen, which teaches LLMs to generate unit test inputs that reveal errors along with their correct expected outputs based on task descriptions and candidate code. We integrate UTGen into UTDebug, a robust debugging pipeline that uses generated tests to help LLMs debug effectively. Since model-generated tests can provide noisy signals (e.g., from incorrectly predicted outputs), UTDebug (i) scales UTGen via test-time compute to improve UT output prediction, and (ii) validates and back-tracks edits based on multiple generated UTs to avoid overfitting. We show that UTGen outperforms UT generation baselines by 7.59% based on a metric measuring the presence of both error-revealing UT inputs and correct UT outputs. When used with UTDebug, we find that feedback from UTGen's unit tests improves pass@1 accuracy of Qwen-2.5 7B on HumanEvalFix and our own harder debugging split of MBPP+ by over 3% and 12.35% (respectively) over other LLM-based UT generation baselines.
Authors: Marco Bombieri, Paolo Fiorini, Simone Paolo Ponzetto, Marco Rospocher
Abstract: Large Language Models (LLMs) have demonstrated remarkable performance across diverse natural language processing tasks, yet their ability to memorize structured knowledge remains underexplored. In this paper, we investigate the extent to which general-purpose pre-trained LLMs retain and correctly reproduce concept identifier (ID)-label associations from publicly available ontologies. We conduct a systematic evaluation across multiple ontological resources, including the Gene Ontology, Uberon, Wikidata, and ICD-10, using LLMs such as Pythia-12B, Gemini-1.5-Flash, GPT-3.5, and GPT-4. Our findings reveal that only a small fraction of ontological concepts is accurately memorized, with GPT-4 demonstrating the highest performance. To understand why certain concepts are memorized more effectively than others, we analyze the relationship between memorization accuracy and concept popularity on the Web. Our results indicate a strong correlation between the frequency of a concept's occurrence online and the likelihood of accurately retrieving its ID from the label. This suggests that LLMs primarily acquire such knowledge through indirect textual exposure rather than directly from structured ontological resources. Furthermore, we introduce new metrics to quantify prediction invariance, demonstrating that the stability of model responses across variations in prompt language and temperature settings can serve as a proxy for estimating memorization robustness.
Authors: Chenkai Sun, Ke Yang, Revanth Gangi Reddy, Yi R. Fung, Hou Pong Chan, Kevin Small, ChengXiang Zhai, Heng Ji
Abstract: The increasing demand for personalized interactions with large language models (LLMs) calls for methodologies capable of accurately and efficiently identifying user opinions and preferences. Retrieval augmentation emerges as an effective strategy, as it can accommodate a vast number of users without the costs from fine-tuning. Existing research, however, has largely focused on enhancing the retrieval stage and devoted limited exploration toward optimizing the representation of the database, a crucial aspect for tasks such as personalization. In this work, we examine the problem from a novel angle, focusing on how data can be better represented for more data-efficient retrieval in the context of LLM customization. To tackle this challenge, we introduce Persona-DB, a simple yet effective framework consisting of a hierarchical construction process to improve generalization across task contexts and collaborative refinement to effectively bridge knowledge gaps among users. In the evaluation of response prediction, Persona-DB demonstrates superior context efficiency in maintaining accuracy with a significantly reduced retrieval size, a critical advantage in scenarios with extensive histories or limited context windows. Our experiments also indicate a marked improvement of over 10% under cold-start scenarios, when users have extremely sparse data. Furthermore, our analysis reveals the increasing importance of collaborative knowledge as the retrieval capacity expands.
Authors: Hanjie Chen, Zhouxiang Fang, Yash Singla, Mark Dredze
Abstract: LLMs have demonstrated impressive performance in answering medical questions, such as achieving passing scores on medical licensing examinations. However, medical board exams or general clinical questions do not capture the complexity of realistic clinical cases. Moreover, the lack of reference explanations means we cannot easily evaluate the reasoning of model decisions, a crucial component of supporting doctors in making complex medical decisions. To address these challenges, we construct two new datasets: JAMA Clinical Challenge and Medbullets.\footnote{Datasets and code are available at \url{https://github.com/HanjieChen/ChallengeClinicalQA}.} JAMA Clinical Challenge consists of questions based on challenging clinical cases, while Medbullets comprises simulated clinical questions. Both datasets are structured as multiple-choice question-answering tasks, accompanied by expert-written explanations. We evaluate seven LLMs on the two datasets using various prompts. Experiments demonstrate that our datasets are harder than previous benchmarks. In-depth automatic and human evaluations of model-generated explanations provide insights into the promise and deficiency of LLMs for explainable medical QA.
Authors: Minbeom Kim, Hwanhee Lee, Joonsuk Park, Hwaran Lee, Kyomin Jung
Abstract: As the integration of large language models into daily life is on the rise, there is a clear gap in benchmarks for advising on subjective and personal dilemmas. To address this, we introduce AdvisorQA, the first benchmark developed to assess LLMs' capability in offering advice for deeply personalized concerns, utilizing the LifeProTips subreddit forum. This forum features a dynamic interaction where users post advice-seeking questions, receiving an average of 8.9 advice per query, with 164.2 upvotes from hundreds of users, embodying a collective intelligence framework. Therefore, we've completed a benchmark encompassing daily life questions, diverse corresponding responses, and majority vote ranking to train our helpfulness metric. Baseline experiments validate the efficacy of AdvisorQA through our helpfulness metric, GPT-4, and human evaluation, analyzing phenomena beyond the trade-off between helpfulness and harmlessness. AdvisorQA marks a significant leap in enhancing QA systems for providing personalized, empathetic advice, showcasing LLMs' improved understanding of human subjectivity.
Authors: Hadi Askari, Anshuman Chhabra, Muhao Chen, Prasant Mohapatra
Abstract: Large Language Models (LLMs) have achieved state-of-the-art performance at zero-shot generation of abstractive summaries for given articles. However, little is known about the robustness of such a process of zero-shot summarization. To bridge this gap, we propose relevance paraphrasing, a simple strategy that can be used to measure the robustness of LLMs as summarizers. The relevance paraphrasing approach identifies the most relevant sentences that contribute to generating an ideal summary, and then paraphrases these inputs to obtain a minimally perturbed dataset. Then, by evaluating model performance for summarization on both the original and perturbed datasets, we can assess the LLM's one aspect of robustness. We conduct extensive experiments with relevance paraphrasing on 4 diverse datasets, as well as 4 LLMs of different sizes (GPT-3.5-Turbo, Llama-2-13B, Mistral-7B, and Dolly-v2-7B). Our results indicate that LLMs are not consistent summarizers for the minimally perturbed articles, necessitating further improvements.
Authors: Hemant Yadav, Sunayana Sitaram, Rajiv Ratn Shah
Abstract: In recent years, self-supervised pre-training methods have gained significant traction in learning high-level information from raw speech. Among these methods, HuBERT has demonstrated SOTA performance in automatic speech recognition (ASR). However, HuBERT's performance lags behind data2vec due to disparities in pre-training strategies. In this paper, we propose (i) a Swap method to address pre-training and inference mismatch observed in HuBERT and (ii) incorporates Multicluster masked prediction loss for more effective utilization of the models capacity. The resulting method is, MS-HuBERT, an end-to-end self-supervised pre-training method for learning robust speech representations. It beats vanilla HuBERT on the ASR Librispeech benchmark on average by a 5% margin when evaluated on different finetuning splits. Additionally, we demonstrate that the learned embeddings obtained during pre-training encode essential information for improving performance of content based tasks such as ASR.
Authors: Xinyan Zhao, Yuan Sun, Wenlin Liu, Chau-Wai Wong
Abstract: This study is among the first to develop different prototypes of generative artificial intelligence (GenAI) chatbots powered by GPT-4 to communicate hurricane preparedness information to diverse residents. Drawing from the Computers Are Social Actors paradigm and the literature on disaster vulnerability and cultural tailoring, we conducted a between-subjects experiment with 441 Black, Hispanic, and Caucasian residents of Florida. Our results suggest that GenAI chatbots varying in tone formality and cultural tailoring significantly influence perceptions of their friendliness and credibility, which, in turn, relate to hurricane preparedness outcomes. These results highlight the potential of using GenAI chatbots to improve diverse communities' disaster preparedness.
Authors: Jessica Zhu, Iain Cruickshank, Michel Cukier
Abstract: News will be biased so long as people have opinions. As social media becomes the primary entry point for news and partisan differences increase, it is increasingly important for informed citizens to be able to recognize bias. If people are aware of the biases of the news they consume, they will be able to take action to avoid polarizing echo chambers. In this paper, we explore an often overlooked aspect of bias detection in media: the semantic structure of news articles. We present DocNet, a novel, inductive, and low-resource document embedding and political bias detection model. We also demonstrate that the semantic structure of news articles from opposing political sides, as represented in document-level graph embeddings, have significant similarities. DocNet bypasses the need for pre-trained language models, reducing resource dependency while achieving comparable performance. It can be used to advance political bias detection in low-resource environments. Our code and data are made available at: https://anonymous.4open.science/r/DocNet/
Authors: Lingrui Mei, Shenghua Liu, Yiwei Wang, Baolong Bi, Jiayi Mao, Xueqi Cheng
Abstract: "Jailbreak" is a major safety concern of Large Language Models (LLMs), which occurs when malicious prompts lead LLMs to produce harmful outputs, raising issues about the reliability and safety of LLMs. Therefore, an effective evaluation of jailbreaks is very crucial to develop its mitigation strategies. However, our research reveals that many jailbreaks identified by current evaluations may actually be hallucinations-erroneous outputs that are mistaken for genuine safety breaches. This finding suggests that some perceived vulnerabilities might not represent actual threats, indicating a need for more precise red teaming benchmarks. To address this problem, we propose the $\textbf{B}$enchmark for reli$\textbf{AB}$ilit$\textbf{Y}$ and jail$\textbf{B}$reak ha$\textbf{L}$l$\textbf{U}$cination $\textbf{E}$valuation (BabyBLUE). BabyBLUE introduces a specialized validation framework including various evaluators to enhance existing jailbreak benchmarks, ensuring outputs are useful malicious instructions. Additionally, BabyBLUE presents a new dataset as an augmentation to the existing red teaming benchmarks, specifically addressing hallucinations in jailbreaks, aiming to evaluate the true potential of jailbroken LLM outputs to cause harm to human society.
Authors: Han Jiang, Xiaoyuan Yi, Zhihua Wei, Ziang Xiao, Shu Wang, Xing Xie
Abstract: Warning: Contains harmful model outputs. Despite significant advancements, the propensity of Large Language Models (LLMs) to generate harmful and unethical content poses critical challenges. Measuring value alignment of LLMs becomes crucial for their regulation and responsible deployment. Although numerous benchmarks have been constructed to assess social bias, toxicity, and ethical issues in LLMs, those static benchmarks suffer from evaluation chronoeffect, in which, as models rapidly evolve, existing benchmarks may leak into training data or become saturated, overestimating ever-developing LLMs. To tackle this problem, we propose GETA, a novel generative evolving testing approach based on adaptive testing methods in measurement theory. Unlike traditional adaptive testing methods that rely on a static test item pool, GETA probes the underlying moral boundaries of LLMs by dynamically generating test items tailored to model capability. GETA co-evolves with LLMs by learning a joint distribution of item difficulty and model value conformity, thus effectively addressing evaluation chronoeffect. We evaluated various popular LLMs with GETA and demonstrated that 1) GETA can dynamically create difficulty-tailored test items and 2) GETA's evaluation results are more consistent with models' performance on unseen OOD and i.i.d. items, laying the groundwork for future evaluation paradigms.
Authors: Tzu-Heng Huang, Catherine Cao, Vaishnavi Bhargava, Frederic Sala
Abstract: Large pretrained models can be used as annotators, helping replace or augment crowdworkers and enabling distilling generalist models into smaller specialist models. Unfortunately, this comes at a cost: employing top-of-the-line models often requires paying thousands of dollars for API calls, while the resulting datasets are static and challenging to audit. To address these challenges, we propose a simple alternative: rather than directly querying labels from pretrained models, we task models to generate programs that can produce labels. These programs can be stored and applied locally, re-used and extended, and cost orders of magnitude less. Our system, Alchemist, obtains comparable to or better performance than large language model-based annotation in a range of tasks for a fraction of the cost: on average, improvements amount to a 12.9% enhancement while the total labeling costs across all datasets are reduced by a factor of approximately 500x.
