Authors: Supriya Nagesh, Justin Y. Chen, Nina Mishra, Tal Wagner
Abstract: We explore how private synthetic text can be generated by suitably prompting a large language model (LLM). This addresses a challenge for organizations like hospitals, which hold sensitive text data like patient medical records, and wish to share it in order to train machine learning models for medical tasks, while preserving patient privacy. Methods that rely on training or finetuning a model may be out of reach, either due to API limits of third-party LLMs, or due to ethical and legal prohibitions on sharing the private data with the LLM itself. We propose Differentially Private Keyphrase Prompt Seeding (DP-KPS), a method that generates a private synthetic text corpus from a sensitive input corpus, by accessing an LLM only through privatized prompts. It is based on seeding the prompts with private samples from a distribution over phrase embeddings, thus capturing the input corpus while achieving requisite output diversity and maintaining differential privacy. We evaluate DP-KPS on downstream ML text classification tasks, and show that the corpora it generates preserve much of the predictive power of the original ones. Our findings offer hope that institutions can reap ML insights by privately sharing data with simple prompts and little compute.
Authors: Leonie Weissweiler, Kyle Mahowald, Adele Goldberg
Abstract: Linguistic evaluations of how well LMs generalize to produce or understand novel text often implicitly take for granted that natural languages are generated by symbolic rules. Grammaticality is thought to be determined by whether or not sentences obey such rules. Interpretation is believed to be compositionally generated by syntactic rules operating on meaningful words. Semantic parsing is intended to map sentences into formal logic. Failures of LMs to obey strict rules have been taken to reveal that LMs do not produce or understand language like humans. Here we suggest that LMs' failures to obey symbolic rules may be a feature rather than a bug, because natural languages are not based on rules. New utterances are produced and understood by a combination of flexible interrelated and context-dependent schemata or constructions. We encourage researchers to reimagine appropriate benchmarks and analyses that acknowledge the rich flexible generalizations that comprise natural languages.
Authors: Giorgio Franceschelli, Mirco Musolesi
Abstract: Despite the increasing use of large language models for creative tasks, their outputs often lack diversity. Common solutions, such as sampling at higher temperatures, can compromise the quality of the results. Drawing on information theory, we propose a context-based score to quantitatively evaluate value and originality. This score incentivizes accuracy and adherence to the request while fostering divergence from the learned distribution. We propose using our score as a reward in a reinforcement learning framework to fine-tune large language models for maximum performance. We validate our strategy through experiments in poetry generation and math problem solving, demonstrating that it enhances the value and originality of the generated solutions.
Authors: Yucheng Shi, Tianze Yang, Canyu Chen, Quanzheng Li, Tianming Liu, Xiang Li, Ninghao Liu
Abstract: Large Language Models (LLMs) have shown remarkable capabilities in general domains but often struggle with tasks requiring specialized knowledge. Conventional Retrieval-Augmented Generation (RAG) techniques typically retrieve external information from static knowledge bases, which can be outdated or incomplete, missing fine-grained clinical details essential for accurate medical question answering. In this work, we propose SearchRAG, a novel framework that overcomes these limitations by leveraging real-time search engines. Our method employs synthetic query generation to convert complex medical questions into search-engine-friendly queries and utilizes uncertainty-based knowledge selection to filter and incorporate the most relevant and informative medical knowledge into the LLM's input. Experimental results demonstrate that our method significantly improves response accuracy in medical question answering tasks, particularly for complex questions requiring detailed and up-to-date knowledge.
Authors: Julia Mendelsohn, Ceren Budak
Abstract: Metaphor, discussing one concept in terms of another, is abundant in politics and can shape how people understand important issues. We develop a computational approach to measure metaphorical language, focusing on immigration discourse on social media. Grounded in qualitative social science research, we identify seven concepts evoked in immigration discourse (e.g. "water" or "vermin"). We propose and evaluate a novel technique that leverages both word-level and document-level signals to measure metaphor with respect to these concepts. We then study the relationship between metaphor, political ideology, and user engagement in 400K US tweets about immigration. While conservatives tend to use dehumanizing metaphors more than liberals, this effect varies widely across concepts. Moreover, creature-related metaphor is associated with more retweets, especially for liberal authors. Our work highlights the potential for computational methods to complement qualitative approaches in understanding subtle and implicit language in political discourse.
Authors: Alfonso Amayuelas, Joy Sain, Simerjot Kaur, Charese Smiley
Abstract: Knowledge Graphs (KGs) are valuable tools for representing relationships between entities in a structured format. Traditionally, these knowledge bases are queried to extract specific information. However, question-answering (QA) over such KGs poses a challenge due to the intrinsic complexity of natural language compared to the structured format and the size of these graphs. Despite these challenges, the structured nature of KGs can provide a solid foundation for grounding the outputs of Large Language Models (LLMs), offering organizations increased reliability and control. Recent advancements in LLMs have introduced reasoning methods at inference time to improve their performance and maximize their capabilities. In this work, we propose integrating these reasoning strategies with KGs to anchor every step or "thought" of the reasoning chains in KG data. Specifically, we evaluate both agentic and automated search methods across several reasoning strategies, including Chain-of-Thought (CoT), Tree-of-Thought (ToT), and Graph-of-Thought (GoT), using GRBench, a benchmark dataset for graph reasoning with domain-specific graphs. Our experiments demonstrate that this approach consistently outperforms baseline models, highlighting the benefits of grounding LLM reasoning processes in structured KG data.
Authors: Difan Deng, Marius Lindauer
Abstract: We present Neural Attention Search (NAtS), a framework that automatically evaluates the importance of each token within a sequence and determines if the corresponding token can be dropped after several steps. This approach can efficiently reduce the KV cache sizes required by transformer-based models during inference and thus reduce inference costs. In this paper, we design a search space that contains three token types: (i) Global Tokens will be preserved and queried by all the following tokens. (ii) Local Tokens survive until the next global token appears. (iii) Sliding Window Tokens have an impact on the inference of a fixed size of the next following tokens. Similar to the One-Shot Neural Architecture Search approach, this token-type information can be learned jointly with the architecture weights via a learnable attention mask. Experiments on both training a new transformer from scratch and fine-tuning existing large language models show that NAtS can efficiently reduce the KV cache size required for the models while maintaining the models' performance.
Authors: Jiayi Wang, Yao Lu, Maurice Weber, Max Ryabinin, David Adelani, Yihong Chen, Raphael Tang, Pontus Stenetorp
Abstract: High-resource languages such as English, enables the pretraining of high-quality large language models (LLMs). The same can not be said for most other languages as LLMs still underperform for non-English languages, likely due to a gap in the quality and diversity of the available multilingual pretraining corpora. In this work, we find that machine-translated texts from a single high-quality source language can contribute significantly to the pretraining quality of multilingual LLMs. We translate FineWeb-Edu, a high-quality English web dataset, into nine languages, resulting in a 1.7-trillion-token dataset, which we call TransWebEdu and pretrain a 1.3B-parameter model, TransWebLLM, from scratch on this dataset. Across nine non-English reasoning tasks, we show that TransWebLLM matches or outperforms state-of-the-art multilingual models trained using closed data, such as Llama3.2, Qwen2.5, and Gemma, despite using an order of magnitude less data. We demonstrate that adding less than 5% of TransWebEdu as domain-specific pretraining data sets a new state-of-the-art in Arabic, Italian, Indonesian, Swahili, and Welsh understanding and commonsense reasoning tasks. To promote reproducibility, we release our corpus, models, and training pipeline under Open Source Initiative-approved licenses.
Authors: Myra Cheng, Sunny Yu, Dan Jurafsky
Abstract: Should LLMs generate language that makes them seem human? Human-like language might improve user experience, but might also lead to overreliance and stereotyping. Assessing these potential impacts requires a systematic way to measure human-like tone in LLM outputs. We introduce HumT and SocioT, metrics for human-like tone and other dimensions of social perceptions in text data based on relative probabilities from an LLM. By measuring HumT across preference and usage datasets, we find that users prefer less human-like outputs from LLMs. HumT also offers insights into the impacts of anthropomorphism: human-like LLM outputs are highly correlated with warmth, social closeness, femininity, and low status, which are closely linked to the aforementioned harms. We introduce DumT, a method using HumT to systematically control and reduce the degree of human-like tone while preserving model performance. DumT offers a practical approach for mitigating risks associated with anthropomorphic language generation.
Authors: Yingqian Cui, Pengfei He, Jingying Zeng, Hui Liu, Xianfeng Tang, Zhenwei Dai, Yan Han, Chen Luo, Jing Huang, Zhen Li, Suhang Wang, Yue Xing, Jiliang Tang, Qi He
Abstract: Chain-of-Thought (CoT) reasoning, which breaks down complex tasks into intermediate reasoning steps, has significantly enhanced the performance of large language models (LLMs) on challenging tasks. However, the detailed reasoning process in CoT often incurs long generation times and high computational costs, partly due to the inclusion of unnecessary steps. To address this, we propose a method to identify critical reasoning steps using perplexity as a measure of their importance: a step is deemed critical if its removal causes a significant increase in perplexity. Our method enables models to focus solely on generating these critical steps. This can be achieved through two approaches: refining demonstration examples in few-shot CoT or fine-tuning the model using selected examples that include only critical steps. Comprehensive experiments validate the effectiveness of our method, which achieves a better balance between the reasoning accuracy and efficiency of CoT.
Authors: Dong-Ho Lee, Adyasha Maharana, Jay Pujara, Xiang Ren, Francesco Barbieri
Abstract: Long-term, open-domain dialogue capabilities are essential for chatbots aiming to recall past interactions and demonstrate emotional intelligence (EI). Yet, most existing research relies on synthetic, LLM-generated data, leaving open questions about real-world conversational patterns. To address this gap, we introduce REALTALK, a 21-day corpus of authentic messaging app dialogues, providing a direct benchmark against genuine human interactions. We first conduct a dataset analysis, focusing on EI attributes and persona consistency to understand the unique challenges posed by real-world dialogues. By comparing with LLM-generated conversations, we highlight key differences, including diverse emotional expressions and variations in persona stability that synthetic dialogues often fail to capture. Building on these insights, we introduce two benchmark tasks: (1) persona simulation where a model continues a conversation on behalf of a specific user given prior dialogue context; and (2) memory probing where a model answers targeted questions requiring long-term memory of past interactions. Our findings reveal that models struggle to simulate a user solely from dialogue history, while fine-tuning on specific user chats improves persona emulation. Additionally, existing models face significant challenges in recalling and leveraging long-term context within real-world conversations.