Authors: Masahiro Suzuki, Hiroki Sakaji
Abstract: Natural language processing (NLP) tasks in English and general domains are widely available and are often used to evaluate pre-trained language models. In contrast, fewer tasks are available for languages other than English and in the financial domain. Particularly, tasks in the Japanese and financial domains are limited. We develop two large datasets using data published by a Japanese central government agency. The datasets provide three Japanese financial NLP tasks, including 3- and 12-class classifications for categorizing sentences, along with a 5-class classification task for sentiment analysis. Our datasets are designed to be comprehensive and updated by leveraging an automatic update framework that ensures that the latest task datasets are publicly always available.
Authors: Kaiser Sun, Mark Dredze
Abstract: The development of large language models leads to the formation of a pre-train-then-align paradigm, in which the model is typically pre-trained on a large text corpus and undergoes a tuning stage to align the model with human preference or downstream tasks. In this work, we investigate the relationship between pre-training and fine-tuning by fine-tuning multiple intermediate pre-trained model checkpoints. Our results on 18 datasets suggest that i) continual pre-training improves the model in a latent way that unveils after fine-tuning; ii) with extra fine-tuning, the datasets that the model does not demonstrate capability gain much more than those that the model performs well during the pre-training stage; iii) although model benefits significantly through supervised fine-tuning, it may forget previously known domain knowledge and the tasks that are not seen during fine-tuning; iv) the model resembles high sensitivity to evaluation prompts after supervised fine-tuning, but this sensitivity can be alleviated by more pre-training.
Authors: Chenwei Wan, Matthieu Labeau, Chlo\'e Clavel
Abstract: Designing emotionally intelligent conversational systems to provide comfort and advice to people experiencing distress is a compelling area of research. Recently, with advancements in large language models (LLMs), end-to-end dialogue agents without explicit strategy prediction steps have become prevalent. However, implicit strategy planning lacks transparency, and recent studies show that LLMs' inherent preference bias towards certain socio-emotional strategies hinders the delivery of high-quality emotional support. To address this challenge, we propose decoupling strategy prediction from language generation, and introduce a novel dialogue strategy prediction framework, EmoDynamiX, which models the discourse dynamics between user fine-grained emotions and system strategies using a heterogeneous graph for better performance and transparency. Experimental results on two ESC datasets show EmoDynamiX outperforms previous state-of-the-art methods with a significant margin (better proficiency and lower preference bias). Our approach also exhibits better transparency by allowing backtracing of decision making.
Authors: Artem Snegirev, Maria Tikhonova, Anna Maksimova, Alena Fenogenova, Alexander Abramov
Abstract: Embedding models play a crucial role in Natural Language Processing (NLP) by creating text embeddings used in various tasks such as information retrieval and assessing semantic text similarity. This paper focuses on research related to embedding models in the Russian language. It introduces a new Russian-focused embedding model called ru-en-RoSBERTa and the ruMTEB benchmark, the Russian version extending the Massive Text Embedding Benchmark (MTEB). Our benchmark includes seven categories of tasks, such as semantic textual similarity, text classification, reranking, and retrieval.The research also assesses a representative set of Russian and multilingual models on the proposed benchmark. The findings indicate that the new model achieves results that are on par with state-of-the-art models in Russian. We release the model ru-en-RoSBERTa, and the ruMTEB framework comes with open-source code, integration into the original framework and a public leaderboard.
Authors: Yunmeng Li, Jun Suzuki, Makoto Morishita, Kaori Abe, Kentaro Inui
Abstract: The complexities of chats pose significant challenges for machine translation models. Recognizing the need for a precise evaluation metric to address the issues of chat translation, this study introduces Multidimensional Quality Metrics for Chat Translation (MQM-Chat). Through the experiments of five models using MQM-Chat, we observed that all models generated certain fundamental errors, while each of them has different shortcomings, such as omission, overly correcting ambiguous source content, and buzzword issues, resulting in the loss of stylized information. Our findings underscore the effectiveness of MQM-Chat in evaluating chat translation, emphasizing the importance of stylized content and dialogue consistency for future studies.
Authors: Kohei Tsuji, Tatsuya Hiraoka, Yuchang Cheng, Tomoya Iwakura
Abstract: NLP datasets may still contain annotation errors, even when they are manually annotated. Researchers have attempted to develop methods to automatically reduce the adverse effect of errors in datasets. However, existing methods are time-consuming because they require many trained models to detect errors. This paper proposes a time-saving method that utilizes a tokenization technique called subword regularization to simulate multiple error detection models for detecting errors. Our proposed method, SubRegWeigh, can perform annotation weighting four to five times faster than the existing method. Additionally, SubRegWeigh improved performance in document classification and named entity recognition tasks. In experiments with pseudo-incorrect labels, SubRegWeigh clearly identifies pseudo-incorrect labels as annotation errors. Our code is available at https://github.com/4ldk/SubRegWeigh .
Authors: Huawei Ji, Cheng Deng, Bo Xue, Zhouyang Jin, Jiaxin Ding, Xiaoying Gan, Luoyi Fu, Xinbing Wang, Chenghu Zhou
Abstract: With the development of data-centric AI, the focus has shifted from model-driven approaches to improving data quality. Academic literature, as one of the crucial types, is predominantly stored in PDF formats and needs to be parsed into texts before further processing. However, parsing diverse structured texts in academic literature remains challenging due to the lack of datasets that cover various text structures. In this paper, we introduce AceParse, the first comprehensive dataset designed to support the parsing of a wide range of structured texts, including formulas, tables, lists, algorithms, and sentences with embedded mathematical expressions. Based on AceParse, we fine-tuned a multimodal model, named AceParser, which accurately parses various structured texts within academic literature. This model outperforms the previous state-of-the-art by 4.1% in terms of F1 score and by 5% in Jaccard Similarity, demonstrating the potential of multimodal models in academic literature parsing. Our dataset is available at https://github.com/JHW5981/AceParse.
Authors: Robert Morabito, Sangmitra Madhusudan, Tyler McDonald, Ali Emami
Abstract: Mitigating explicit and implicit biases in Large Language Models (LLMs) has become a critical focus in the field of natural language processing. However, many current methodologies evaluate scenarios in isolation, without considering the broader context or the spectrum of potential biases within each situation. To address this, we introduce the Sensitivity Testing on Offensive Progressions (STOP) dataset, which includes 450 offensive progressions containing 2,700 unique sentences of varying severity that progressively escalate from less to more explicitly offensive. Covering a broad spectrum of 9 demographics and 46 sub-demographics, STOP ensures inclusivity and comprehensive coverage. We evaluate several leading closed- and open-source models, including GPT-4, Mixtral, and Llama 3. Our findings reveal that even the best-performing models detect bias inconsistently, with success rates ranging from 19.3% to 69.8%. We also demonstrate how aligning models with human judgments on STOP can improve model answer rates on sensitive tasks such as BBQ, StereoSet, and CrowS-Pairs by up to 191%, while maintaining or even improving performance. STOP presents a novel framework for assessing the complex nature of biases in LLMs, which will enable more effective bias mitigation strategies and facilitates the creation of fairer language models.
Authors: Jinman Zhao (University of Toronto), Zifan Qian (University of Alberta), Linbo Cao (University of Waterloo), Yining Wang (University of Toronto), Yitian Ding (McGill University), Yulan Hu (Renmin University of China), Zeyu Zhang (The Australian National University), Zeyong Jin
Abstract: Role-play in large language models (LLMs) enhances their ability to generate contextually relevant and high-quality responses by simulating diverse cognitive perspectives. However, our study identifies significant risks associated with this technique. First, we demonstrate that autotuning, a method used to auto-select models' roles based on the question, can lead to the generation of harmful outputs, even when the model is tasked with adopting neutral roles. Second, we investigate how different roles affect the likelihood of generating biased or harmful content. Through testing on benchmarks containing stereotypical and harmful questions, we find that role-play consistently amplifies the risk of biased outputs. Our results underscore the need for careful consideration of both role simulation and tuning processes when deploying LLMs in sensitive or high-stakes contexts.
Authors: Zhiwei Liu, Xin Zhang, Kailai Yang, Qianqian Xie, Jimin Huang, Sophia Ananiadou
Abstract: The emergence of social media has made the spread of misinformation easier. In the financial domain, the accuracy of information is crucial for various aspects of financial market, which has made financial misinformation detection (FMD) an urgent problem that needs to be addressed. Large language models (LLMs) have demonstrated outstanding performance in various fields. However, current studies mostly rely on traditional methods and have not explored the application of LLMs in the field of FMD. The main reason is the lack of FMD instruction tuning datasets and evaluation benchmarks. In this paper, we propose FMDLlama, the first open-sourced instruction-following LLMs for FMD task based on fine-tuning Llama3.1 with instruction data, the first multi-task FMD instruction dataset (FMDID) to support LLM instruction tuning, and a comprehensive FMD evaluation benchmark (FMD-B) with classification and explanation generation tasks to test the FMD ability of LLMs. We compare our models with a variety of LLMs on FMD-B, where our model outperforms other open-sourced LLMs as well as OpenAI's products. This project is available at https://github.com/lzw108/FMD.
Authors: Mahmoud Abdelrahman, Edgardo Macatulad, Binyu Lei, Matias Quintana, Clayton Miller, Filip Biljecki
Abstract: The concept of digital twins has attracted significant attention across various domains, particularly within the built environment. However, there is a sheer volume of definitions and the terminological consensus remains out of reach. The lack of a universally accepted definition leads to ambiguities in their conceptualization and implementation, and may cause miscommunication for both researchers and practitioners. We employed Natural Language Processing (NLP) techniques to systematically extract and analyze definitions of digital twins from a corpus of more than 15,000 full-text articles spanning diverse disciplines. The study compares these findings with insights from an expert survey that included 52 experts. The study identifies concurrence on the components that comprise a ``Digital Twin'' from a practical perspective across various domains, contrasting them with those that do not, to identify deviations. We investigate the evolution of digital twin definitions over time and across different scales, including manufacturing, building, and urban/geospatial perspectives. We extracted the main components of Digital Twins using Text Frequency Analysis and N-gram analysis. Subsequently, we identified components that appeared in the literature and conducted a Chi-square test to assess the significance of each component in different domains. Our analysis identified key components of digital twins and revealed significant variations in definitions based on application domains, such as manufacturing, building, and urban contexts. The analysis of DT components reveal two major groups of DT types: High-Performance Real-Time (HPRT) DTs, and Long-Term Decision Support (LTDS) DTs. Contrary to common assumptions, we found that components such as simulation, AI/ML, real-time capabilities, and bi-directional data flow are not yet fully mature in the digital twins of the built environment.
Authors: Zhe Xu, Kaveh Hassani, Si Zhang, Hanqing Zeng, Michihiro Yasunaga, Limei Wang, Dongqi Fu, Ning Yao, Bo Long, Hanghang Tong
Abstract: Language Models (LMs) are increasingly challenging the dominance of domain-specific models, such as Graph Neural Networks (GNNs) and Graph Transformers (GTs), in graph learning tasks. Following this trend, we propose a novel approach that empowers off-the-shelf LMs to achieve performance comparable to state-of-the-art (SOTA) GNNs on node classification tasks, without requiring any architectural modification. By preserving the LM's original architecture, our approach retains a key benefit of LM instruction tuning: the ability to jointly train on diverse datasets, fostering greater flexibility and efficiency. To achieve this, we introduce two key augmentation strategies: (1) Enriching LMs' input using topological and semantic retrieval methods, which provide richer contextual information, and (2) guiding the LMs' classification process through a lightweight GNN classifier that effectively prunes class candidates. Our experiments on real-world datasets show that backbone Flan-T5 LMs equipped with these augmentation strategies outperform SOTA text-output node classifiers and are comparable to top-performing vector-output node classifiers. By bridging the gap between specialized node classifiers and general LMs, this work paves the way for more versatile and widely applicable graph learning models. We will open-source the code upon publication.
Authors: Ignacio Hounie, Charilaos Kanatsoulis, Arnuv Tandon, Alejandro Ribeiro
Abstract: Low Rank Adaptation (LoRA) is a popular Parameter Efficient Fine Tuning (PEFT) method that effectively adapts large pre-trained models for downstream tasks. LoRA parameterizes model updates using low-rank matrices at each layer, significantly reducing the number of trainable parameters and, consequently, resource requirements during fine-tuning. However, the lower bound on the number of trainable parameters remains high due to the use of the low-rank matrix model. Recent works have addressed this limitation by proposing low rank tensor parameterizations for model updates. However, they only exploit redundancy across layers, or tensorize individual matrices using ad-hoc schemes that introduce additional hyperparameters. In this work, we propose a higher-order Candecomp/Parafac (CP) decomposition, enabling a more compact and flexible representation compared to existing matrix and tensor based PEFT methods. Our experiments on Natural Language Understanding, Instruction Tuning, Preference Optimization and Protein Folding benchmarks demonstrate that our method can achieve a reduction in the number of parameters while maintaining comparable performance.
Authors: Aiwei Liu, Haoping Bai, Zhiyun Lu, Yanchao Sun, Xiang Kong, Simon Wang, Jiulong Shan, Albin Madappally Jose, Xiaojiang Liu, Lijie Wen, Philip S. Yu, Meng Cao
Abstract: Direct Preference Optimization (DPO) has been widely adopted for preference alignment of Large Language Models (LLMs) due to its simplicity and effectiveness. However, DPO is derived as a bandit problem in which the whole response is treated as a single arm, ignoring the importance differences between tokens, which may affect optimization efficiency and make it difficult to achieve optimal results. In this work, we propose that the optimal data for DPO has equal expected rewards for each token in winning and losing responses, as there is no difference in token importance. However, since the optimal dataset is unavailable in practice, we propose using the original dataset for importance sampling to achieve unbiased optimization. Accordingly, we propose a token-level importance sampling DPO objective named TIS-DPO that assigns importance weights to each token based on its reward. Inspired by previous works, we estimate the token importance weights using the difference in prediction probabilities from a pair of contrastive LLMs. We explore three methods to construct these contrastive LLMs: (1) guiding the original LLM with contrastive prompts, (2) training two separate LLMs using winning and losing responses, and (3) performing forward and reverse DPO training with winning and losing responses. Experiments show that TIS-DPO significantly outperforms various baseline methods on harmlessness and helpfulness alignment and summarization tasks. We also visualize the estimated weights, demonstrating their ability to identify key token positions.
Authors: Zhuoran Zhang, Yongxiang Li, Zijian Kan, Keyuan Cheng, Lijie Hu, Di Wang
Abstract: The locate-then-edit paradigm has shown significant promise for knowledge editing (KE) in Large Language Models (LLMs). While previous methods perform well on single-hop fact recall tasks, they consistently struggle with multi-hop factual recall tasks involving newly edited knowledge. In this paper, leveraging tools in mechanistic interpretability, we first identify that in multi-hop tasks, LLMs tend to retrieve knowledge with implicit subject information from deeper MLP layers, unlike single-hop tasks, which rely on shallow layers. This distinction explains the poor performance of current methods in multi-hop queries, as they primarily focus on editing shallow layers with single-hop edit prompts, leaving deeper layers unchanged. To address this, we propose IFMET, a novel locate-then-edit KE approach designed to edit both shallow and deep MLP layers. Beyond single-hop editing prompts, IFMET further incorporates multi-hop editing prompts to locate and modify knowledge across different stages of reasoning. Experimental results demonstrate that IFMET significantly improves performance on multi-hop factual recall tasks, overcoming the limitations of previous locate-then-edit methods
Authors: Philipp Guldimann, Alexander Spiridonov, Robin Staab, Nikola Jovanovi\'c, Mark Vero, Velko Vechev, Anna-Maria Gueorguieva, Mislav Balunovi\'c, Nikola Konstantinov, Pavol Bielik, Petar Tsankov, Martin Vechev
Abstract: The EU's Artificial Intelligence Act (AI Act) is a significant step towards responsible AI development, but lacks clear technical interpretation, making it difficult to assess models' compliance. This work presents COMPL-AI, a comprehensive framework consisting of (i) the first technical interpretation of the EU AI Act, translating its broad regulatory requirements into measurable technical requirements, with the focus on large language models (LLMs), and (ii) an open-source Act-centered benchmarking suite, based on thorough surveying and implementation of state-of-the-art LLM benchmarks. By evaluating 12 prominent LLMs in the context of COMPL-AI, we reveal shortcomings in existing models and benchmarks, particularly in areas like robustness, safety, diversity, and fairness. This work highlights the need for a shift in focus towards these aspects, encouraging balanced development of LLMs and more comprehensive regulation-aligned benchmarks. Simultaneously, COMPL-AI for the first time demonstrates the possibilities and difficulties of bringing the Act's obligations to a more concrete, technical level. As such, our work can serve as a useful first step towards having actionable recommendations for model providers, and contributes to ongoing efforts of the EU to enable application of the Act, such as the drafting of the GPAI Code of Practice.
Authors: Jiamu Zheng, Jinghuai Zhang, Tianyu Du, Xuhong Zhang, Jianwei Yin, Tao Lin
Abstract: Collaborative learning of large language models (LLMs) has emerged as a new paradigm for utilizing private data from different parties to guarantee efficiency and privacy. Meanwhile, Knowledge Editing (KE) for LLMs has also garnered increased attention due to its ability to manipulate the behaviors of LLMs explicitly, yet leaves the collaborative KE case (in which knowledge edits of multiple parties are aggregated in a privacy-preserving and continual manner) unexamined. To this end, this manuscript dives into the first investigation of collaborative KE, in which we start by carefully identifying the unique three challenges therein, including knowledge overlap, knowledge conflict, and knowledge forgetting. We then propose a non-destructive collaborative KE framework, COLLABEDIT, which employs a novel model merging mechanism to mimic the global KE behavior while preventing the severe performance drop. Extensive experiments on two canonical datasets demonstrate the superiority of COLLABEDIT compared to other destructive baselines, and results shed light on addressing three collaborative KE challenges and future applications. Our code is available at https://github.com/LINs-lab/CollabEdit.
Authors: Daniele Gambetta, Gizem Gezici, Fosca Giannotti, Dino Pedreschi, Alistair Knott, Luca Pappalardo
Abstract: As synthetic content increasingly infiltrates the web, generative AI models may experience an autophagy process, where they are fine-tuned using their own outputs. This autophagy could lead to a phenomenon known as model collapse, which entails a degradation in the performance and diversity of generative AI models over successive generations. Recent studies have explored the emergence of model collapse across various generative AI models and types of data. However, the current characterizations of model collapse tend to be simplistic and lack comprehensive evaluation. In this article, we conduct a thorough investigation of model collapse across three text datasets, utilizing semantic networks to analyze text repetitiveness and diversity, while employing next-token probabilities to quantify the loss of diversity. We also examine how the proportions of synthetic tokens affect the severity of model collapse and perform cross-dataset evaluations to identify domain-specific variations. By proposing metrics and strategies for a more detailed assessment of model collapse, our study provides new insights for the development of robust generative AI systems.
Authors: Zhengyu Hu, Jieyu Zhang, Zhihan Xiong, Alexander Ratner, Hui Xiong, Ranjay Krishna
Abstract: Despite the remarkable success of Large Language Models (LLMs), evaluating their outputs' quality regarding *preference* remains a critical challenge. Existing works usually leverage an LLM as the judge for comparing LLMs' output pairwisely, yet such model-based evaluator is *weak evaluator* due to *conflicting preference*, i.e., output A is better than B, B than C, but C than A, causing contradictory evaluation results. To address this, we introduce GED (Preference Graph Ensemble and Denoise), a novel approach that leverages multiple model-based evaluators to construct preference graphs, and then ensemble and denoise these graphs for better, non-contradictory evaluation results. In particular, our method consists of two primary stages: aggregating evaluations into a unified graph and applying a denoising process to eliminate cyclic inconsistencies, ensuring a directed acyclic graph (DAG) structure. We provide theoretical guarantees for our framework, demonstrating its efficacy in recovering the ground truth preference structure. Extensive experiments on ten benchmarks demonstrate GED's superiority in three applications: model ranking, response selection, and model alignment tasks. Notably, GED combines small LLM evaluators (e.g., Llama3-8B, Mistral-7B, Qwen2-7B) to outperform stronger ones (e.g., Qwen2-72B), showcasing its effectiveness in enhancing evaluation reliability and improving model performance.
Authors: Shwai He, Tao Ge, Guoheng Sun, Bowei Tian, Xiaoyang Wang, Ang Li, Dong Yu
Abstract: Traditional transformer models often allocate a fixed amount of computational resources to every input token, leading to inefficient and unnecessary computation. To address this, the Mixture of Depths (MoD) was introduced to dynamically adjust the computational depth by skipping less important layers. Despite its promise, current MoD approaches remain under-explored and face two main challenges: (1) \textit{high training costs due to the need to train the entire model along with the routers that determine which layers to skip}, and (2) \textit{the risk of performance degradation when important layers are bypassed}. In response to the first issue, we propose Router-Tuning, a method that fine-tunes only the router on a small dataset, drastically reducing the computational overhead associated with full model training. For the second challenge, we propose MindSkip, which deploys \textit{Attention with Dynamic Depths}. This method preserves the model's performance while significantly enhancing computational and memory efficiency. Extensive experiments demonstrate that our approach delivers competitive results while dramatically improving the computation efficiency, e.g., 21\% speedup and only a 0.2\% performance drop. The code is released at \url{https://github.com/CASE-Lab-UMD/Router-Tuning}.
Authors: Zefang Liu, Yinzhu Quan
Abstract: Retrieving temporal event sequences from textual descriptions is crucial for applications such as analyzing e-commerce behavior, monitoring social media activities, and tracking criminal incidents. To advance this task, we introduce TESRBench, a comprehensive benchmark for temporal event sequence retrieval (TESR) from textual descriptions. TESRBench includes diverse real-world datasets with synthesized and reviewed textual descriptions, providing a strong foundation for evaluating retrieval performance and addressing challenges in this domain. Building on this benchmark, we propose TPP-Embedding, a novel model for embedding and retrieving event sequences. The model leverages the TPP-LLM framework, integrating large language models (LLMs) with temporal point processes (TPPs) to encode both event texts and times. By pooling representations and applying a contrastive loss, it unifies temporal dynamics and event semantics in a shared embedding space, aligning sequence-level embeddings of event sequences and their descriptions. TPP-Embedding demonstrates superior performance over baseline models across TESRBench datasets, establishing it as a powerful solution for the temporal event sequence retrieval task.
Authors: Rui Yang, Boming Yang, Aosong Feng, Sixun Ouyang, Moritz Blum, Tianwei She, Yuang Jiang, Freddy Lecue, Jinghui Lu, Irene Li
Abstract: Knowledge Graphs (KGs) are crucial in the field of artificial intelligence and are widely used in downstream tasks, such as question-answering (QA). The construction of KGs typically requires significant effort from domain experts. Large Language Models (LLMs) have recently been used for Knowledge Graph Construction (KGC). However, most existing approaches focus on a local perspective, extracting knowledge triplets from individual sentences or documents, missing a fusion process to combine the knowledge in a global KG. This work introduces Graphusion, a zero-shot KGC framework from free text. It contains three steps: in Step 1, we extract a list of seed entities using topic modeling to guide the final KG includes the most relevant entities; in Step 2, we conduct candidate triplet extraction using LLMs; in Step 3, we design the novel fusion module that provides a global view of the extracted knowledge, incorporating entity merging, conflict resolution, and novel triplet discovery. Results show that Graphusion achieves scores of 2.92 and 2.37 out of 3 for entity extraction and relation recognition, respectively. Moreover, we showcase how Graphusion could be applied to the Natural Language Processing (NLP) domain and validate it in an educational scenario. Specifically, we introduce TutorQA, a new expert-verified benchmark for QA, comprising six tasks and a total of 1,200 QA pairs. Using the Graphusion-constructed KG, we achieve a significant improvement on the benchmark, for example, a 9.2% accuracy improvement on sub-graph completion.
Authors: Mufei Li, Siqi Miao, Pan Li
Abstract: Large Language Models (LLMs) demonstrate strong reasoning abilities but face limitations such as hallucinations and outdated knowledge. Knowledge Graph (KG)-based Retrieval-Augmented Generation (RAG) addresses these issues by grounding LLM outputs in structured external knowledge from KGs. However, current KG-based RAG frameworks still struggle to optimize the trade-off between retrieval effectiveness and efficiency in identifying a suitable amount of relevant graph information for the LLM to digest. We introduce SubgraphRAG, extending the KG-based RAG framework that retrieves subgraphs and leverages LLMs for reasoning and answer prediction. Our approach innovatively integrates a lightweight multilayer perceptron with a parallel triple-scoring mechanism for efficient and flexible subgraph retrieval while encoding directional structural distances to enhance retrieval effectiveness. The size of retrieved subgraphs can be flexibly adjusted to match the query's need and the downstream LLM's capabilities. This design strikes a balance between model complexity and reasoning power, enabling scalable and generalizable retrieval processes. Notably, based on our retrieved subgraphs, smaller LLMs like Llama3.1-8B-Instruct deliver competitive results with explainable reasoning, while larger models like GPT-4o achieve state-of-the-art accuracy compared with previous baselines -- all without fine-tuning. Extensive evaluations on the WebQSP and CWQ benchmarks highlight SubgraphRAG's strengths in efficiency, accuracy, and reliability by reducing hallucinations and improving response grounding.