Authors: Sirisha Velampalli, Chandrashekar Muniyappa, Ashutosh Saxena
Abstract: Emojis are being frequently used in todays digital world to express from simple to complex thoughts more than ever before. Hence, they are also being used in sentiment analysis and targeted marketing campaigns. In this work, we performed sentiment analysis of Tweets as well as on emoji dataset from the Kaggle. Since tweets are sentences we have used Universal Sentence Encoder (USE) and Sentence Bidirectional Encoder Representations from Transformers (SBERT) end-to-end sentence embedding models to generate the embeddings which are used to train the Standard fully connected Neural Networks (NN), and LSTM NN models. We observe the text classification accuracy was almost the same for both the models around 98 percent. On the contrary, when the validation set was built using emojis that were not present in the training set then the accuracy of both the models reduced drastically to 70 percent. In addition, the models were also trained using the distributed training approach instead of a traditional singlethreaded model for better scalability. Using the distributed training approach, we were able to reduce the run-time by roughly 15% without compromising on accuracy. Finally, as part of explainable AI the Shap algorithm was used to explain the model behaviour and check for model biases for the given feature set.
Authors: Yunpeng Xiao, Youpeng Zhao, Kai Shu
Abstract: Natural language understanding (NLU) is a task that enables machines to understand human language. Some tasks, such as stance detection and sentiment analysis, are closely related to individual subjective perspectives, thus termed individual-level NLU. Previously, these tasks are often simplified to text-level NLU tasks, ignoring individual factors. This not only makes inference difficult and unexplainable but often results in a large number of label errors when creating datasets. To address the above limitations, we propose a new NLU annotation guideline based on individual-level factors. Specifically, we incorporate other posts by the same individual and then annotate individual subjective perspectives after considering all individual posts. We use this guideline to expand and re-annotate the stance detection and topic-based sentiment analysis datasets. We find that error rates in the samples were as high as 31.7\% and 23.3\%. We further use large language models to conduct experiments on the re-annotation datasets and find that the large language models perform well on both datasets after adding individual factors. Both GPT-4o and Llama3-70B can achieve an accuracy greater than 87\% on the re-annotation datasets. We also verify the effectiveness of individual factors through ablation studies. We call on future researchers to add individual factors when creating such datasets. Our re-annotation dataset can be found at https://github.com/24yearsoldstudent/Individual-NLU
Authors: Moghis Fereidouni, Md Sajid Ahmed, Adib Mosharrof, A. B. Siddique
Abstract: Task-oriented dialog (TOD) systems facilitate users in accomplishing complex, multi-turn tasks through natural language. While traditional approaches rely on extensive fine-tuning and annotated data for each domain, instruction-tuned large language models (LLMs) offer a more flexible alternative. However, LLMs struggle to reliably handle multi-turn task completion, particularly with accurately generating API calls and adapting to new domains without explicit demonstrations. To address these challenges, we propose RealTOD, a novel framework that enhances TOD systems through prompt chaining and fine-grained feedback mechanisms. Prompt chaining enables zero-shot domain adaptation via a two-stage prompting strategy, eliminating the need for human-curated demonstrations. Meanwhile, the fine-grained feedback mechanism improves task completion by verifying API calls against domain schemas and providing precise corrective feedback when errors are detected. We conduct extensive experiments on the SGD and BiTOD benchmarks using four LLMs. RealTOD improves API accuracy, surpassing AutoTOD by 37.74% on SGD and SimpleTOD by 11.26% on BiTOD. Human evaluations further confirm that LLMs integrated with RealTOD achieve superior task completion, fluency, and informativeness compared to existing methods.
Authors: Adib Mosharrof, Moghis Fereidouni, A. B. Siddique
Abstract: Traditional task-oriented dialog (ToD) systems rely heavily on labor-intensive turn-level annotations, such as dialogue states and policy labels, for training. This work explores whether large language models (LLMs) can be fine-tuned solely on natural language dialogs to perform ToD tasks, without requiring such annotations. We evaluate their ability to generalize to unseen domains and compare their performance with models trained on fully annotated data. Through extensive experiments with three open-source LLMs of varying sizes and two diverse ToD datasets, we find that models fine-tuned without turn-level annotations generate coherent and contextually appropriate responses. However, their task completion performance - measured by accurate execution of API calls - remains suboptimal, with the best models achieving only around 53% success in unseen domains. To improve task completion, we propose ZeroToD, a framework that incorporates a schema augmentation mechanism to enhance API call accuracy and overall task completion rates, particularly in out-of-domain settings. We also compare ZeroToD with fine-tuning-free alternatives, such as prompting off-the-shelf LLMs, and find that our framework enables smaller, fine-tuned models that outperform large-scale proprietary LLMs in task completion. Additionally, a human study evaluating informativeness, fluency, and task completion confirms our empirical findings. These findings suggest the feasibility of developing cost-effective, scalable, and zero-shot generalizable ToD systems for real-world applications.
Authors: Jian Wang, Yinpei Dai, Yichi Zhang, Ziqiao Ma, Wenjie Li, Joyce Chai
Abstract: Intelligent tutoring agents powered by large language models (LLMs) have been increasingly explored to deliver personalized guidance in areas such as language learning and science education. However, their capabilities in guiding users to solve complex real-world tasks remain underexplored. To address this limitation, in this work, we focus on coding tutoring, a challenging problem that requires tutors to proactively guide students toward completing predefined coding tasks. We propose a novel agent workflow, Trace-and-Verify (TRAVER), which combines knowledge tracing to estimate a student's knowledge state and turn-by-turn verification to ensure effective guidance toward task completion. We introduce DICT, an automatic evaluation protocol that assesses tutor agents holistically using controlled student simulation and code generation tests. Extensive experiments reveal the challenges of coding tutoring and demonstrate that TRAVER achieves a significantly higher success rate. Although we use code tutoring as an example in this paper, our results and findings can be extended beyond coding, providing valuable insights into advancing tutoring agents for a variety of tasks.
Authors: Hiba Ahsan, Arnab Sen Sharma, Silvio Amir, David Bau, Byron C. Wallace
Abstract: We know from prior work that LLMs encode social biases, and that this manifests in clinical tasks. In this work we adopt tools from mechanistic interpretability to unveil sociodemographic representations and biases within LLMs in the context of healthcare. Specifically, we ask: Can we identify activations within LLMs that encode sociodemographic information (e.g., gender, race)? We find that gender information is highly localized in middle MLP layers and can be reliably manipulated at inference time via patching. Such interventions can surgically alter generated clinical vignettes for specific conditions, and also influence downstream clinical predictions which correlate with gender, e.g., patient risk of depression. We find that representation of patient race is somewhat more distributed, but can also be intervened upon, to a degree. To our knowledge, this is the first application of mechanistic interpretability methods to LLMs for healthcare.
Authors: Vasudha Varadarajan, Syeda Mahwish, Xiaoran Liu, Julia Buffolino, Christian C. Luhmann, Ryan L. Boyd, H. Andrew Schwartz
Abstract: While NLP models often seek to capture cognitive states via language, the validity of predicted states is determined by comparing them to annotations created without access the cognitive states of the authors. In behavioral sciences, cognitive states are instead measured via experiments. Here, we introduce an experiment-based framework for evaluating language-based cognitive style models against human behavior. We explore the phenomenon of decision making, and its relationship to the linguistic style of an individual talking about a recent decision they made. The participants then follow a classical decision-making experiment that captures their cognitive style, determined by how preferences change during a decision exercise. We find that language features, intended to capture cognitive style, can predict participants' decision style with moderate-to-high accuracy (AUC ~ 0.8), demonstrating that cognitive style can be partly captured and revealed by discourse patterns.
Authors: Dhananjay Ashok, Jonathan May
Abstract: Autoregressive Language Models output text by sequentially predicting the next token to generate, with modern methods like Chain-of-Thought (CoT) prompting achieving state-of-the-art reasoning capabilities by scaling the number of generated tokens. However, are there times when we can infer how the model will behave (e.g. abstain from answering a question) early in the computation, making generation unnecessary? We show that internal representation of input tokens alone can often precisely predict, not just the next token, but eventual behavior over the entire output sequence. We leverage this capacity and learn probes on internal states to create early warning (and exit) systems. Specifically, if the probes can confidently estimate the way the LM is going to behave, then the system will avoid generating tokens altogether and return the estimated behavior instead. On 27 text classification datasets spanning five different tasks, we apply this method to estimate the eventual answer of an LM under CoT prompting, reducing inference costs by 65% (average) while suffering an accuracy loss of no more than 1.4% (worst case). We demonstrate the potential of this method to pre-emptively identify when a model will abstain from answering a question, fail to follow output format specifications, or give a low-confidence response. We explore the limits of this capability, showing that probes generalize to unseen datasets, but perform worse when LM outputs are longer and struggle to predict properties that require access to knowledge that the models themselves lack. Encouragingly, performance scales with model size, suggesting applicability to the largest of models
Authors: Chenyan Zhao, Mariana Silva, Seth Poulsen
Abstract: Providing evaluations to student work is a critical component of effective student learning, and automating its process can significantly reduce the workload on human graders. Automatic Short Answer Grading (ASAG) systems, enabled by advancements in Large Language Models (LLMs), offer a promising solution for assessing and providing instant feedback for open-ended student responses. In this paper, we present an ASAG pipeline leveraging state-of-the-art LLMs. Our new LLM-based ASAG pipeline achieves better performances than existing custom-built models on the same datasets. We also compare the grading performance of three OpenAI models: GPT-4, GPT-4o, and o1-preview. Our results demonstrate that GPT-4o achieves the best balance between accuracy and cost-effectiveness. On the other hand, o1-preview, despite higher accuracy, exhibits a larger variance in error that makes it less practical for classroom use. We investigate the effects of incorporating instructor-graded examples into prompts using no examples, random selection, and Retrieval-Augmented Generation (RAG)-based selection strategies. Our findings indicate that providing graded examples enhances grading accuracy, with RAG-based selection outperforming random selection. Additionally, integrating grading rubrics improves accuracy by offering a structured standard for evaluation.