Authors: Haimanti Bhattacharya, Subhasish Dugar, Sanchaita Hazra, Bodhisattwa Prasad Majumder
Abstract: We investigate how low-quality AI advisors, lacking quality disclosures, can help spread text-based lies while seeming to help people detect lies. Participants in our experiment discern truth from lies by evaluating transcripts from a game show that mimicked deceptive social media exchanges on topics with objective truths. We find that when relying on low-quality advisors without disclosures, participants' truth-detection rates fall below their own abilities, which recovered once the AI's true effectiveness was revealed. Conversely, high-quality advisor enhances truth detection, regardless of disclosure. We discover that participants' expectations about AI capabilities contribute to their undue reliance on opaque, low-quality advisors.
Authors: Haritz Puerto, Martin Gubri, Sangdoo Yun, Seong Joon Oh
Abstract: Membership inference attacks (MIA) attempt to verify the membership of a given data sample in the training set for a model. MIA has become relevant in recent years, following the rapid development of large language models (LLM). Many are concerned about the usage of copyrighted materials for training them and call for methods for detecting such usage. However, recent research has largely concluded that current MIA methods do not work on LLMs. Even when they seem to work, it is usually because of the ill-designed experimental setup where other shortcut features enable "cheating." In this work, we argue that MIA still works on LLMs, but only when multiple documents are presented for testing. We construct new benchmarks that measure the MIA performances at a continuous scale of data samples, from sentences (n-grams) to a collection of documents (multiple chunks of tokens). To validate the efficacy of current MIA approaches at greater scales, we adapt a recent work on Dataset Inference (DI) for the task of binary membership detection that aggregates paragraph-level MIA features to enable MIA at document and collection of documents level. This baseline achieves the first successful MIA on pre-trained and fine-tuned LLMs.
Authors: Guangxuan Xu, Kai Xu, Shivchander Sudalairaj, Hao Wang, Akash Srivastava
Abstract: Preference tuning relies on high-quality human preference data, which is often expensive and time-consuming to gather. In this paper, we introduce Dr.SoW (Density Ratio of Strong over Weak) a cost-effective method that eliminates the reliance for human annotation by leveraging off-the-shelf LLMs for preference data annotation. Dr.SoW uses the log-density ratio between a better-aligned and a less-aligned LLM as a reward signal. We evaluate Dr.SoW across 221 different LLM pairs and empirically find a strong correlation between the performance gap of the paired models and the quality of the reward signal. This insight provides a practical guideline for selecting LLMs for data annotation. Additionally, we introduce an end-to-end pipeline that customizes reward functions based on user query domains. Without fine-tuning, it improves accuracy on domain-specific evaluations. With a pair of Mistral-7B models, Dr.SoW achieves a RewardBench score of 82.6, outperforming the best trained reward functions from same model class and demonstrating competitive performance against SoTA models in Safety (91.0) and Reasoning (88.0) domains. Further, we preference-tune Llama-3-8B-Instruct using data annotated by Dr.SoW. Our approach pushes Llama-3-8B to achieve a 37.4 % (+15.1 %) win rate on ArenaHard and a 40.7 % (+17.8 %) win rate on length-controlled AlpacaEval 2.0.
Authors: Yihan Wang, Andrew Bai, Nanyun Peng, Cho-Jui Hsieh
Abstract: Pre-trained Large Language Models (LLMs) require post-training methods such as supervised fine-tuning (SFT) on instruction-response pairs to enable instruction following. However, this process can potentially harm existing capabilities learned during pre-training. In this paper, we investigate the loss of context awareness after SFT, where context awareness is defined as the ability to extract and understand information from user-provided context and respond accordingly. We identify and demonstrate that the loss of context awareness, particularly in open-source models, occurs in instruction fine-tuned LLMs when the chat template is applied to input prompts. We identify that the performance decline is associated with a bias toward different roles learned during conversational instruction fine-tuning. We demonstrate this correlation by visualizing changes in attention allocation after the chat template is applied and manually steering the attention heads. The bias can be learned from training examples that align with the model's internal knowledge and rely less on the user-provided context to generate correct responses. Based on these observations, we propose a metric to identify context-dependent examples from general instruction fine-tuning datasets. We then apply conditional instruction fine-tuning with a context-dependency indicator, enabling the model to preserve context awareness after SFT. Empirical experiments on four context-dependent downstream tasks and three pre-trained LLMs of different sizes show that our method effectively mitigates the loss of context awareness without compromising general instruction-following capabilities.
Authors: Zhi Zhang, Yan Liu, Sheng-hua Zhong, Gong Chen, Yu Yang, Jiannong Cao
Abstract: Literature reviews play a crucial role in scientific research for understanding the current state of research, identifying gaps, and guiding future studies on specific topics. However, the process of conducting a comprehensive literature review is yet time-consuming. This paper proposes a novel framework, collaborative knowledge minigraph agents (CKMAs), to automate scholarly literature reviews. A novel prompt-based algorithm, the knowledge minigraph construction agent (KMCA), is designed to identify relations between concepts from academic literature and automatically constructs knowledge minigraphs. By leveraging the capabilities of large language models on constructed knowledge minigraphs, the multiple path summarization agent (MPSA) efficiently organizes concepts and relations from different viewpoints to generate literature review paragraphs. We evaluate CKMAs on three benchmark datasets. Experimental results show the effectiveness of the proposed method, further revealing promising applications of LLMs in scientific research.
Authors: Jinming Xing, Dongwen Luo, Chang Xue, Ruilin Xing
Abstract: Large Language Models (LLMs) have revolutionized natural language processing (NLP) by delivering state-of-the-art performance across a variety of tasks. Among these, Transformer-based models like BERT and GPT rely on pooling layers to aggregate token-level embeddings into sentence-level representations. Common pooling mechanisms such as Mean, Max, and Weighted Sum play a pivotal role in this aggregation process. Despite their widespread use, the comparative performance of these strategies on different LLM architectures remains underexplored. To address this gap, this paper investigates the effects of these pooling mechanisms on two prominent LLM families -- BERT and GPT, in the context of sentence-level sentiment analysis. Comprehensive experiments reveal that each pooling mechanism exhibits unique strengths and weaknesses depending on the task's specific requirements. Our findings underline the importance of selecting pooling methods tailored to the demands of particular applications, prompting a re-evaluation of common assumptions regarding pooling operations. By offering actionable insights, this study contributes to the optimization of LLM-based models for downstream tasks.
Authors: Jiawei Gu, Xuhui Jiang, Zhichao Shi, Hexiang Tan, Xuehao Zhai, Chengjin Xu, Wei Li, Yinghan Shen, Shengjie Ma, Honghao Liu, Saizhuo Wang, Kun Zhang, Yuanzhuo Wang, Wen Gao, Lionel Ni, Jian Guo
Abstract: Accurate and consistent evaluation is crucial for decision-making across numerous fields, yet it remains a challenging task due to inherent subjectivity, variability, and scale. Large Language Models (LLMs) have achieved remarkable success across diverse domains, leading to the emergence of "LLM-as-a-Judge," where LLMs are employed as evaluators for complex tasks. With their ability to process diverse data types and provide scalable, cost-effective, and consistent assessments, LLMs present a compelling alternative to traditional expert-driven evaluations. However, ensuring the reliability of LLM-as-a-Judge systems remains a significant challenge that requires careful design and standardization. This paper provides a comprehensive survey of LLM-as-a-Judge, addressing the core question: How can reliable LLM-as-a-Judge systems be built? We explore strategies to enhance reliability, including improving consistency, mitigating biases, and adapting to diverse assessment scenarios. Additionally, we propose methodologies for evaluating the reliability of LLM-as-a-Judge systems, supported by a novel benchmark designed for this purpose. To advance the development and real-world deployment of LLM-as-a-Judge systems, we also discussed practical applications, challenges, and future directions. This survey serves as a foundational reference for researchers and practitioners in this rapidly evolving field.
Authors: Jamshid Mozafari, Florian Gerhold, Adam Jatowt
Abstract: The use of Large Language Models (LLMs) has increased significantly with users frequently asking questions to chatbots. In the time when information is readily accessible, it is crucial to stimulate and preserve human cognitive abilities and maintain strong reasoning skills. This paper addresses such challenges by promoting the use of hints as an alternative or a supplement to direct answers. We first introduce a manually constructed hint dataset, WikiHint, which is based on Wikipedia and includes 5,000 hints created for 1,000 questions. We then finetune open-source LLMs such as LLaMA-3.1 for hint generation in answer-aware and answeragnostic contexts. We assess the effectiveness of the hints with human participants who answer questions with and without the aid of hints. Additionally, we introduce a lightweight evaluation method, HintRank, to evaluate and rank hints in both answeraware and answer-agnostic settings. Our findings show that (a) the dataset helps generate more effective hints, (b) including answer information along with questions generally improves quality of generated hints, and (c) encoder-based models perform better than decoder-based models in hint ranking.
Authors: Yerram Varun, Rahul Madhavan, Sravanti Addepalli, Arun Suggala, Karthikeyan Shanmugam, Prateek Jain
Abstract: Large Language Models (LLMs) are typically trained to predict in the forward direction of time. However, recent works have shown that prompting these models to look back and critique their own generations can produce useful feedback. Motivated by this, we explore the question of whether LLMs can be empowered to think (predict and score) backwards to provide unsupervised feedback that complements forward LLMs. Towards this, we introduce Time Reversed Language Models (TRLMs), which can score and generate queries when conditioned on responses, effectively functioning in the reverse direction of time. Further, to effectively infer in the response to query direction, we pre-train and fine-tune a language model (TRLM-Ba) in the reverse token order from scratch. We show empirically (and theoretically in a stylized setting) that time-reversed models can indeed complement forward model predictions when used to score the query given response for re-ranking multiple forward generations. We obtain up to 5\% improvement on the widely used AlpacaEval Leaderboard over the competent baseline of best-of-N re-ranking using self log-perplexity scores. We further show that TRLM scoring outperforms conventional forward scoring of response given query, resulting in significant gains in applications such as citation generation and passage retrieval. We next leverage the generative ability of TRLM to augment or provide unsupervised feedback to input safety filters of LLMs, demonstrating a drastic reduction in false negative rate with negligible impact on false positive rates against several attacks published on the popular JailbreakBench leaderboard.
Authors: David Samuel, Vladislav Mikhailov, Erik Velldal, Lilja {\O}vrelid, Lucas Georges Gabriel Charpentier, Andrey Kutuzov, Stephan Oepen
Abstract: Training large language models requires vast amounts of data, posing a challenge for less widely spoken languages like Norwegian and even more so for truly low-resource languages like Northern S\'ami. To address this issue, we present a novel three-stage continual training approach that substantially improves the downstream performance together with the inference efficiency for the target languages. Based on our findings, we train, evaluate, and openly release a new generative language model for Norwegian Bokm\r{a}l, Nynorsk, and Northern S\'ami with 11.4 billion parameters: NorMistral-11B.
Authors: Amirhossein Abaskohi, Spandana Gella, Giuseppe Carenini, Issam H. Laradji
Abstract: Multimodal multihop question answering is a complex task that requires reasoning over multiple sources of information, such as images and text, to answer questions. While there has been significant progress in visual question answering, the multihop setting remains unexplored due to the lack of high-quality datasets. Current methods focus on single-hop question answering or a single modality, which makes them unsuitable for real-world scenarios such as analyzing multimodal educational materials, summarizing lengthy academic articles, or interpreting scientific studies that combine charts, images, and text. To address this gap, we propose a novel methodology, introducing the first framework for creating a high-quality dataset that enables training models for multimodal multihop question answering. Our approach consists of a 5-stage pipeline that involves acquiring relevant multimodal documents from Wikipedia, synthetically generating high-level questions and answers, and validating them through rigorous criteria to ensure quality data. We evaluate our methodology by training models on our synthesized dataset and testing on two benchmarks, our results demonstrate that, with an equal sample size, models trained on our synthesized data outperform those trained on human-collected data by 1.9 in exact match (EM) on average. We believe our data synthesis method will serve as a strong foundation for training and evaluating multimodal multihop question answering models.