Authors: Ibrahim Baroud, Lisa Raithel, Sebastian M\"oller, Roland Roller
Abstract: Sharing sensitive texts for scientific purposes requires appropriate techniques to protect the privacy of patients and healthcare personnel. Anonymizing textual data is particularly challenging due to the presence of diverse unstructured direct and indirect identifiers. To mitigate the risk of re-identification, this work introduces a schema of nine categories of indirect identifiers designed to account for different potential adversaries, including acquaintances, family members and medical staff. Using this schema, we annotate 100 MIMIC-III discharge summaries and propose baseline models for identifying indirect identifiers. We will release the annotation guidelines, annotation spans (6,199 annotations in total) and the corresponding MIMIC-III document IDs to support further research in this area.
Authors: Shi Yu, Zhiyuan Liu, Chenyan Xiong
Abstract: Web crawl is a main source of large language models' (LLMs) pretraining data, but the majority of crawled web pages are discarded in pretraining due to low data quality. This paper presents Crawl4LLM, an efficient web crawling method that explores the web graph based on the preference of LLM pretraining. Specifically, it leverages the influence of a webpage in LLM pretraining as the priority score of the web crawler's scheduler, replacing the standard graph connectivity based priority. Our experiments on a web graph containing 900 million webpages from a commercial search engine's index demonstrate the efficiency of Crawl4LLM in obtaining high-quality pretraining data. With just 21% URLs crawled, LLMs pretrained on Crawl4LLM data reach the same downstream performances of previous crawls, significantly reducing the crawling waste and alleviating the burdens on websites. Our code is publicly available at https://github.com/cxcscmu/Crawl4LLM.
Authors: Ryan A. Panela (Rotman Research Institute, Baycrest Academy for Research and Education, Department of Psychology, University of Toronto), Alex J. Barnett (Department of Psychology, University of Toronto, Department of Neurology and Neurosurgery, Montreal Neurological Institute and Hospital, McGill University), Morgan D. Barense (Rotman Research Institute, Baycrest Academy for Research and Education, Department of Psychology, University of Toronto), Bj\"orn Herrmann (Rotman Research Institute, Baycrest Academy for Research and Education, Department of Psychology, University of Toronto)
Abstract: Understanding how individuals perceive and recall information in their natural environments is critical to understanding potential failures in perception (e.g., sensory loss) and memory (e.g., dementia). Event segmentation, the process of identifying distinct events within dynamic environments, is central to how we perceive, encode, and recall experiences. This cognitive process not only influences moment-to-moment comprehension but also shapes event specific memory. Despite the importance of event segmentation and event memory, current research methodologies rely heavily on human judgements for assessing segmentation patterns and recall ability, which are subjective and time-consuming. A few approaches have been introduced to automate event segmentation and recall scoring, but validity with human responses and ease of implementation require further advancements. To address these concerns, we leverage Large Language Models (LLMs) to automate event segmentation and assess recall, employing chat completion and text-embedding models, respectively. We validated these models against human annotations and determined that LLMs can accurately identify event boundaries, and that human event segmentation is more consistent with LLMs than among humans themselves. Using this framework, we advanced an automated approach for recall assessments which revealed semantic similarity between segmented narrative events and participant recall can estimate recall performance. Our findings demonstrate that LLMs can effectively simulate human segmentation patterns and provide recall evaluations that are a scalable alternative to manual scoring. This research opens novel avenues for studying the intersection between perception, memory, and cognitive impairment using methodologies driven by artificial intelligence.
Authors: Yiming Zeng, Wanhao Yu, Zexin Li, Tao Ren, Yu Ma, Jinghan Cao, Xiyan Chen, Tingting Yu
Abstract: Large Language Models (LLMs) have transformed natural language processing, yet they still struggle with direct text editing tasks that demand precise, context-aware modifications. While models like ChatGPT excel in text generation and analysis, their editing abilities often fall short, addressing only superficial issues rather than deeper structural or logical inconsistencies. In this work, we introduce a dual approach to enhance LLMs editing performance. First, we present InstrEditBench, a high-quality benchmark dataset comprising over 20,000 structured editing tasks spanning Wiki articles, LaTeX documents, code, and database Domain-specific Languages (DSL). InstrEditBench is generated using an innovative automated workflow that accurately identifies and evaluates targeted edits, ensuring that modifications adhere strictly to specified instructions without altering unrelated content. Second, we propose FineEdit, a specialized model trained on this curated benchmark. Experimental results demonstrate that FineEdit achieves significant improvements around {10\%} compared with Gemini on direct editing tasks, convincingly validating its effectiveness.
Authors: Sichu Liang, Linhai Zhang, Hongyu Zhu, Wenwen Wang, Yulan He, Deyu Zhou
Abstract: Medical question answering requires extensive access to specialized conceptual knowledge. The current paradigm, Retrieval-Augmented Generation (RAG), acquires expertise medical knowledge through large-scale corpus retrieval and uses this knowledge to guide a general-purpose large language model (LLM) for generating answers. However, existing retrieval approaches often overlook the importance of factual knowledge, which limits the relevance of retrieved conceptual knowledge and restricts its applicability in real-world scenarios, such as clinical decision-making based on Electronic Health Records (EHRs). This paper introduces RGAR, a recurrence generation-augmented retrieval framework that retrieves both relevant factual and conceptual knowledge from dual sources (i.e., EHRs and the corpus), allowing them to interact and refine each another. Through extensive evaluation across three factual-aware medical question answering benchmarks, RGAR establishes a new state-of-the-art performance among medical RAG systems. Notably, the Llama-3.1-8B-Instruct model with RGAR surpasses the considerably larger, RAG-enhanced GPT-3.5. Our findings demonstrate the benefit of extracting factual knowledge for retrieval, which consistently yields improved generation quality.
Authors: Aditya Sharma, Luis Lara, Amal Zouaq, Christopher J. Pal
Abstract: The ability to generate SPARQL queries from natural language questions is crucial for ensuring efficient and accurate retrieval of structured data from knowledge graphs (KG). While large language models (LLMs) have been widely adopted for SPARQL query generation, they are often susceptible to hallucinations and out-of-distribution errors when producing KG elements like Uniform Resource Identifiers (URIs) based on internal parametric knowledge. This often results in content that appears plausible but is factually incorrect, posing significant challenges for their use in real-world information retrieval (IR) applications. This has led to increased research aimed at detecting and mitigating such errors. In this paper, we introduce PGMR (Post-Generation Memory Retrieval), a modular framework that incorporates a non-parametric memory module to retrieve KG elements and enhance LLM-based SPARQL query generation. Our experimental results indicate that PGMR consistently delivers strong performance across diverse datasets, data distributions, and LLMs. Notably, PGMR significantly mitigates URI hallucinations, nearly eliminating the problem in several scenarios.
Authors: Barys Liskavets, Shuvendu Roy, Maxim Ushakov, Mark Klibanov, Ali Etemad, Shane Luke
Abstract: The rise of Large Language Models (LLMs) has led to significant interest in prompt compression, a technique aimed at reducing the length of input prompts while preserving critical information. However, the prominent approaches in prompt compression often require explicit questions or handcrafted templates for compression, limiting their generalizability. We propose Task-agnostic Prompt Compression (TPC), a novel framework that generalizes compression across tasks and domains without requiring input questions or templates. TPC generates a context-relevant task description using a task descriptor trained on a curated dataset of context and query pairs, and fine-tuned via reinforcement learning with a reward function designed to capture the most relevant information. The task descriptor is then utilized to compute the relevance of each sentence in the prompt to generate the compressed prompt. We introduce 3 model sizes (Base, Large, and Huge), where the largest model outperforms the existing state-of-the-art methods on LongBench and ZeroSCROLLS benchmarks, and our smallest model performs comparable to the existing solutions while being considerably smaller.
Authors: Linzhuang Sun, Hao Liang, Jingxuan Wei, Bihui Yu, Tianpeng Li, Fan Yang, Zenan Zhou, Wentao Zhang
Abstract: According to the Test-Time Scaling, the integration of External Slow-Thinking with the Verify mechanism has been demonstrated to enhance multi-round reasoning in large language models (LLMs). However, in the multimodal (MM) domain, there is still a lack of a strong MM-Verifier. In this paper, we introduce MM-Verifier and MM-Reasoner to enhance multimodal reasoning through longer inference and more robust verification. First, we propose a two-step MM verification data synthesis method, which combines a simulation-based tree search with verification and uses rejection sampling to generate high-quality Chain-of-Thought (COT) data. This data is then used to fine-tune the verification model, MM-Verifier. Additionally, we present a more efficient method for synthesizing MMCOT data, bridging the gap between text-based and multimodal reasoning. The synthesized data is used to fine-tune MM-Reasoner. Our MM-Verifier outperforms all larger models on the MathCheck, MathVista, and MathVerse benchmarks. Moreover, MM-Reasoner demonstrates strong effectiveness and scalability, with performance improving as data size increases. Finally, our approach achieves strong performance when combining MM-Reasoner and MM-Verifier, reaching an accuracy of 65.3 on MathVista, surpassing GPT-4o (63.8) with 12 rollouts.
Authors: Wenwen Xie, Gray Gwizdz, Dongji Feng
Abstract: While Large Language Models (LLMs) have emerged as promising tools for evaluating Natural Language Generation (NLG) tasks, their effectiveness is limited by their inability to appropriately weigh the importance of different topics, often overemphasizing minor details while undervaluing critical information, leading to misleading assessments. Our work proposes an efficient prompt design mechanism to address this specific limitation and provide a case study. Through strategic prompt engineering that incorporates explicit importance weighting mechanisms, we enhance using LLM-as-a-Judge ability to prioritize relevant information effectively, as demonstrated by an average improvement of 6% in the Human Alignment Rate (HAR) metric.