Authors: Jiya Manchanda, Laura Boettcher, Matheus Westphalen, Jasser Jasser
Abstract: Large language models (LLMs) have rapidly advanced natural language processing, driving significant breakthroughs in tasks such as text generation, machine translation, and domain-specific reasoning. The field now faces a critical dilemma in its approach: closed-source models like GPT-4 deliver state-of-the-art performance but restrict reproducibility, accessibility, and external oversight, while open-source frameworks like LLaMA and Mixtral democratize access, foster collaboration, and support diverse applications, achieving competitive results through techniques like instruction tuning and LoRA. Hybrid approaches address challenges like bias mitigation and resource accessibility by combining the scalability of closed-source systems with the transparency and inclusivity of open-source framework. However, in this position paper, we argue that open-source remains the most robust path for advancing LLM research and ethical deployment.
Authors: Xingyu Chen, Jiahao Xu, Tian Liang, Zhiwei He, Jianhui Pang, Dian Yu, Linfeng Song, Qiuzhi Liu, Mengfei Zhou, Zhuosheng Zhang, Rui Wang, Zhaopeng Tu, Haitao Mi, Dong Yu
Abstract: The remarkable performance of models like the OpenAI o1 can be attributed to their ability to emulate human-like long-time thinking during inference. These models employ extended chain-of-thought (CoT) processes, exploring multiple strategies to enhance problem-solving capabilities. However, a critical question remains: How to intelligently and efficiently scale computational resources during testing. This paper presents the first comprehensive study on the prevalent issue of overthinking in these models, where excessive computational resources are allocated for simple problems with minimal benefit. We introduce novel efficiency metrics from both outcome and process perspectives to evaluate the rational use of computational resources by o1-like models. Using a self-training paradigm, we propose strategies to mitigate overthinking, streamlining reasoning processes without compromising accuracy. Experimental results show that our approach successfully reduces computational overhead while preserving model performance across a range of testsets with varying difficulty levels, such as GSM8K, MATH500, GPQA, and AIME.
Authors: Antonin Poch\'e (IRIT), Alon Jacovi (CERCO UMR5549, ANITI), Agustin Martin Picard (CERCO UMR5549, ANITI), Victor Boutin (CERCO UMR5549, ANITI), Fanny Jourdan
Abstract: Concept-based explanations work by mapping complex model computations to human-understandable concepts. Evaluating such explanations is very difficult, as it includes not only the quality of the induced space of possible concepts but also how effectively the chosen concepts are communicated to users. Existing evaluation metrics often focus solely on the former, neglecting the latter. We introduce an evaluation framework for measuring concept explanations via automated simulatability: a simulator's ability to predict the explained model's outputs based on the provided explanations. This approach accounts for both the concept space and its interpretation in an end-to-end evaluation. Human studies for simulatability are notoriously difficult to enact, particularly at the scale of a wide, comprehensive empirical evaluation (which is the subject of this work). We propose using large language models (LLMs) as simulators to approximate the evaluation and report various analyses to make such approximations reliable. Our method allows for scalable and consistent evaluation across various models and datasets. We report a comprehensive empirical evaluation using this framework and show that LLMs provide consistent rankings of explanation methods. Code available at https://github.com/AnonymousConSim/ConSim.
Authors: Zenon Lamprou, Frank Polick, Yashar Moshfeghi
Abstract: This research examines the congruence between neural activity and advanced transformer models, emphasizing the semantic significance of punctuation in text understanding. Utilizing an innovative approach originally proposed by Toneva and Wehbe, we evaluate four advanced transformer models RoBERTa, DistiliBERT, ALBERT, and ELECTRA against neural activity data. Our findings indicate that RoBERTa exhibits the closest alignment with neural activity, surpassing BERT in accuracy. Furthermore, we investigate the impact of punctuation removal on model performance and neural alignment, revealing that BERT's accuracy enhances in the absence of punctuation. This study contributes to the comprehension of how neural networks represent language and the influence of punctuation on semantic processing within the human brain.
Authors: Kun-Peng Ning, Jia-Yu Yao, Yu-Yang Liu, Mu-Nan Ning, Li Yuan
Abstract: Large Language Models (LLMs), such as GPT, are considered to learn the latent distributions within large-scale web-crawl datasets and accomplish natural language processing (NLP) tasks by predicting the next token. However, this mechanism of latent distribution modeling lacks quantitative understanding and analysis. In this paper, we propose a novel perspective that any language dataset can be represented by a Monte Carlo Language Tree (abbreviated as ``Data-Tree''), where each node denotes a token, each edge denotes a token transition probability, and each sequence has a unique path. Any GPT-like language model can also be flattened into another Monte Carlo Language Tree (abbreviated as ``GPT-Tree''). Our experiments show that different GPT models trained on the same dataset exhibit significant structural similarity in GPT-Tree visualization, and larger models converge more closely to the Data-Tree. More than 87\% GPT output tokens can be recalled by Data-Tree. These findings may confirm that the reasoning process of LLMs is more likely to be probabilistic pattern-matching rather than formal reasoning, as each model inference seems to find a context pattern with maximum probability from the Data-Tree. Furthermore, we provide deeper insights into issues such as hallucination, Chain-of-Thought (CoT) reasoning, and token bias in LLMs.
Authors: Nikos Kanakaris, Heng Ping, Xiongye Xiao, Nesreen K. Ahmed, Luca Luceri, Emilio Ferrara, Paul Bogdan
Abstract: Detecting organized political campaigns is of paramount importance in fighting against disinformation on social media. Existing approaches for the identification of such organized actions employ techniques mostly from network science, graph machine learning and natural language processing. Their ultimate goal is to analyze the relationships and interactions (e.g. re-posting) among users and the textual similarities of their posts. Despite their effectiveness in recognizing astroturf campaigns, these methods face significant challenges, notably the class imbalance in available training datasets. To mitigate this issue, recent methods usually resort to data augmentation or increasing the number of positive samples, which may not always be feasible or sufficient in real-world settings. Following a different path, in this paper, we propose a novel framework for identifying astroturf campaigns based solely on large language models (LLMs), introducing a Balanced Retrieval-Augmented Generation (Balanced RAG) component. Our approach first gives both textual information concerning the posts (in our case tweets) and the user interactions of the social network as input to a language model. Then, through prompt engineering and the proposed Balanced RAG method, it effectively detects coordinated disinformation campaigns on X (Twitter). The proposed framework does not require any training or fine-tuning of the language model. Instead, by strategically harnessing the strengths of prompt engineering and Balanced RAG, it facilitates LLMs to overcome the effects of class imbalance and effectively identify coordinated political campaigns. The experimental results demonstrate that by incorporating the proposed prompt engineering and Balanced RAG methods, our framework outperforms the traditional graph-based baselines, achieving 2x-3x improvements in terms of precision, recall and F1 scores.
Authors: Anna Kruspe
Abstract: Large Language Models (LLMs) reflect the biases in their training data and, by extension, those of the people who created this training data. Detecting, analyzing, and mitigating such biases is becoming a focus of research. One type of bias that has been understudied so far are geocultural biases. Those can be caused by an imbalance in the representation of different geographic regions and cultures in the training data, but also by value judgments contained therein. In this paper, we make a first step towards analyzing musical biases in LLMs, particularly ChatGPT and Mixtral. We conduct two experiments. In the first, we prompt LLMs to provide lists of the "Top 100" musical contributors of various categories and analyze their countries of origin. In the second experiment, we ask the LLMs to numerically rate various aspects of the musical cultures of different countries. Our results indicate a strong preference of the LLMs for Western music cultures in both experiments.
Authors: Erica Coppolillo, Giuseppe Manco, Luca Maria Aiello
Abstract: Detecting biases in the outputs produced by generative models is essential to reduce the potential risks associated with their application in critical settings. However, the majority of existing methodologies for identifying biases in generated text consider the models in isolation and neglect their contextual applications. Specifically, the biases that may arise in multi-agent systems involving generative models remain under-researched. To address this gap, we present a framework designed to quantify biases within multi-agent systems of conversational Large Language Models (LLMs). Our approach involves simulating small echo chambers, where pairs of LLMs, initialized with aligned perspectives on a polarizing topic, engage in discussions. Contrary to expectations, we observe significant shifts in the stance expressed in the generated messages, particularly within echo chambers where all agents initially express conservative viewpoints, in line with the well-documented political bias of many LLMs toward liberal positions. Crucially, the bias observed in the echo-chamber experiment remains undetected by current state-of-the-art bias detection methods that rely on questionnaires. This highlights a critical need for the development of a more sophisticated toolkit for bias detection and mitigation for AI multi-agent systems. The code to perform the experiments is publicly available at https://anonymous.4open.science/r/LLMsConversationalBias-7725.
URLs: https://anonymous.4open.science/r/LLMsConversationalBias-7725.
Authors: Hakaze Cho, Naoya Inoue
Abstract: Classification tasks are widely investigated in the In-Context Learning (ICL) paradigm. However, current efforts are evaluated on disjoint benchmarks and settings, while their performances are significantly influenced by some trivial variables, such as prompt templates, data sampling, instructions, etc., which leads to significant inconsistencies in the results reported across various literature, preventing fair comparison or meta-analysis across different papers. Therefore, this paper proposes a standardized and easy-to-use evaluation toolkit (StaICC) for in-context classification. Including, for the normal classification task, we provide StaICC-Normal, selecting 10 widely used datasets, and generating prompts with a fixed form, to mitigate the variance among the experiment implementations. To enrich the usage of our benchmark, we also provide a sub-benchmark StaICC-Diag for diagnosing ICL from several aspects, aiming for a more robust inference processing.
Authors: Beiming Liu, Zhizhuo Cui, Siteng Hu, Xiaohua Li, Haifeng Lin, Zhengxin Zhang
Abstract: Aerospace manufacturing demands exceptionally high precision in technical parameters. The remarkable performance of Large Language Models (LLMs), such as GPT-4 and QWen, in Natural Language Processing has sparked industry interest in their application to tasks including process design, material selection, and tool information retrieval. However, LLMs are prone to generating "hallucinations" in specialized domains, producing inaccurate or false information that poses significant risks to the quality of aerospace products and flight safety. This paper introduces a set of evaluation metrics tailored for LLMs in aerospace manufacturing, aiming to assess their accuracy by analyzing their performance in answering questions grounded in professional knowledge. Firstly, key information is extracted through in-depth textual analysis of classic aerospace manufacturing textbooks and guidelines. Subsequently, utilizing LLM generation techniques, we meticulously construct multiple-choice questions with multiple correct answers of varying difficulty. Following this, different LLM models are employed to answer these questions, and their accuracy is recorded. Experimental results demonstrate that the capabilities of LLMs in aerospace professional knowledge are in urgent need of improvement. This study provides a theoretical foundation and practical guidance for the application of LLMs in aerospace manufacturing, addressing a critical gap in the field.
Authors: Wonjun Lee, Solee Im, Heejin Do, Yunsu Kim, Jungseul Ok, Gary Geunbae Lee
Abstract: Dysarthric speech recognition often suffers from performance degradation due to the intrinsic diversity of dysarthric severity and extrinsic disparity from normal speech. To bridge these gaps, we propose a Dynamic Phoneme-level Contrastive Learning (DyPCL) method, which leads to obtaining invariant representations across diverse speakers. We decompose the speech utterance into phoneme segments for phoneme-level contrastive learning, leveraging dynamic connectionist temporal classification alignment. Unlike prior studies focusing on utterance-level embeddings, our granular learning allows discrimination of subtle parts of speech. In addition, we introduce dynamic curriculum learning, which progressively transitions from easy negative samples to difficult-to-distinguishable negative samples based on phonetic similarity of phoneme. Our approach to training by difficulty levels alleviates the inherent variability of speakers, better identifying challenging speeches. Evaluated on the UASpeech dataset, DyPCL outperforms baseline models, achieving an average 22.10\% relative reduction in word error rate (WER) across the overall dysarthria group.
Authors: Dang Huu-Tien, Hoang Thanh-Tung, Le-Minh Nguyen, Naoya Inoue
Abstract: Representation Misdirection (RM) and variants are established large language model (LLM) unlearning methods with state-of-the-art performance. In this paper, we show that RM methods inherently reduce models' robustness, causing them to misbehave even when a single non-adversarial forget-token is in the retain-query. Toward understanding underlying causes, we reframe the unlearning process as backdoor attacks and defenses: forget-tokens act as backdoor triggers that, when activated in retain-queries, cause disruptions in RM models' behaviors, similar to successful backdoor attacks. To mitigate this vulnerability, we propose Random Noise Augmentation -- a model and method agnostic approach with theoretical guarantees for improving the robustness of RM methods. Extensive experiments demonstrate that RNA significantly improves the robustness of RM models while enhancing the unlearning performances.
Authors: Niklas Muennighoff, Zitong Yang, Weijia Shi, Xiang Lisa Li, Li Fei-Fei, Hannaneh Hajishirzi, Luke Zettlemoyer, Percy Liang, Emmanuel Cand\`es, Tatsunori Hashimoto
Abstract: Test-time scaling is a promising new approach to language modeling that uses extra test-time compute to improve performance. Recently, OpenAI's o1 model showed this capability but did not publicly share its methodology, leading to many replication efforts. We seek the simplest approach to achieve test-time scaling and strong reasoning performance. First, we curate a small dataset s1K of 1,000 questions paired with reasoning traces relying on three criteria we validate through ablations: difficulty, diversity, and quality. Second, we develop budget forcing to control test-time compute by forcefully terminating the model's thinking process or lengthening it by appending "Wait" multiple times to the model's generation when it tries to end. This can lead the model to double-check its answer, often fixing incorrect reasoning steps. After supervised finetuning the Qwen2.5-32B-Instruct language model on s1K and equipping it with budget forcing, our model s1-32B exceeds o1-preview on competition math questions by up to 27% (MATH and AIME24). Further, scaling s1-32B with budget forcing allows extrapolating beyond its performance without test-time intervention: from 50% to 57% on AIME24. Our model, data, and code are open-source at https://github.com/simplescaling/s1
Authors: Hejie Cui, Jiaying Lu, Ran Xu, Shiyu Wang, Wenjing Ma, Yue Yu, Shaojun Yu, Xuan Kan, Chen Ling, Liang Zhao, Zhaohui S. Qin, Joyce C. Ho, Tianfan Fu, Jing Ma, Mengdi Huai, Fei Wang, Carl Yang
Abstract: This comprehensive review aims to provide an overview of the current state of Healthcare Knowledge Graphs (HKGs), including their construction, utilization models, and applications across various healthcare and biomedical research domains. We thoroughly analyzed existing literature on HKGs, covering their construction methodologies, utilization techniques, and applications in basic science research, pharmaceutical research and development, clinical decision support, and public health. The review encompasses both model-free and model-based utilization approaches and the integration of HKGs with large language models (LLMs). We searched Google Scholar for relevant papers on HKGs and classified them into the following topics: HKG construction, HKG utilization, and their downstream applications in various domains. We also discussed their special challenges and the promise for future work. The review highlights the potential of HKGs to significantly impact biomedical research and clinical practice by integrating vast amounts of biomedical knowledge from multiple domains. The synergy between HKGs and LLMs offers promising opportunities for constructing more comprehensive knowledge graphs and improving the accuracy of healthcare applications. HKGs have emerged as a powerful tool for structuring medical knowledge, with broad applications across biomedical research, clinical decision-making, and public health. This survey serves as a roadmap for future research and development in the field of HKGs, highlighting the potential of combining knowledge graphs with advanced machine learning models for healthcare transformation.
Authors: Edmund Mills, Shiye Su, Stuart Russell, Scott Emmons
Abstract: How do we measure the efficacy of language model explainability methods? While many explainability methods have been developed, they are typically evaluated on bespoke tasks, preventing an apples-to-apples comparison. To help fill this gap, we present ALMANACS, a language model explainability benchmark. ALMANACS scores explainability methods on simulatability, i.e., how well the explanations improve behavior prediction on new inputs. The ALMANACS scenarios span twelve safety-relevant topics such as ethical reasoning and advanced AI behaviors; they have idiosyncratic premises to invoke model-specific behavior; and they have a train-test distributional shift to encourage faithful explanations. By using another language model to predict behavior based on the explanations, ALMANACS is a fully automated benchmark. While not a replacement for human evaluations, we aim for ALMANACS to be a complementary, automated tool that allows for fast, scalable evaluation. Using ALMANACS, we evaluate counterfactual, rationalization, attention, and Integrated Gradients explanations. Our results are sobering: when averaged across all topics, no explanation method outperforms the explanation-free control. We conclude that despite modest successes in prior work, developing an explanation method that aids simulatability in ALMANACS remains an open challenge.
Authors: Jiajia Liu, Mengyuan Yang, Yankai Yu, Haixia Xu, Tiangang Wang, Kang Li, Xiaobo Zhou
Abstract: Large language models (LLMs) are a class of artificial intelligence models based on deep learning, which have great performance in various tasks, especially in natural language processing (NLP). Large language models typically consist of artificial neural networks with numerous parameters, trained on large amounts of unlabeled input using self-supervised or semi-supervised learning. However, their potential for solving bioinformatics problems may even exceed their proficiency in modeling human language. In this review, we will provide a comprehensive overview of the essential components of large language models (LLMs) in bioinformatics, spanning genomics, transcriptomics, proteomics, drug discovery, and single-cell analysis. Key aspects covered include tokenization methods for diverse data types, the architecture of transformer models, the core attention mechanism, and the pre-training processes underlying these models. Additionally, we will introduce currently available foundation models and highlight their downstream applications across various bioinformatics domains. Finally, drawing from our experience, we will offer practical guidance for both LLM users and developers, emphasizing strategies to optimize their use and foster further innovation in the field.
Authors: James Flamino, Mohammed Shahid Modi, Boleslaw K. Szymanski, Brendan Cross, Colton Mikolajczyk
Abstract: Large Language Models (LLMs) have shown remarkable promise in communicating with humans. Their potential use as artificial partners with humans in sociological experiments involving conversation is an exciting prospect. But how viable is it? Here, we rigorously test the limits of agents that debate using LLMs in a preregistered study that runs multiple debate-based opinion consensus games. Each game starts with six humans, six agents, or three humans and three agents. We found that agents can blend in and concentrate on a debate's topic better than humans, improving the productivity of all players. Yet, humans perceive agents as less convincing and confident than other humans, and several behavioral metrics of humans and agents we collected deviate measurably from each other. We observed that agents are already decent debaters, but their behavior generates a pattern distinctly different from the human-generated data.
Authors: Robin Staab, Mark Vero, Mislav Balunovi\'c, Martin Vechev
Abstract: Recent privacy research on large language models (LLMs) has shown that they achieve near-human-level performance at inferring personal data from online texts. With ever-increasing model capabilities, existing text anonymization methods are currently lacking behind regulatory requirements and adversarial threats. In this work, we take two steps to bridge this gap: First, we present a new setting for evaluating anonymization in the face of adversarial LLM inferences, allowing for a natural measurement of anonymization performance while remedying some of the shortcomings of previous metrics. Then, within this setting, we develop a novel LLM-based adversarial anonymization framework leveraging the strong inferential capabilities of LLMs to inform our anonymization procedure. We conduct a comprehensive experimental evaluation of adversarial anonymization across 13 LLMs on real-world and synthetic online texts, comparing it against multiple baselines and industry-grade anonymizers. Our evaluation shows that adversarial anonymization outperforms current commercial anonymizers both in terms of the resulting utility and privacy. We support our findings with a human study (n=50) highlighting a strong and consistent human preference for LLM-anonymized texts.
Authors: Wonjoong Kim, Sangwu Park, Yeonjun In, Seokwon Han, Chanyoung Park
Abstract: Recently, interpreting complex charts with logical reasoning has emerged as challenges due to the development of vision-language models. A prior state-of-the-art (SOTA) model has presented an end-to-end method that leverages the vision-language model to convert charts into table format utilizing Large Language Model (LLM) for reasoning. However, unlike natural images, charts contain a mix of essential and irrelevant information required for chart reasoning, and we discover that this characteristic can lower the performance of chart-to-table extraction. In this paper, we introduce SIMPLOT, a method designed to extract only the elements necessary for chart reasoning. The proposed method involves two steps: 1) training to mimic a simple plot that contains only the essential information from a complex chart for table extraction, followed by 2) performing reasoning based on the table. Our model enables accurate chart reasoning without the need for additional annotations or datasets, and its effectiveness is demonstrated through various experiments. Furthermore, we propose a novel prompt mimicking how human interpret charts for more accurate reasoning. Our source code is available at https://github.com/sangwu99/Simplot.
Authors: Tobias Alt, Andrea Ibisch, Clemens Meiser, Anna Wilhelm, Raphael Zimmer, Jonas Ditz, Dominique Dresen, Christoph Droste, Jens Karschau, Friederike Laus, Oliver M\"uller, Matthias Neu, Rainer Plaga, Carola Plesch, Britta Sennewald, Thomas Thaeren, Kristina Unverricht, Steffen Waurick
Abstract: Generative AI models are capable of performing a wide variety of tasks that have traditionally required creativity and human understanding. During training, they learn patterns from existing data and can subsequently generate new content such as texts, images, audio, and videos that align with these patterns. Due to their versatility and generally high-quality results, they represent, on the one hand, an opportunity for digitalisation. On the other hand, the use of generative AI models introduces novel IT security risks that must be considered as part of a comprehensive analysis of the IT security threat landscape. In response to this risk potential, companies or authorities intending to use generative AI should conduct an individual risk analysis before integrating it into their workflows. The same applies to developers and operators, as many risks associated with generative AI must be addressed during development or can only be influenced by the operating organisation. Based on this, existing security measures can be adapted, and additional measures implemented.
Authors: Jiachen Jiang, Jinxin Zhou, Zhihui Zhu
Abstract: Analyzing the similarity of internal representations has been an important technique for understanding the behavior of deep neural networks. Most existing methods for analyzing the similarity between representations of high dimensions, such as those based on Centered Kernel Alignment (CKA), rely on statistical properties of the representations for a set of data points. In this paper, we focus on transformer models and study the similarity of representations between the hidden layers of individual transformers. In this context, we show that a simple sample-wise cosine similarity metric is capable of capturing the similarity and aligns with the complicated CKA. Our experimental results on common transformers reveal that representations across layers are positively correlated, with similarity increasing when layers get closer. We provide a theoretical justification for this phenomenon under the geodesic curve assumption for the learned transformer. We then show that an increase in representation similarity implies an increase in predicted probability when directly applying the last-layer classifier to any hidden layer representation. We then propose an aligned training method to improve the effectiveness of shallow layer by enhancing the similarity between internal representations, with trained models that enjoy the following properties: (1) more early saturation events, (2) layer-wise accuracies monotonically increase and reveal the minimal depth needed for the given task, (3) when served as multi-exit models, they achieve on-par performance with standard multi-exit architectures which consist of additional classifiers designed for early exiting in shallow layers. To our knowledge, our work is the first to show that one common classifier is sufficient for multi-exit models. We conduct experiments on both vision and NLP tasks to demonstrate the performance of the proposed aligned training.
Authors: Rickard Br\"uel-Gabrielsson, Jiacheng Zhu, Onkar Bhardwaj, Leshem Choshen, Kristjan Greenewald, Mikhail Yurochkin, Justin Solomon
Abstract: Fine-tuning large language models (LLMs) with low-rank adaptations (LoRAs) has become common practice, often yielding numerous copies of the same LLM differing only in their LoRA updates. This paradigm presents challenges for systems that serve real-time responses to queries that each involve a different LoRA. Prior works optimize the design of such systems but still require continuous loading and offloading of LoRAs, as it is infeasible to store thousands of LoRAs in GPU memory. To mitigate this issue, we investigate the efficacy of compression when serving LoRAs. We propose a method for the joint compression of LoRAs into a shared basis paired with LoRA-specific scaling matrices. We extend our algorithm to learn clusters of LoRAs that are amenable to joint compression, allowing it to scale gracefully to large LoRA collections. Our experiments with up to 1000 LoRAs demonstrate that compressed LoRAs preserve performance while offering major throughput gains in realistic serving scenarios with over a thousand LoRAs, maintaining 80% of the throughput of serving a single LoRA.