Authors: Ningke Li, Yahui Song, Kailong Wang, Yuekang Li, Ling Shi, Yi Liu, Haoyu Wang
Abstract: Large language models (LLMs) face the challenge of hallucinations -- outputs that seem coherent but are actually incorrect. A particularly damaging type is fact-conflicting hallucination (FCH), where generated content contradicts established facts. Addressing FCH presents three main challenges: 1) Automatically constructing and maintaining large-scale benchmark datasets is difficult and resource-intensive; 2) Generating complex and efficient test cases that the LLM has not been trained on -- especially those involving intricate temporal features -- is challenging, yet crucial for eliciting hallucinations; and 3) Validating the reasoning behind LLM outputs is inherently difficult, particularly with complex logical relationships, as it requires transparency in the model's decision-making process. This paper presents Drowzee, an innovative end-to-end metamorphic testing framework that utilizes temporal logic to identify fact-conflicting hallucinations (FCH) in large language models (LLMs). Drowzee builds a comprehensive factual knowledge base by crawling sources like Wikipedia and uses automated temporal-logic reasoning to convert this knowledge into a large, extensible set of test cases with ground truth answers. LLMs are tested using these cases through template-based prompts, which require them to generate both answers and reasoning steps. To validate the reasoning, we propose two semantic-aware oracles that compare the semantic structure of LLM outputs to the ground truths. Across nine LLMs in nine different knowledge domains, experimental results show that Drowzee effectively identifies rates of non-temporal-related hallucinations ranging from 24.7% to 59.8%, and rates of temporal-related hallucinations ranging from 16.7% to 39.2%.
Authors: Yifei Xu, Tusher Chakraborty, Emre K{\i}c{\i}man, Bibek Aryal, Eduardo Rodrigues, Srinagesh Sharma, Roberto Estevao, Maria Angels de Luis Balaguer, Jessica Wolk, Rafael Padilha, Leonardo Nunes, Shobana Balakrishnan, Songwu Lu, Ranveer Chandra
Abstract: Fine-tuning large language models (LLMs) to align with user preferences is challenging due to the high cost of quality human annotations in Reinforcement Learning from Human Feedback (RLHF) and the generalizability limitations of AI Feedback. To address these challenges, we propose RLTHF, a human-AI hybrid framework that combines LLM-based initial alignment with selective human annotations to achieve full-human annotation alignment with minimal effort. RLTHF identifies hard-to-annotate samples mislabeled by LLMs using a reward model's reward distribution and iteratively enhances alignment by integrating strategic human corrections while leveraging LLM's correctly labeled samples. Evaluations on HH-RLHF and TL;DR datasets show that RLTHF reaches full-human annotation-level alignment with only 6-7% of the human annotation effort. Furthermore, models trained on RLTHF's curated datasets for downstream tasks outperform those trained on fully human-annotated datasets, underscoring the effectiveness of RLTHF's strategic data curation.
Authors: Yuxiang Wang, Junhao Gan, Jianzhong Qi
Abstract: Question answering on free-form tables (TableQA) is challenging due to the absence of predefined schemas and the presence of noise in large tables. While Large Language Models (LLMs) have shown promise in TableQA, they struggle with large free-form tables and noise sensitivity. To address these challenges, we propose TabSD, a SQL-based decomposition model that enhances LLMs' ability to process large free-form tables. TabSD generates SQL queries to guide the table decomposition, remove noise, and processes sub-tables for better answer generation. Additionally, SQL Verifier refines SQL outputs to enhance decomposition accuracy. We introduce two TableQA datasets with large free-form tables, SLQA and SEQA, which consist solely of large free-form tables and will be publicly available. Experimental results on four benchmark datasets demonstrate that TABSD outperforms the best-existing baseline models by 23.07%, 2.84%, 23.24% and 9.32% in accuracy, respectively, highlighting its effectiveness in handling large and noisy free-form tables.
Authors: Guanming Xiong, Haochen Li, Wen Zhao
Abstract: This study explores how to enhance the reasoning capabilities of large language models (LLMs) in knowledge base question answering (KBQA) by leveraging Monte Carlo Tree Search (MCTS). Semantic parsing-based KBQA methods are particularly challenging as these approaches require locating elements from knowledge bases and generating logical forms, demanding not only extensive annotated data but also strong reasoning capabilities. Although recent approaches leveraging LLMs as agents have demonstrated considerable potential, these studies are inherently constrained by their linear decision-making processes. To address this limitation, we propose a MCTS-based framework that enhances LLMs' reasoning capabilities through tree search methodology. We design a carefully designed step-wise reward mechanism that requires only direct prompting of open-source instruction LLMs without additional fine-tuning. Experimental results demonstrate that our approach significantly outperforms linear decision-making methods, particularly in low-resource scenarios. Additionally, we contribute new data resources to the KBQA community by annotating intermediate reasoning processes for existing question-SPARQL datasets using distant supervision. Experimental results on the extended dataset demonstrate that our method achieves comparable performance to fully supervised models while using significantly less training data.
Authors: Yutao Sun, Mingshuai Chen, Tiancheng Zhao, Ruochen Xu, Zilun Zhang, Jianwei Yin
Abstract: Self-improving large language models (LLMs) -- i.e., to improve the performance of an LLM by fine-tuning it with synthetic data generated by itself -- is a promising way to advance the capabilities of LLMs while avoiding extensive supervision. Existing approaches to self-improvement often rely on external supervision signals in the form of seed data and/or assistance from third-party models. This paper presents Crescent -- a simple yet effective framework for generating high-quality synthetic question-answer data in a fully autonomous manner. Crescent first elicits the LLM to generate raw questions via a bait prompt, then diversifies these questions leveraging a rejection sampling-based self-deduplication, and finally feeds the questions to the LLM and collects the corresponding answers by means of majority voting. We show that Crescent sheds light on the potential of true self-improvement with zero external supervision signals for math reasoning; in particular, Crescent-generated question-answer pairs suffice to (i) improve the reasoning capabilities of an LLM while preserving its general performance (especially in the 0-shot setting); and (ii) distil LLM knowledge to weaker models more effectively than existing methods based on seed-dataset augmentation.
Authors: Jialin Ouyang
Abstract: Large language models (LLMs) now achieve near-human performance on standard math word problem benchmarks (e.g., GSM8K), yet their true reasoning ability remains disputed. A key concern is that models often produce confident, yet unfounded, answers to unanswerable problems. We introduce TreeCut, a synthetic dataset that systematically generates infinite unanswerable math word problems and their answerable counterparts, by representing each question as a tree and removing chosen necessary conditions. Experiments show TreeCut effectively induce hallucinations in large language models, including GPT-4o and o3-mini, with rates of 61% and 42% in their respective worst-case scenarios. Further analysis highlights that deeper or more complex trees, composite item names, and removing necessary condition near the middle of a path all increase the likelihood of hallucinations, underscoring the persistent challenges LLMs face in identifying unanswerable math problems.
Authors: Xiaofei Wen, Wenxuan Zhou, Wenjie Jacky Mo, Muhao Chen
Abstract: Ensuring the safety of large language models (LLMs) is critical as they are deployed in real-world applications. Existing guardrails rely on rule-based filtering or single-pass classification, limiting their ability to handle nuanced safety violations. To address this, we propose ThinkGuard, a critique-augmented guardrail model that distills knowledge from high-capacity LLMs by generating structured critiques alongside safety labels. Fine-tuned on critique-augmented data, the captured deliberative thinking ability drastically enhances the guardrail's cautiousness and interpretability. Evaluated on multiple safety benchmarks, ThinkGuard achieves the highest average F1 and AUPRC, outperforming all baselines. Compared to LLaMA Guard 3, ThinkGuard improves accuracy by 16.1% and macro F1 by 27.0%. Moreover, it surpasses label-only fine-tuned models, confirming that structured critiques enhance both classification precision and nuanced safety reasoning while maintaining computational efficiency.
Authors: Wanqing Cui, Keping Bi, Jiafeng Guo, Xueqi Cheng
Abstract: Commonsense plausibility estimation is critical for evaluating language models (LMs), yet existing generative approaches--reliant on likelihoods or verbalized judgments--struggle with fine-grained discrimination. In this paper, we propose ComPaSS, a novel discriminative framework that quantifies commonsense plausibility by measuring semantic shifts when augmenting sentences with commonsense-related information. Plausible augmentations induce minimal shifts in semantics, while implausible ones result in substantial deviations. Evaluations on two types of fine-grained commonsense plausibility estimation tasks across different backbones, including LLMs and vision-language models (VLMs), show that ComPaSS consistently outperforms baselines. It demonstrates the advantage of discriminative approaches over generative methods in fine-grained commonsense plausibility evaluation. Experiments also show that (1) VLMs yield superior performance to LMs, when integrated with ComPaSS, on vision-grounded commonsense tasks. (2) contrastive pre-training sharpens backbone models' ability to capture semantic nuances, thereby further enhancing ComPaSS.
Authors: Borui Liao, Yulong Xu, Jiao Ou, Kaiyuan Yang, Weihua Jian, Pengfei Wan, Di Zhang
Abstract: Full-Duplex Speech Dialogue Systems (Full-Duplex SDS) have significantly enhanced the naturalness of human-machine interaction by enabling real-time bidirectional communication. However, existing approaches face challenges such as difficulties in independent module optimization and contextual noise interference due to highly coupled architectural designs and oversimplified binary state modeling. This paper proposes FlexDuo, a flexible full-duplex control module that decouples duplex control from spoken dialogue systems through a plug-and-play architectural design. Furthermore, inspired by human information-filtering mechanisms in conversations, we introduce an explicit Idle state. On one hand, the Idle state filters redundant noise and irrelevant audio to enhance dialogue quality. On the other hand, it establishes a semantic integrity-based buffering mechanism, reducing the risk of mutual interruptions while ensuring accurate response transitions. Experimental results on the Fisher corpus demonstrate that FlexDuo reduces the false interruption rate by 24.9% and improves response accuracy by 7.6% compared to integrated full-duplex dialogue system baselines. It also outperforms voice activity detection (VAD) controlled baseline systems in both Chinese and English dialogue quality. The proposed modular architecture and state-based dialogue model provide a novel technical pathway for building flexible and efficient duplex dialogue systems.