Authors: Ahmed Frikha, Nassim Walha, Krishna Kanth Nakka, Ricardo Mendes, Xue Jiang, Xuebing Zhou
Abstract: In this work, we address the problem of text anonymization where the goal is to prevent adversaries from correctly inferring private attributes of the author, while keeping the text utility, i.e., meaning and semantics. We propose IncogniText, a technique that anonymizes the text to mislead a potential adversary into predicting a wrong private attribute value. Our empirical evaluation shows a reduction of private attribute leakage by more than 90% across 8 different private attributes. Finally, we demonstrate the maturity of IncogniText for real-world applications by distilling its anonymization capability into a set of LoRA parameters associated with an on-device model. Our results show the possibility of reducing privacy leakage by more than half with limited impact on utility.
Authors: Jinwei Hu, Yi Dong, Xiaowei Huang
Abstract: Guardrail, an emerging mechanism designed to ensure that large language models (LLMs) align with human values by moderating harmful or toxic responses, requires a sociotechnical approach in their design. This paper addresses a critical issue: existing guardrails lack a well-founded methodology to accommodate the diverse needs of different user groups, particularly concerning access rights. Supported by trust modeling (primarily on `social' aspect) and enhanced with online in-context learning via retrieval-augmented generation (on `technical' aspect), we introduce an adaptive guardrail mechanism, to dynamically moderate access to sensitive content based on user trust metrics. User trust metrics, defined as a novel combination of direct interaction trust and authority-verified trust, enable the system to precisely tailor the strictness of content moderation by aligning with the user's credibility and the specific context of their inquiries. Our empirical evaluation demonstrates the effectiveness of the adaptive guardrail in meeting diverse user needs, outperforming existing guardrails while securing sensitive information and precisely managing potentially hazardous content through a context-aware knowledge base. To the best of our knowledge, this work is the first to introduce trust-oriented concept into a guardrail system, offering a scalable solution that enriches the discourse on ethical deployment for next-generation LLM service.
Authors: Hoang-Thang Ta, Duy-Quy Thai, Abu Bakar Siddiqur Rahman, Grigori Sidorov, Alexander Gelbukh
Abstract: In this paper, we introduce FC-KAN, a Kolmogorov-Arnold Network (KAN) that leverages combinations of popular mathematical functions such as B-splines, wavelets, and radial basis functions on low-dimensional data through element-wise operations. We explore several methods for combining the outputs of these functions, including sum, element-wise product, the addition of sum and element-wise product, representations of quadratic and cubic functions, concatenation, linear transformation of the concatenated output, and others. In our experiments, we compare FC-KAN with a multi-layer perceptron network (MLP) and other existing KANs, such as BSRBF-KAN, EfficientKAN, FastKAN, and FasterKAN, on the MNIST and Fashion-MNIST datasets. Two variants of FC-KAN, which use a combination of outputs from B-splines and Difference of Gaussians (DoG) and from B-splines and linear transformations in the form of a quadratic function, outperformed overall other models on the average of 5 independent training runs. We expect that FC-KAN can leverage function combinations to design future KANs. Our repository is publicly available at: https://github.com/hoangthangta/FC_KAN.
Authors: Akshat Gupta, Atahan Ozdemir, Gopala Anumanchipalli
Abstract: This paper presents a novel geometric interpretation of LayerNorm and explores how LayerNorm influences the norm and orientation of hidden vectors in the representation space. With these geometric insights, we prepare the foundation for comparing LayerNorm with RMSNorm. We show that the definition of LayerNorm is innately linked to the uniform vector, defined as $\boldsymbol{1} = [1, 1, 1, 1, \cdots, 1]^T \in \mathbb{R}^d$. We then show that the standardization step in LayerNorm can be understood in three simple steps: (i) remove the component of a vector along the uniform vector, (ii) normalize the remaining vector, and (iii) scale the resultant vector by $\sqrt{d}$, where $d$ is the dimensionality of the representation space. We also provide additional insights into how LayerNorm operates at inference time. Finally, we compare the hidden representations of LayerNorm-based LLMs with models trained using RMSNorm and show that all LLMs naturally operate orthogonal to the uniform vector at inference time, that is, on average they do not have a component along the uniform vector during inference. This presents the first mechanistic evidence that removing the component along the uniform vector in LayerNorm is a redundant step. These results advocate for using RMSNorm over LayerNorm which is also more computationally efficient.
Authors: Zhangcheng Qiang, Kerry Taylor, Weiqing Wang, Jing Jiang
Abstract: Hallucinations of large language models (LLMs) commonly occur in domain-specific downstream tasks, with no exception in ontology matching (OM). The prevalence of using LLMs for OM raises the need for benchmarks to better understand LLM hallucinations. The OAEI-LLM dataset is an extended version of the Ontology Alignment Evaluation Initiative (OAEI) datasets that evaluate LLM-specific hallucinations in OM tasks. We outline the methodology used in dataset construction and schema extension, and provide examples of potential use cases.
Authors: Ruizhe Shi, Runlong Zhou, Simon S. Du
Abstract: Direct Preference Optimization (DPO) has emerged as a stable, scalable, and efficient solution for language model alignment. Despite its empirical success, the optimization properties, particularly the impact of samplers on its convergence rates, remain under-explored. In this paper, we provide a rigorous analysis of DPO's convergence rates with different sampling strategies under the exact gradient setting, revealing a surprising separation: uniform sampling achieves $\textbf{linear}$ convergence, while our proposed online sampler achieves $\textbf{quadratic}$ convergence. We further adapt the sampler to practical settings by incorporating posterior distributions and logit mixing, demonstrating improvements over previous methods. For example, it outperforms vanilla DPO by over $7.4$% on Safe-RLHF dataset. Our results not only offer insights into the theoretical understanding of DPO but also pave the way for further algorithm designs.
Authors: Oussama Zekri, Ambroise Odonnat, Abdelhakim Benechehab, Linus Bleistein, Nicolas Boull\'e, Ievgen Redko
Abstract: Large language models (LLMs) are remarkably efficient across a wide range of natural language processing tasks and well beyond them. However, a comprehensive theoretical analysis of the LLMs' generalization capabilities remains elusive. In our paper, we approach this task by drawing an equivalence between autoregressive transformer-based language models and Markov chains defined on a finite state space. This allows us to study the multi-step inference mechanism of LLMs from first principles. We relate the obtained results to the pathological behavior observed with LLMs such as repetitions and incoherent replies with high temperature. Finally, we leverage the proposed formalization to derive pre-training and in-context learning generalization bounds for LLMs under realistic data and model assumptions. Experiments with the most recent Llama and Gemma herds of models show that our theory correctly captures their behavior in practice.
Authors: Haodong Zhao, Jinming Hu, Peixuan Li, Fangqi Li, Jinrui Sha, Tianjie Ju, Peixuan Chen, Zhuosheng Zhang, Gongshen Liu
Abstract: Language models (LMs) have emerged as critical intellectual property (IP) assets that necessitate protection. Although various watermarking strategies have been proposed, they remain vulnerable to Linear Functionality Equivalence Attack (LFEA), which can invalidate most existing white-box watermarks without prior knowledge of the watermarking scheme or training data. This paper analyzes and extends the attack scenarios of LFEA to the commonly employed black-box settings for LMs by considering Last-Layer outputs (dubbed LL-LFEA). We discover that the null space of the output matrix remains invariant against LL-LFEA attacks. Based on this finding, we propose NSmark, a black-box watermarking scheme that is task-agnostic and capable of resisting LL-LFEA attacks. NSmark consists of three phases: (i) watermark generation using the digital signature of the owner, enhanced by spread spectrum modulation for increased robustness; (ii) watermark embedding through an output mapping extractor that preserves the LM performance while maximizing watermark capacity; (iii) watermark verification, assessed by extraction rate and null space conformity. Extensive experiments on both pre-training and downstream tasks confirm the effectiveness, scalability, reliability, fidelity, and robustness of our approach. Code is available at https://github.com/dongdongzhaoUP/NSmark.
Authors: Xin Li, Qizhi Chu, Yubin Chen, Yang Liu, Yaoqi Liu, Zekai Yu, Weize Chen, Chen Qian, Chuan Shi, Cheng Yang
Abstract: Graphs are widely used for modeling relational data in real-world scenarios, such as social networks and urban computing. Existing LLM-based graph analysis approaches either integrate graph neural networks (GNNs) for specific machine learning tasks, limiting their transferability, or rely solely on LLMs' internal reasoning ability, resulting in suboptimal performance. To address these limitations, we take advantage of recent advances in LLM-based agents, which have shown capabilities of utilizing external knowledge or tools for problem solving. By simulating human problem-solving strategies such as analogy and collaboration, we propose a multi-agent system based on LLMs named GraphTeam, for graph analysis. GraphTeam consists of five LLM-based agents from three modules, and the agents with different specialities can collaborate with each other to address complex problems. Specifically, (1) input-output normalization module: the question agent extracts and refines four key arguments from the original question, facilitating the problem understanding, and the answer agent organizes the results to meet the output requirement; (2) external knowledge retrieval module: we first build a knowledge base consisting of relevant documentation and experience information, and then the search agent retrieves the most relevant entries for each question. (3) problem-solving module: given the retrieved information from search agent, the coding agent uses established algorithms via programming to generate solutions, and in case the coding agent does not work, the reasoning agent will directly compute the results without programming. Extensive experiments on six graph analysis benchmarks demonstrate that GraphTeam achieves state-of-the-art performance with an average 25.85% improvement over the best baseline in terms of accuracy. The code and data are available at https://github.com/BUPT-GAMMA/GraphTeam.
Authors: Yuqing Zhou, Ziwei Zhu
Abstract: In text classification tasks, models often rely on spurious correlations for predictions, incorrectly associating irrelevant features with the target labels. This issue limits the robustness and generalization of models, especially when faced with out-of-distribution data where such spurious correlations no longer hold. To address this challenge, we propose the Causally Calibrated Robust Classifier (CCR), which aims to reduce models' reliance on spurious correlations and improve model robustness. Our approach integrates a causal feature selection method based on counterfactual reasoning, along with an unbiased inverse propensity weighting (IPW) loss function. By focusing on selecting causal features, we ensure that the model relies less on spurious features during prediction. We theoretically justify our approach and empirically show that CCR achieves state-of-the-art performance among methods without group labels, and in some cases, it can compete with the models that utilize group labels.
Authors: Qiyao Ma, Menglin Yang, Mingxuan Ju, Tong Zhao, Neil Shah, Rex Ying
Abstract: Modern recommender systems often create information cocoons, restricting users' exposure to diverse content. A key challenge lies in balancing content exploration and exploitation while allowing users to adjust their recommendation preferences. Intuitively, this balance can be modeled as a tree-structured representation, where depth search facilitates exploitation and breadth search enables exploration. However, existing approaches face two fundamental limitations: Euclidean methods struggle to capture hierarchical structures, while hyperbolic methods, despite their superior hierarchical modeling, lack semantic understanding of user and item profiles and fail to provide a principled mechanism for balancing exploration and exploitation. To address these challenges, we propose HERec, a hyperbolic graph-LLM framework that effectively balances exploration and exploitation in recommender systems. Our framework introduces two key innovations: (1) a hierarchical-aware graph-LLM mechanism that jointly aligns textual descriptions with user-item collaborative information in hyperbolic space, and (2) a hierarchical representation structure that enables user-adjustable exploration-exploitation trade-offs. Extensive experiments demonstrate that HERec consistently outperforms both Euclidean and hyperbolic baselines, achieving up to 5.49% improvement in utility metrics and 11.39% increase in diversity metrics, effectively mitigating information cocoons. We open-source our model implementation at https://github.com/Martin-qyma/HERec.
Authors: Yang Xu, Yi Wang, Hao Wang
Abstract: Understanding training dynamics and feature evolution is crucial for the mechanistic interpretability of large language models (LLMs). Although sparse autoencoders (SAEs) have been used to identify features within LLMs, a clear picture of how these features evolve during training remains elusive. In this study, we: (1) introduce SAE-Track, a novel method to efficiently obtain a continual series of SAEs; (2) mechanistically investigate feature formation and develop a progress measure for it ; and (3) analyze and visualize feature drift during training. Our work provides new insights into the dynamics of features in LLMs, enhancing our understanding of training mechanisms and feature evolution.