Authors: Chenyu Zhu, Yefeng Liu, Chenyang Lyu, Xue Yang, Guanhua Chen, Longyue Wang, Weihua Luo, Kaifu Zhang
Abstract: Multi-aspect controllable text generation aims to control text generation in attributes from multiple aspects, making it a complex but powerful task in natural language processing. Supervised fine-tuning methods are often employed for this task due to their simplicity and effectiveness. However, they still have some limitations: low rank adaptation (LoRA) only fine-tunes a few parameters and has suboptimal control effects, while full fine-tuning (FFT) requires significant computational resources and is susceptible to overfitting, particularly when data is limited. Moreover, existing works typically train multi-aspect controllable text generation models using only single-aspect annotated data, which results in discrepancies in data distribution; at the same time, accurately generating text with specific attributes is a challenge that requires strong attribute-aware capabilities. To address these limitations, we propose a lightweight, adaptive and attribute-aware framework for multi-aspect controllable text generation. Our framework can dynamically adjust model parameters according to different aspects of data to achieve controllable text generation, aiming to optimize performance across multiple aspects. Experimental results show that our framework outperforms other strong baselines, achieves state-of-the-art performance, adapts well to data discrepancies, and is more accurate in attribute perception.
Authors: Haun Leung, ZiNan Wang
Abstract: It is popular lately to train large language models to be used as chat assistants, but in the conversation between the user and the chat assistant, there are prompts, require multi-turns between the chat assistant and the user. However, there are a number of issues with the multi-turns conversation: The response of the chat assistant is prone to errors and cannot help users achieve their goals; It is difficult for chat assistant to generate responses with different processes based on actual needs for the same command or request; Chat assistant require the use of tools, but the current approach is not elegant and efficient, and the number of tool calls that can be supported is limited. The main reason for these issues is that large language models do not have the thinking ability as a human, lack the reasoning ability and planning ability, and lack the ability to execute plans. To solve these issues, we propose a thinking method based on a built-in chain of thought: In the multi-turns conversation, for each user prompt, the large language model thinks based on elements such as chat history, thinking context, action calls, memory and knowledge, makes detailed reasoning and planning, and actions according to the plan. We also explored how the large language model enhances thinking ability through this thinking method: Collect training datasets according to the thinking method and fine tune the large language model through supervised learning; Train a consistency reward model and use it as a reward function to fine tune the large language model using reinforcement learning, and the reinforced large language model outputs according to this way of thinking. Our experimental results show that the reasoning ability and planning ability of the large language model are enhanced, and the issues in the multi-turns conversation are solved.
Authors: Chen-An Li, Tzu-Han Lin, Yun-Nung Chen, Hung-yi Lee
Abstract: Large vision-language models (LVLMs) perform outstandingly across various multimodal tasks. However, their ability to evaluate generated content remains limited, and training vision-language reward models (VLRMs) with preference data is computationally expensive. This paper explores a training-free alternative by merging text-based reward models (RMs) with LVLMs to create VLRMs. Our approach shows that integrating these models leads to improved performance over LVLMs' scoring and text-based RMs, offering an efficient method for incorporating textual preferences into LVLMs.
Authors: Peiran Wang, Yang Liu, Yunfei Lu, Jue Hong, Ye Wu
Abstract: Large language model (LLM) systems suffer from the models' unstable ability to generate valid and factual content, resulting in hallucination generation. Current hallucination detection methods heavily rely on out-of-model information sources, such as RAG to assist the detection, thus bringing heavy additional latency. Recently, internal states of LLMs' inference have been widely used in numerous research works, such as prompt injection detection, etc. Considering the interpretability of LLM internal states and the fact that they do not require external information sources, we introduce such states into LLM hallucination detection. In this paper, we systematically analyze different internal states' revealing features during inference forward and comprehensively evaluate their ability in hallucination detection. Specifically, we cut the forward process of a large language model into three stages: understanding, query, generation, and extracting the internal state from these stages. By analyzing these states, we provide a deep understanding of why the hallucinated content is generated and what happened in the internal state of the models. Then, we introduce these internal states into hallucination detection and conduct comprehensive experiments to discuss the advantages and limitations.
Authors: Piyawat Lertvittayakumjorn, David Kinney, Vinodkumar Prabhakaran, Donald Martin, Sunipa Dev
Abstract: Generative large language models (LLMs) have been demonstrated to have gaps in diverse, cultural knowledge across the globe. We investigate the effect of retrieval augmented generation and search-grounding techniques on the ability of LLMs to display familiarity with a diverse range of national cultures. Specifically, we compare the performance of standard LLMs, LLMs augmented with retrievals from a bespoke knowledge base (i.e., KB grounding), and LLMs augmented with retrievals from a web search (i.e., search grounding) on a series of cultural familiarity benchmarks. We find that search grounding significantly improves the LLM performance on multiple-choice benchmarks that test propositional knowledge (e.g., the norms, artifacts, and institutions of national cultures), while KB grounding's effectiveness is limited by inadequate knowledge base coverage and a suboptimal retriever. However, search grounding also increases the risk of stereotypical judgments by language models, while failing to improve evaluators' judgments of cultural familiarity in a human evaluation with adequate statistical power. These results highlight the distinction between propositional knowledge about a culture and open-ended cultural fluency when it comes to evaluating the cultural familiarity of generative LLMs.
Authors: Burc Gokden
Abstract: We show that Large Language Model from Power Law Decoder Representations (PLDR-LLM) is a foundational model whose deductive outputs are invariant tensors up to a small perturbation. PLDR-LLM learns a singularity condition for the deductive outputs that enable the once-inferred energy-curvature tensor $\mathbf{G}_{LM}$ to replace the deep neural network of power law graph attention (PLGA) generating the deductive outputs at inference. We demonstrate that a cache for $\mathbf{G}_{LM}$ (G-cache) and KV-cache can be implemented in a straightforward manner to improve the inference time. The invariance and generalizable nature of deductive outputs is at a very high fidelity where deductive outputs have same RMSE and determinant values up to 15 decimal places after caching, and zero-shot benchmark scores remain unchanged. Ablation studies show that learned deductive outputs have distinct loss and accuracy characteristics from models pretrained with transferred, randomly initialized or identity tensors as a constant tensor operator and an LLM with scaled-dot product attention (SDPA) is a special case of PLDR-LLM where $\mathbf{G}_{LM}$ is predefined as identity. The observed invariance characteristic introduces a novel asymmetry between training and inference phases with caching. We outline observed common characteristics of the deductive outputs for the learned singularity condition. We provide an implementation of a training and inference framework for PLDR-LLM with KV-cache and G-cache.
Authors: Shuai Niu, Jing Ma, Hongzhan Lin, Liang Bai, Zhihua Wang, Wei Bi, Yida Xu, Guo Li, Xian Yang
Abstract: Large language models (LLMs) have shown remarkable performance in vision-language tasks, but their application in the medical field remains underexplored, particularly for integrating structured time series data with unstructured clinical notes. In clinical practice, dynamic time series data such as lab test results capture critical temporal patterns, while clinical notes provide rich semantic context. Merging these modalities is challenging due to the inherent differences between continuous signals and discrete text. To bridge this gap, we introduce ProMedTS, a novel self-supervised multimodal framework that employs prompt-guided learning to unify these heterogeneous data types. Our approach leverages lightweight anomaly detection to generate anomaly captions that serve as prompts, guiding the encoding of raw time series data into informative embeddings. These embeddings are aligned with textual representations in a shared latent space, preserving fine-grained temporal nuances alongside semantic insights. Furthermore, our framework incorporates tailored self-supervised objectives to enhance both intra- and inter-modal alignment. We evaluate ProMedTS on disease diagnosis tasks using real-world datasets, and the results demonstrate that our method consistently outperforms state-of-the-art approaches.
Authors: Minlong Peng, Jingyi Yang, Zhongjun He, Hua Wu
Abstract: This work presents a swift method to assess the efficacy of particular types of instruction-tuning data, utilizing just a handful of probe examples and eliminating the need for model retraining. This method employs the idea of gradient-based data influence estimation, analyzing the gradient projections of probe examples from the chosen strategy onto evaluation examples to assess its advantages. Building upon this method, we conducted three swift studies to investigate the potential of Chain-of-thought (CoT) data, query clarification data, and response evaluation data in enhancing model generalization. Subsequently, we embarked on a validation study to corroborate the findings of these swift studies. In this validation study, we developed training datasets tailored to each studied strategy and compared model performance with and without the use of these datasets. The results of the validation study aligned with the findings of the swift studies, validating the efficacy of our proposed method.
Authors: Khalid N. Elmadani, Nizar Habash, Hanada Taha-Thomure
Abstract: This paper introduces the Balanced Arabic Readability Evaluation Corpus BAREC, a large-scale, fine-grained dataset for Arabic readability assessment. BAREC consists of 68,182 sentences spanning 1+ million words, carefully curated to cover 19 readability levels, from kindergarten to postgraduate comprehension. The corpus balances genre diversity, topical coverage, and target audiences, offering a comprehensive resource for evaluating Arabic text complexity. The corpus was fully manually annotated by a large team of annotators. The average pairwise inter-annotator agreement, measured by Quadratic Weighted Kappa, is 81.3%, reflecting a high level of substantial agreement. Beyond presenting the corpus, we benchmark automatic readability assessment across different granularity levels, comparing a range of techniques. Our results highlight the challenges and opportunities in Arabic readability modeling, demonstrating competitive performance across various methods. To support research and education, we will make BAREC openly available, along with detailed annotation guidelines and benchmark results.
Authors: Qingfa Xiao, Jiachuan Wang, Haoyang Li, Cheng Deng, Jiaqi Tang, Shuangyin Li, Yongqi Zhang, Jun Wang, Lei Chen
Abstract: Recent advances in large language models (LLMs) have showcased exceptional performance in long-context tasks, while facing significant inference efficiency challenges with limited GPU memory. Existing solutions first proposed the sliding-window approach to accumulate a set of historical \textbf{key-value} (KV) pairs for reuse, then further improvements selectively retain its subsets at each step. However, due to the sparse attention distribution across a long context, it is hard to identify and recall relevant KV pairs, as the attention is distracted by massive candidate pairs. Additionally, we found it promising to select representative tokens as probe-Query in each sliding window to effectively represent the entire context, which is an approach overlooked by existing methods. Thus, we propose \textbf{ActQKV}, a training-free, \textbf{Act}ivation-aware approach that dynamically determines probe-\textbf{Q}uery and leverages it to retrieve the relevant \textbf{KV} pairs for inference. Specifically, ActQKV monitors a token-level indicator, Activation Bias, within each context window, enabling the proper construction of probe-Query for retrieval at pre-filling stage. To accurately recall the relevant KV pairs and minimize the irrelevant ones, we design a dynamic KV cut-off mechanism guided by information density across layers at the decoding stage. Experiments on the Long-Bench and $\infty$ Benchmarks demonstrate its state-of-the-art performance with competitive inference quality and resource efficiency.
Authors: Peiwen Yuan, Chuyi Tan, Shaoxiong Feng, Yiwei Li, Xinglin Wang, Yueqi Zhang, Jiayi Shi, Boyuan Pan, Yao Hu, Kan Li
Abstract: Despite the rapid progress of large language models (LLMs), their length-controllable text generation (LCTG) ability remains below expectations, posing a major limitation for practical applications. Existing methods mainly focus on end-to-end training to reinforce adherence to length constraints. However, the lack of decomposition and targeted enhancement of LCTG sub-abilities restricts further progress.To bridge this gap, we conduct a bottom-up decomposition of LCTG sub-abilities with human patterns as reference and perform a detailed error analysis.On this basis, we propose MarkerGen, a simple-yet-effective plug-and-play approach that:(1) mitigates LLM fundamental deficiencies via external tool integration;(2) conducts explicit length modeling with dynamically inserted markers;(3) employs a three-stage generation scheme to better align length constraints while maintaining content quality.Comprehensive experiments demonstrate that MarkerGen significantly improves LCTG across various settings, exhibiting outstanding effectiveness and generalizability.
Authors: Milena de Swart, Floris den Hengst, Jieying Chen
Abstract: This paper addresses the critical need for detecting bias in government documents, an underexplored area with significant implications for governance. Existing methodologies often overlook the unique context and far-reaching impacts of governmental documents, potentially obscuring embedded biases that shape public policy and citizen-government interactions. To bridge this gap, we introduce the Dutch Government Data for Bias Detection (DGDB), a dataset sourced from the Dutch House of Representatives and annotated for bias by experts. We fine-tune several BERT-based models on this dataset and compare their performance with that of generative language models. Additionally, we conduct a comprehensive error analysis that includes explanations of the models' predictions. Our findings demonstrate that fine-tuned models achieve strong performance and significantly outperform generative language models, indicating the effectiveness of DGDB for bias detection. This work underscores the importance of labeled datasets for bias detection in various languages and contributes to more equitable governance practices.
Authors: Mingqian He, Yongliang Shen, Wenqi Zhang, Qiuying Peng, Jun Wang, Weiming Lu
Abstract: Generating step-by-step "chain-of-thought" rationales has proven effective for improving the performance of large language models on complex reasoning tasks. However, applying such techniques to structured tasks, such as text-to-SQL, remains largely unexplored. In this paper, we introduce Self-Taught Reasoner for text-to-SQL (STaR-SQL), a novel approach that reframes SQL query generation as a reasoning-driven process. Our method prompts the LLM to produce detailed reasoning steps for SQL queries and fine-tunes it on rationales that lead to correct outcomes. Unlike traditional methods, STaR-SQL dedicates additional test-time computation to reasoning, thereby positioning LLMs as spontaneous reasoners rather than mere prompt-based agents. To further scale the inference process, we incorporate an outcome-supervised reward model (ORM) as a verifier, which enhances SQL query accuracy. Experimental results on the challenging Spider benchmark demonstrate that STaR-SQL significantly improves text-to-SQL performance, achieving an execution accuracy of 86.6%. This surpasses a few-shot baseline by 31.6% and a baseline fine-tuned to predict answers directly by 18.0%. Additionally, STaR-SQL outperforms agent-like prompting methods that leverage more powerful yet closed-source models such as GPT-4. These findings underscore the potential of reasoning-augmented training for structured tasks and open the door to extending self-improving reasoning models to text-to-SQL generation and beyond.
Authors: Guangwei Li, Yuansen Zhang, Yinggui Wang, Shoumeng Yan, Lei Wang, Tao Wei
Abstract: The rapid development of large language models (LLMs) is redefining the landscape of human-computer interaction, and their integration into various user-service applications is becoming increasingly prevalent. However, transmitting user data to cloud-based LLMs presents significant risks of data breaches and unauthorized access to personal identification information. In this paper, we propose a privacy preservation pipeline for protecting privacy and sensitive information during interactions between users and LLMs in practical LLM usage scenarios. We construct SensitiveQA, the first privacy open-ended question-answering dataset. It comprises 57k interactions in Chinese and English, encompassing a diverse range of user-sensitive information within the conversations. Our proposed solution employs a multi-stage strategy aimed at preemptively securing user information while simultaneously preserving the response quality of cloud-based LLMs. Experimental validation underscores our method's efficacy in balancing privacy protection with maintaining robust interaction quality. The code and dataset are available at https://github.com/ligw1998/PRIV-QA.
Authors: Joonatan Laato, Jenna Kanerva, John Loehr, Virpi Lummaa, Filip Ginter
Abstract: We performed a zero-shot information extraction study on a historical collection of 89,339 brief Finnish-language interviews of refugee families relocated post-WWII from Finnish Eastern Karelia. Our research objective is two-fold. First, we aim to extract social organizations and hobbies from the free text of the interviews, separately for each family member. These can act as a proxy variable indicating the degree of social integration of refugees in their new environment. Second, we aim to evaluate several alternative ways to approach this task, comparing a number of generative models and a supervised learning approach, to gain a broader insight into the relative merits of these different approaches and their applicability in similar studies. We find that the best generative model (GPT-4) is roughly on par with human performance, at an F-score of 88.8%. Interestingly, the best open generative model (Llama-3-70B-Instruct) reaches almost the same performance, at 87.7% F-score, demonstrating that open models are becoming a viable alternative for some practical tasks even on non-English data. Additionally, we test a supervised learning alternative, where we fine-tune a Finnish BERT model (FinBERT) using GPT-4 generated training data. By this method, we achieved an F-score of 84.1% already with 6K interviews up to an F-score of 86.3% with 30k interviews. Such an approach would be particularly appealing in cases where the computational resources are limited, or there is a substantial mass of data to process.
Authors: Nicol\`o Penzo, Marco Guerini, Bruno Lepri, Goran Glava\v{s}, Sara Tonelli
Abstract: Multi-Party Conversations (MPCs) are widely studied across disciplines, with social media as a primary data source due to their accessibility. However, these datasets raise privacy concerns and often reflect platform-specific properties. For example, interactions between speakers may be limited due to rigid platform structures (e.g., threads, tree-like discussions), which yield overly simplistic interaction patterns (e.g., as a consequence of ``reply-to'' links). This work explores the feasibility of generating diverse MPCs with instruction-tuned Large Language Models (LLMs) by providing deterministic constraints such as dialogue structure and participants' stance. We investigate two complementary strategies of leveraging LLMs in this context: (i.) LLMs as MPC generators, where we task the LLM to generate a whole MPC at once and (ii.) LLMs as MPC parties, where the LLM generates one turn of the conversation at a time, provided the conversation history. We next introduce an analytical framework to evaluate compliance with the constraints, content quality, and interaction complexity for both strategies. Finally, we assess the quality of obtained MPCs via human annotation and LLM-as-a-judge evaluations. We find stark differences among LLMs, with only some being able to generate high-quality MPCs. We also find that turn-by-turn generation yields better conformance to constraints and higher linguistic variability than generating MPCs in one pass. Nonetheless, our structural and qualitative evaluation indicates that both generation strategies can yield high-quality MPCs.
Authors: Kenneth Enevoldsen, Isaac Chung, Imene Kerboua, M\'arton Kardos, Ashwin Mathur, David Stap, Jay Gala, Wissam Siblini, Dominik Krzemi\'nski, Genta Indra Winata, Saba Sturua, Saiteja Utpala, Mathieu Ciancone, Marion Schaeffer, Gabriel Sequeira, Diganta Misra, Shreeya Dhakal, Jonathan Rystr{\o}m, Roman Solomatin, \"Omer \c{C}a\u{g}atan, Akash Kundu, Martin Bernstorff, Shitao Xiao, Akshita Sukhlecha, Bhavish Pahwa, Rafa{\l} Po\'swiata, Kranthi Kiran GV, Shawon Ashraf, Daniel Auras, Bj\"orn Pl\"uster, Jan Philipp Harries, Lo\"ic Magne, Isabelle Mohr, Mariya Hendriksen, Dawei Zhu, Hippolyte Gisserot-Boukhlef, Tom Aarsen, Jan Kostkan, Konrad Wojtasik, Taemin Lee, Marek \v{S}uppa, Crystina Zhang, Roberta Rocca, Mohammed Hamdy, Andrianos Michail, John Yang, Manuel Faysse, Aleksei Vatolin, Nandan Thakur, Manan Dey, Dipam Vasani, Pranjal Chitale, Simone Tedeschi, Nguyen Tai, Artem Snegirev, Michael G\"unther, Mengzhou Xia, Weijia Shi, Xing Han L\`u, Jordan Clive, Gayatri Krishnakumar, Anna Maksimova, Silvan Wehrli, Maria Tikhonova, Henil Panchal, Aleksandr Abramov, Malte Ostendorff, Zheng Liu, Simon Clematide, Lester James Miranda, Alena Fenogenova, Guangyu Song, Ruqiya Bin Safi, Wen-Ding Li, Alessia Borghini, Federico Cassano, Hongjin Su, Jimmy Lin, Howard Yen, Lasse Hansen, Sara Hooker, Chenghao Xiao, Vaibhav Adlakha, Orion Weller, Siva Reddy, Niklas Muennighoff
Abstract: Text embeddings are typically evaluated on a limited set of tasks, which are constrained by language, domain, and task diversity. To address these limitations and provide a more comprehensive evaluation, we introduce the Massive Multilingual Text Embedding Benchmark (MMTEB) - a large-scale, community-driven expansion of MTEB, covering over 500 quality-controlled evaluation tasks across 250+ languages. MMTEB includes a diverse set of challenging, novel tasks such as instruction following, long-document retrieval, and code retrieval, representing the largest multilingual collection of evaluation tasks for embedding models to date. Using this collection, we develop several highly multilingual benchmarks, which we use to evaluate a representative set of models. We find that while large language models (LLMs) with billions of parameters can achieve state-of-the-art performance on certain language subsets and task categories, the best-performing publicly available model is multilingual-e5-large-instruct with only 560 million parameters. To facilitate accessibility and reduce computational cost, we introduce a novel downsampling method based on inter-task correlation, ensuring a diverse selection while preserving relative model rankings. Furthermore, we optimize tasks such as retrieval by sampling hard negatives, creating smaller but effective splits. These optimizations allow us to introduce benchmarks that drastically reduce computational demands. For instance, our newly introduced zero-shot English benchmark maintains a ranking order similar to the full-scale version but at a fraction of the computational cost.