Authors: Muru Zhang, Mayank Mishra, Zhongzhu Zhou, William Brandon, Jue Wang, Yoon Kim, Jonathan Ragan-Kelley, Shuaiwen Leon Song, Ben Athiwaratkun, Tri Dao
Abstract: Large language model inference is both memory-intensive and time-consuming, often requiring distributed algorithms to efficiently scale. Various model parallelism strategies are used in multi-gpu training and inference to partition computation across multiple devices, reducing memory load and computation time. However, using model parallelism necessitates communication of information between GPUs, which has been a major bottleneck and limits the gains obtained by scaling up the number of devices. We introduce Ladder Residual, a simple architectural modification applicable to all residual-based models that enables straightforward overlapping that effectively hides the latency of communication. Our insight is that in addition to systems optimization, one can also redesign the model architecture to decouple communication from computation. While Ladder Residual can allow communication-computation decoupling in conventional parallelism patterns, we focus on Tensor Parallelism in this paper, which is particularly bottlenecked by its heavy communication. For a Transformer model with 70B parameters, applying Ladder Residual to all its layers can achieve 29% end-to-end wall clock speed up at inference time with TP sharding over 8 devices. We refer the resulting Transformer model as the Ladder Transformer. We train a 1B and 3B Ladder Transformer from scratch and observe comparable performance to a standard dense transformer baseline. We also show that it is possible to convert parts of the Llama-3.1 8B model to our Ladder Residual architecture with minimal accuracy degradation by only retraining for 3B tokens. We release our code for training and inference for easier replication of experiments.
Authors: Niklas Pfister, V\'aclav Volhejn, Manuel Knott, Santiago Arias, Julia Bazi\'nska, Mykhailo Bichurin, Alan Commike, Janet Darling, Peter Dienes, Matthew Fiedler, David Haber, Matthias Kraft, Marco Lancini, Max Mathys, Dami\'an Pascual-Ortiz, Jakub Podolak, Adri\`a Romero-L\'opez, Kyriacos Shiarlis, Andreas Signer, Zsolt Terek, Athanasios Theocharis, Daniel Timbrell, Samuel Trautwein, Samuel Watts, Yun-Han Wu, Mateo Rojas-Carulla
Abstract: Current evaluations of defenses against prompt attacks in large language model (LLM) applications often overlook two critical factors: the dynamic nature of adversarial behavior and the usability penalties imposed on legitimate users by restrictive defenses. We propose D-SEC (Dynamic Security Utility Threat Model), which explicitly separates attackers from legitimate users, models multi-step interactions, and expresses the security-utility in an optimizable form. We further address the shortcomings in existing evaluations by introducing Gandalf, a crowd-sourced, gamified red-teaming platform designed to generate realistic, adaptive attack. Using Gandalf, we collect and release a dataset of 279k prompt attacks. Complemented by benign user data, our analysis reveals the interplay between security and utility, showing that defenses integrated in the LLM (e.g., system prompts) can degrade usability even without blocking requests. We demonstrate that restricted application domains, defense-in-depth, and adaptive defenses are effective strategies for building secure and useful LLM applications.
Authors: Tim Grams, Patrick Betz, Christian Bartelt
Abstract: Exploration is a crucial skill for self-improvement and open-ended problem-solving. However, it remains unclear if large language models can effectively explore the state-space within an unknown environment. This work isolates exploration as the sole objective, tasking the agent with delivering information that enhances future returns. Within this framework, we argue that measuring agent returns is not sufficient for a fair evaluation and decompose missing rewards into exploration and exploitation components based on the optimal achievable return. Comprehensive experiments with various models reveal that most struggle to sufficiently explore the state-space and weak exploration is insufficient. We observe a positive correlation between parameter count and exploration performance, with larger models demonstrating superior capabilities. Furthermore, we show that our decomposition provides insights into differences in behaviors driven by prompt engineering, offering a valuable tool for refining performance in exploratory tasks.
Authors: Aditi Singh, Abul Ehtesham, Saket Kumar, Tala Talaei Khoei
Abstract: Large Language Models (LLMs) have revolutionized artificial intelligence (AI) by enabling human like text generation and natural language understanding. However, their reliance on static training data limits their ability to respond to dynamic, real time queries, resulting in outdated or inaccurate outputs. Retrieval Augmented Generation (RAG) has emerged as a solution, enhancing LLMs by integrating real time data retrieval to provide contextually relevant and up-to-date responses. Despite its promise, traditional RAG systems are constrained by static workflows and lack the adaptability required for multistep reasoning and complex task management. Agentic Retrieval-Augmented Generation (Agentic RAG) transcends these limitations by embedding autonomous AI agents into the RAG pipeline. These agents leverage agentic design patterns reflection, planning, tool use, and multiagent collaboration to dynamically manage retrieval strategies, iteratively refine contextual understanding, and adapt workflows to meet complex task requirements. This integration enables Agentic RAG systems to deliver unparalleled flexibility, scalability, and context awareness across diverse applications. This survey provides a comprehensive exploration of Agentic RAG, beginning with its foundational principles and the evolution of RAG paradigms. It presents a detailed taxonomy of Agentic RAG architectures, highlights key applications in industries such as healthcare, finance, and education, and examines practical implementation strategies. Additionally, it addresses challenges in scaling these systems, ensuring ethical decision making, and optimizing performance for real-world applications, while providing detailed insights into frameworks and tools for implementing Agentic RAG.
Authors: Yuefan Cao, Chengyue Gong, Xiaoyu Li, Yingyu Liang, Zhizhou Sha, Zhenmei Shi, Zhao Song
Abstract: Text-to-video generation models have made impressive progress, but they still struggle with generating videos with complex features. This limitation often arises from the inability of the text encoder to produce accurate embeddings, which hinders the video generation model. In this work, we propose a novel approach to overcome this challenge by selecting the optimal text embedding through interpolation in the embedding space. We demonstrate that this method enables the video generation model to produce the desired videos. Additionally, we introduce a simple algorithm using perpendicular foot embeddings and cosine similarity to identify the optimal interpolation embedding. Our findings highlight the importance of accurate text embeddings and offer a pathway for improving text-to-video generation performance.
Authors: Xiaoyu Li, Yingyu Liang, Jiangxuan Long, Zhenmei Shi, Zhao Song, Zhen Zhuang
Abstract: Looped Transformers have shown exceptional neural algorithmic reasoning capability in simulating traditional graph algorithms, but their application to more complex structures like hypergraphs remains underexplored. Hypergraphs generalize graphs by modeling higher-order relationships among multiple entities, enabling richer representations but introducing significant computational challenges. In this work, we extend the Loop Transformer architecture's neural algorithmic reasoning capability to simulate hypergraph algorithms, addressing the gap between neural networks and combinatorial optimization over hypergraphs. Specifically, we propose a novel degradation mechanism for reducing hypergraphs to graph representations, enabling the simulation of graph-based algorithms, such as Dijkstra's shortest path. Furthermore, we introduce a hyperedge-aware encoding scheme to simulate hypergraph-specific algorithms, exemplified by Helly's algorithm. We establish theoretical guarantees for these simulations, demonstrating the feasibility of processing high-dimensional and combinatorial data using Loop Transformers. This work highlights the potential of Transformers as general-purpose algorithmic solvers for structured data.
Authors: Zunhai Su, Zhe Chen, Wang Shen, Hanyu Wei, Linge Li, Huangqi Yu, Kehong Yuan
Abstract: Key-Value (KV) cache facilitates efficient large language models (LLMs) inference by avoiding recomputation of past KVs. As the batch size and context length increase, the oversized KV caches become a significant memory bottleneck, highlighting the need for efficient compression. Existing KV quantization rely on fine-grained quantization or the retention of a significant portion of high bit-widths caches, both of which compromise compression ratio and often fail to maintain robustness at extremely low average bit-widths. In this work, we explore the potential of rotation technique for 2-bit KV quantization and propose RotateKV, which achieves accurate and robust performance through the following innovations: (i) Outlier-Aware Rotation, which utilizes channel-reordering to adapt the rotations to varying channel-wise outlier distributions without sacrificing the computational efficiency of the fast Walsh-Hadamard transform (FWHT); (ii) Pre-RoPE Grouped-Head Rotation, which mitigates the impact of rotary position embedding (RoPE) on proposed outlier-aware rotation and further smooths outliers across heads; (iii) Attention-Sink-Aware Quantization, which leverages the massive activations to precisely identify and protect attention sinks. RotateKV achieves less than 0.3 perplexity (PPL) degradation with 2-bit quantization on WikiText-2 using LLaMA-2-13B, maintains strong CoT reasoning and long-context capabilities, with less than 1.7\% degradation on GSM8K, outperforming existing methods even at lower average bit-widths. RotateKV also showcases a 3.97x reduction in peak memory usage, supports 5.75x larger batch sizes, and achieves a 2.32x speedup in decoding stage.
Authors: Sergey Berezin, Reza Farahbakhsh, Noel Crespi
Abstract: We present a novel class of jailbreak adversarial attacks on LLMs, termed Task-in-Prompt (TIP) attacks. Our approach embeds sequence-to-sequence tasks (e.g., cipher decoding, riddles, code execution) into the model's prompt to indirectly generate prohibited inputs. To systematically assess the effectiveness of these attacks, we introduce the PHRYGE benchmark. We demonstrate that our techniques successfully circumvent safeguards in six state-of-the-art language models, including GPT-4o and LLaMA 3.2. Our findings highlight critical weaknesses in current LLM safety alignments and underscore the urgent need for more sophisticated defence strategies. Warning: this paper contains examples of unethical inquiries used solely for research purposes.
Authors: Ahmed K. Kadhim, Lei Jiao, Rishad Shafik, Ole-Christoffer Granmo, Bimal Bhattarai
Abstract: The Tsetlin Machine (TM) architecture has recently demonstrated effectiveness in Machine Learning (ML), particularly within Natural Language Processing (NLP). It has been utilized to construct word embedding using conjunctive propositional clauses, thereby significantly enhancing our understanding and interpretation of machine-derived decisions. The previous approach performed the word embedding over a sequence of input words to consolidate the information into a cohesive and unified representation. However, that approach encounters scalability challenges as the input size increases. In this study, we introduce a novel approach incorporating two-phase training to discover contextual embeddings of input sequences. Specifically, this method encapsulates the knowledge for each input word within the dataset's vocabulary, subsequently constructing embeddings for a sequence of input words utilizing the extracted knowledge. This technique not only facilitates the design of a scalable model but also preserves interpretability. Our experimental findings revealed that the proposed method yields competitive performance compared to the previous approaches, demonstrating promising results in contrast to human-generated benchmarks. Furthermore, we applied the proposed approach to sentiment analysis on the IMDB dataset, where the TM embedding and the TM classifier, along with other interpretable classifiers, offered a transparent end-to-end solution with competitive performance.
Authors: Jiefeng Chen, Jie Ren, Xinyun Chen, Chengrun Yang, Ruoxi Sun, Sercan \"O Ar{\i}k
Abstract: Recent advancements in Large Language Models (LLMs) have created new opportunities to enhance performance on complex reasoning tasks by leveraging test-time computation. However, conventional approaches such as repeated sampling with majority voting or reward model scoring, often face diminishing returns as test-time compute scales, in addition to requiring costly task-specific reward model training. In this paper, we present Self-Enhanced Test-Time Scaling (SETS), a novel method that leverages the self-verification and self-correction capabilities of recent advanced LLMs to overcome these limitations. SETS integrates sampling, self-verification, and self-correction into a unified framework, enabling efficient and scalable test-time computation for improved capabilities at complex tasks. Through extensive experiments on challenging planning and reasoning benchmarks, compared to the alternatives, we demonstrate that SETS achieves significant performance improvements and more favorable test-time scaling laws.
Authors: Dominik Wagner, Alexander Churchill, Siddharth Sigtia, Erik Marchi
Abstract: In this work, we present and evaluate SELMA, a Speech-Enabled Language Model for virtual Assistant interactions that integrates audio and text as inputs to a Large Language Model (LLM). SELMA is designed to handle three primary and two auxiliary tasks related to interactions with virtual assistants simultaneously within a single end-to-end model. We employ low-rank adaptation modules for parameter-efficient training of both the audio encoder and the LLM. Additionally, we implement a feature pooling strategy enabling the system to recognize global patterns and improve accuracy on tasks less reliant on individual sequence elements. Experimental results on Voice Trigger (VT) detection, Device-Directed Speech Detection (DDSD), and Automatic Speech Recognition (ASR), demonstrate that our approach both simplifies the typical input processing pipeline of virtual assistants significantly and also improves performance compared to dedicated models for each individual task. SELMA yields relative Equal-Error Rate improvements of 64% on the VT detection task, and 22% on DDSD, while also achieving word error rates close to the baseline.