Authors: Dario Garcia-Gasulla, Anna Arias-Duart, Adrian Tormos, Daniel Hinjos, Oscar Molina-Sedano, Ashwin Kumar Gururajan, Maria Eugenia Cardello
Abstract: Direct Preference Optimization (DPO) is an efficient alignment technique that steers LLMs towards preferable outputs by training on preference data, bypassing the need for explicit reward models. Its simplicity enables easy adaptation to various domains and safety requirements. This paper examines DPO's effectiveness in model safety against jailbreaking attacks while minimizing data requirements and training costs. We introduce Egida, a dataset expanded from multiple sources, which includes 27 different safety topics and 18 different attack styles, complemented with synthetic and human labels. This data is used to boost the safety of state-of-the-art LLMs (Llama-3.1-8B/70B-Instruct, Qwen-2.5-7B/72B-Instruct) across topics and attack styles. In addition to safety evaluations, we assess their post-alignment performance degradation in general purpose tasks, and their tendency to over refusal. Following the proposed methodology, trained models reduce their Attack Success Rate by 10%-30%, using small training efforts (2,000 samples) with low computational cost (3\$ for 8B models, 20\$ for 72B models). Safety aligned models generalize to unseen topics and attack styles, with the most successful attack style reaching a success rate around 5%. Size and family are found to strongly influence model malleability towards safety, pointing at the importance of pre-training choices. To validate our findings, a large independent assessment of human preference agreement with Llama-Guard-3-8B is conducted by the authors and the associated dataset Egida-HSafe is released. Overall, this study illustrates how affordable and accessible it is to enhance LLM safety using DPO while outlining its current limitations. All datasets and models are released to enable reproducibility and further research.
Authors: Naibin Gu, Zhenyu Zhang, Xiyu Liu, Peng Fu, Zheng Lin, Shuohuan Wang, Yu Sun, Hua Wu, Weiping Wang, Haifeng Wang
Abstract: Due to the demand for efficient fine-tuning of large language models, Low-Rank Adaptation (LoRA) has been widely adopted as one of the most effective parameter-efficient fine-tuning methods. Nevertheless, while LoRA improves efficiency, there remains room for improvement in accuracy. Herein, we adopt a novel perspective to assess the characteristics of LoRA ranks. The results reveal that different ranks within the LoRA modules not only exhibit varying levels of importance but also evolve dynamically throughout the fine-tuning process, which may limit the performance of LoRA. Based on these findings, we propose BeamLoRA, which conceptualizes each LoRA module as a beam where each rank naturally corresponds to a potential sub-solution, and the fine-tuning process becomes a search for the optimal sub-solution combination. BeamLoRA dynamically eliminates underperforming sub-solutions while expanding the parameter space for promising ones, enhancing performance with a fixed rank. Extensive experiments across three base models and 12 datasets spanning math reasoning, code generation, and commonsense reasoning demonstrate that BeamLoRA consistently enhances the performance of LoRA, surpassing the other baseline methods.
Authors: Guilherme Sousa, Rinaldo Lima, Cassia Trojahn
Abstract: Ontology, and more broadly, Knowledge Graph Matching is a challenging task in which expressiveness has not been fully addressed. Despite the increasing use of embeddings and language models for this task, approaches for generating expressive correspondences still do not take full advantage of these models, in particular, large language models (LLMs). This paper proposes to integrate LLMs into an approach for generating expressive correspondences based on alignment need and ABox-based relation discovery. The generation of correspondences is performed by matching similar surroundings of instance sub-graphs. The integration of LLMs results in different architectural modifications, including label similarity, sub-graph matching, and entity matching. The performance word embeddings, sentence embeddings, and LLM-based embeddings, was compared. The results demonstrate that integrating LLMs surpasses all other models, enhancing the baseline version of the approach with a 45\% increase in F-measure.
Authors: DongGeon Lee, Hwanjo Yu
Abstract: Hallucinations in large language model (LLM) outputs severely limit their reliability in knowledge-intensive tasks such as question answering. To address this challenge, we introduce REFIND (Retrieval-augmented Factuality hallucINation Detection), a novel framework that detects hallucinated spans within LLM outputs by directly leveraging retrieved documents. As part of the REFIND, we propose the Context Sensitivity Ratio (CSR), a novel metric that quantifies the sensitivity of LLM outputs to retrieved evidence. This innovative approach enables REFIND to efficiently and accurately detect hallucinations, setting it apart from existing methods. In the evaluation, REFIND demonstrated robustness across nine languages, including low-resource settings, and significantly outperformed baseline models, achieving superior IoU scores in identifying hallucinated spans. This work highlights the effectiveness of quantifying context sensitivity for hallucination detection, thereby paving the way for more reliable and trustworthy LLM applications across diverse languages.
Authors: Darpan Aswal, Manjira Sinha
Abstract: Transformer-based models dominate NLP tasks like sentiment analysis, machine translation, and claim verification. However, their massive computational demands and lack of interpretability pose challenges for real-world applications requiring efficiency and transparency. In this work, we explore Graph Neural Networks (GNNs) and Hyperbolic Graph Neural Networks (HGNNs) as lightweight yet effective alternatives for Environmental Claim Detection, reframing it as a graph classification problem. We construct dependency parsing graphs to explicitly model syntactic structures, using simple word embeddings (word2vec) for node features with dependency relations encoded as edge features. Our results demonstrate that these graph-based models achieve comparable or superior performance to state-of-the-art transformers while using 30x fewer parameters. This efficiency highlights the potential of structured, interpretable, and computationally efficient graph-based approaches.
Authors: Maiya Goloburda, Nurkhan Laiyk, Diana Turmakhan, Yuxia Wang, Mukhammed Togmanov, Jonibek Mansurov, Askhat Sametov, Nurdaulet Mukhituly, Minghan Wang, Daniil Orel, Zain Muhammad Mujahid, Fajri Koto, Timothy Baldwin, Preslav Nakov
Abstract: Large language models (LLMs) are known to have the potential to generate harmful content, posing risks to users. While significant progress has been made in developing taxonomies for LLM risks and safety evaluation prompts, most studies have focused on monolingual contexts, primarily in English. However, language- and region-specific risks in bilingual contexts are often overlooked, and core findings can diverge from those in monolingual settings. In this paper, we introduce Qorgau, a novel dataset specifically designed for safety evaluation in Kazakh and Russian, reflecting the unique bilingual context in Kazakhstan, where both Kazakh (a low-resource language) and Russian (a high-resource language) are spoken. Experiments with both multilingual and language-specific LLMs reveal notable differences in safety performance, emphasizing the need for tailored, region-specific datasets to ensure the responsible and safe deployment of LLMs in countries like Kazakhstan. Warning: this paper contains example data that may be offensive, harmful, or biased.
Authors: Ori Shapira, Shlomo E. Chazan, Amir DN Cohen
Abstract: With the increasing prevalence of recorded human speech, spoken language understanding (SLU) is essential for its efficient processing. In order to process the speech, it is commonly transcribed using automatic speech recognition technology. This speech-to-text transition introduces errors into the transcripts, which subsequently propagate to downstream NLP tasks, such as dialogue summarization. While it is known that transcript noise affects downstream tasks, a systematic approach to analyzing its effects across different noise severities and types has not been addressed. We propose a configurable framework for assessing task models in diverse noisy settings, and for examining the impact of transcript-cleaning techniques. The framework facilitates the investigation of task model behavior, which can in turn support the development of effective SLU solutions. We exemplify the utility of our framework on three SLU tasks and four task models, offering insights regarding the effect of transcript noise on tasks in general and models in particular. For instance, we find that task models can tolerate a certain level of noise, and are affected differently by the types of errors in the transcript.
Authors: Qi Zhang, Zhiqing Xiao, Ruixuan Xiao, Lirong Gao, Junbo Zhao
Abstract: In-context learning (ICL) has demonstrated significant potential in enhancing the capabilities of large language models (LLMs) during inference. It's well-established that ICL heavily relies on selecting effective demonstrations to generate outputs that better align with the expected results. As for demonstration selection, previous approaches have typically relied on intuitive metrics to evaluate the effectiveness of demonstrations, which often results in limited robustness and poor cross-model generalization capabilities. To tackle these challenges, we propose a novel method, \textbf{D}emonstration \textbf{VA}lidation (\textbf{D.Va}), which integrates a demonstration validation perspective into this field. By introducing the demonstration validation mechanism, our method effectively identifies demonstrations that are both effective and highly generalizable. \textbf{D.Va} surpasses all existing demonstration selection techniques across both natural language understanding (NLU) and natural language generation (NLG) tasks. Additionally, we demonstrate the robustness and generalizability of our approach across various language models with different retrieval models.
Authors: Nurkhan Laiyk, Daniil Orel, Rituraj Joshi, Maiya Goloburda, Yuxia Wang, Preslav Nakov, Fajri Koto
Abstract: Instruction tuning in low-resource languages remains underexplored due to limited text data, particularly in government and cultural domains. To address this, we introduce and open-source a large-scale (10,600 samples) instruction-following (IFT) dataset, covering key institutional and cultural knowledge relevant to Kazakhstan. Our dataset enhances LLMs' understanding of procedural, legal, and structural governance topics. We employ LLM-assisted data generation, comparing open-weight and closed-weight models for dataset construction, and select GPT-4o as the backbone. Each entity of our dataset undergoes full manual verification to ensure high quality. We also show that fine-tuning Qwen, Falcon, and Gemma on our dataset leads to consistent performance improvements in both multiple-choice and generative tasks, demonstrating the potential of LLM-assisted instruction tuning for low-resource languages.
Authors: Youna Kim, Minjoon Choi, Sungmin Cho, Hyuhng Joon Kim, Sang-goo Lee, Taeuk Kim
Abstract: Large Language Models (LLMs) enhance their problem-solving capability by leveraging both parametric and external knowledge. Beyond leveraging external knowledge to improve response accuracy, they require key capabilities for reliable knowledge-handling: resolving conflicts between knowledge sources, avoiding distraction from uninformative external knowledge, and abstaining when sufficient knowledge is unavailable. Prior studies have examined these scenarios in isolation or with limited scope. To systematically evaluate these capabilities, we introduce a comprehensive framework for analyzing knowledge-handling based on two key dimensions: the presence of parametric knowledge and the informativeness of external knowledge. Through analysis, we identify biases in knowledge utilization and examine how the ability to handle one scenario impacts performance in others. Furthermore, we demonstrate that training on data constructed based on the knowledge-handling scenarios improves LLMs' reliability in integrating and utilizing knowledge.
Authors: Feiye Huo, Jianchao Tan, Kefeng Zhang, Xunliang Cai, Shengli Sun
Abstract: The growing scale of Large Language Models (LLMs) has exacerbated inference latency and computational costs. Speculative decoding methods, which aim to mitigate these issues, often face inefficiencies in the construction of token trees and the verification of candidate tokens. Existing strategies, including chain mode, static tree, and dynamic tree approaches, have limitations in accurately preparing candidate token trees for verification. We propose a novel method named C2T that adopts a lightweight classifier to generate and prune token trees dynamically. Our classifier considers additional feature variables beyond the commonly used joint probability to predict the confidence score for each draft token to determine whether it is the candidate token for verification. This method outperforms state-of-the-art (SOTA) methods such as EAGLE-2 on multiple benchmarks, by reducing the total number of candidate tokens by 25% while maintaining or even improving the acceptance length.
Authors: Liyang He, Chenglong Liu, Rui Li, Zhenya Huang, Shulan Ruan, Jun Zhou, Enhong Chen
Abstract: Sentence embedding is essential for many NLP tasks, with contrastive learning methods achieving strong performance using annotated datasets like NLI. Yet, the reliance on manual labels limits scalability. Recent studies leverage large language models (LLMs) to generate sentence pairs, reducing annotation dependency. However, they overlook ranking information crucial for fine-grained semantic distinctions. To tackle this challenge, we propose a method for controlling the generation direction of LLMs in the latent space. Unlike unconstrained generation, the controlled approach ensures meaningful semantic divergence. Then, we refine exist sentence embedding model by integrating ranking information and semantic information. Experiments on multiple benchmarks demonstrate that our method achieves new SOTA performance with a modest cost in ranking sentence synthesis.
Authors: Tim Baumg\"artner, Ted Briscoe, Iryna Gurevych
Abstract: We present PeerQA, a real-world, scientific, document-level Question Answering (QA) dataset. PeerQA questions have been sourced from peer reviews, which contain questions that reviewers raised while thoroughly examining the scientific article. Answers have been annotated by the original authors of each paper. The dataset contains 579 QA pairs from 208 academic articles, with a majority from ML and NLP, as well as a subset of other scientific communities like Geoscience and Public Health. PeerQA supports three critical tasks for developing practical QA systems: Evidence retrieval, unanswerable question classification, and answer generation. We provide a detailed analysis of the collected dataset and conduct experiments establishing baseline systems for all three tasks. Our experiments and analyses reveal the need for decontextualization in document-level retrieval, where we find that even simple decontextualization approaches consistently improve retrieval performance across architectures. On answer generation, PeerQA serves as a challenging benchmark for long-context modeling, as the papers have an average size of 12k tokens. Our code and data is available at https://github.com/UKPLab/peerqa.
Authors: Song Duong, Florian Le Bronnec, Alexandre Allauzen, Vincent Guigue, Alberto Lumbreras, Laure Soulier, Patrick Gallinari
Abstract: Large Language Models (LLMs), when used for conditional text generation, often produce hallucinations, i.e., information that is unfaithful or not grounded in the input context. This issue arises in typical conditional text generation tasks, such as text summarization and data-to-text generation, where the goal is to produce fluent text based on contextual input. When fine-tuned on specific domains, LLMs struggle to provide faithful answers to a given context, often adding information or generating errors. One underlying cause of this issue is that LLMs rely on statistical patterns learned from their training data. This reliance can interfere with the model's ability to stay faithful to a provided context, leading to the generation of ungrounded information. We build upon this observation and introduce a novel self-supervised method for generating a training set of unfaithful samples. We then refine the model using a training process that encourages the generation of grounded outputs over unfaithful ones, drawing on preference-based training. Our approach leads to significantly more grounded text generation, outperforming existing self-supervised techniques in faithfulness, as evaluated through automatic metrics, LLM-based assessments, and human evaluations.
Authors: Jusen Du, Weigao Sun, Disen Lan, Jiaxi Hu, Yu Cheng
Abstract: Linear sequence modeling methods, such as linear attention, state space modeling, and linear RNNs, offer significant efficiency improvements by reducing the complexity of training and inference. However, these methods typically compress the entire input sequence into a single fixed-size memory state, which leads to suboptimal performance on recall-intensive downstream tasks. Drawing inspiration from neuroscience, particularly the brain's ability to maintain robust long-term memory while mitigating "memory interference", we introduce a novel architecture called Mixture-of-Memories (MoM). MoM utilizes multiple independent memory states, with a router network directing input tokens to specific memory states. This approach greatly enhances the overall memory capacity while minimizing memory interference. As a result, MoM performs exceptionally well on recall-intensive tasks, surpassing existing linear sequence modeling techniques. Despite incorporating multiple memory states, the computation of each memory state remains linear in complexity, allowing MoM to retain the linear-complexity advantage during training, while constant-complexity during inference. Our experimental results show that MoM significantly outperforms current linear sequence models on downstream language tasks, particularly recall-intensive tasks, and even achieves performance comparable to Transformer models. The code is released at https://github.com/OpenSparseLLMs/MoM and is also released as a part of https://github.com/OpenSparseLLMs/Linear-MoE.
URLs: https://github.com/OpenSparseLLMs/MoM, https://github.com/OpenSparseLLMs/Linear-MoE.
Authors: Tristan Karch, Luca Engel, Philippe Schwaller, Fr\'ed\'eric Kaplan
Abstract: As large language models (LLMs) converge towards similar capabilities, the key to advancing their performance lies in identifying and incorporating valuable new information sources. However, evaluating which text collections are worth the substantial investment required for digitization, preprocessing, and integration into LLM systems remains a significant challenge. We present a novel approach to this challenge: an automated pipeline that evaluates the potential information gain from text collections without requiring model training or fine-tuning. Our method generates multiple choice questions (MCQs) from texts and measures an LLM's performance both with and without access to the source material. The performance gap between these conditions serves as a proxy for the collection's information potential. We validate our approach using three strategically selected datasets: EPFL PhD manuscripts (likely containing novel specialized knowledge), Wikipedia articles (presumably part of training data), and a synthetic baseline dataset. Our results demonstrate that this method effectively identifies collections containing valuable novel information, providing a practical tool for prioritizing data acquisition and integration efforts.
Authors: Chengyan Wu, Bolei Ma, Ningyuan Deng, Yanqing He, Yun Xue
Abstract: Aspect-based sentiment analysis (ABSA) is a sequence labeling task that has garnered growing research interest in multilingual contexts. However, recent studies lack more robust feature alignment and finer aspect-level alignment. In this paper, we propose a novel framework, Multi-Scale and Multi-Objective optimization (MSMO) for cross-lingual ABSA. During multi-scale alignment, we achieve cross-lingual sentence-level and aspect-level alignment, aligning features of aspect terms in different contextual environments. Specifically, we introduce code-switched bilingual sentences into the language discriminator and consistency training modules to enhance the model's robustness. During multi-objective optimization, we design two optimization objectives: supervised training and consistency training, aiming to enhance cross-lingual semantic alignment. To further improve model performance, we incorporate distilled knowledge of the target language into the model. Results show that MSMO significantly enhances cross-lingual ABSA by achieving state-of-the-art performance across multiple languages and models.
Authors: Hongbo Zhang, Han Cui, Guangsheng Bao, Linyi Yang, Jun Wang, Yue Zhang
Abstract: We introduce Direct Value Optimization (DVO), an innovative reinforcement learning framework for enhancing large language models in complex reasoning tasks. Unlike traditional methods relying on preference labels, DVO utilizes value signals at individual reasoning steps, optimizing models via a mean squared error loss. The key benefit of DVO lies in its fine-grained supervision, circumventing the need for labor-intensive human annotations. Target values within the DVO are estimated using either Monte Carlo Tree Search or an outcome value model. Our empirical analysis on both mathematical and commonsense reasoning tasks shows that DVO consistently outperforms existing offline preference optimization techniques, even with fewer training steps. These findings underscore the importance of value signals in advancing reasoning capabilities and highlight DVO as a superior methodology under scenarios lacking explicit human preference information.
Authors: Juyuan Zhang, Wei Zhu, Jiechao Gao
Abstract: Time series modeling holds significant importance in many real-world applications and has been extensively studied. While pre-trained foundation models have made impressive strides in the fields of natural language processing (NLP) and computer vision (CV), their development in time series domains has been constrained by data sparsity. A series of recent studies have demonstrated that large language models (LLMs) possess robust pattern recognition and reasoning abilities over complex sequences of tokens. However, the current literature have yet striked a high-quality balance between (a) effectively aligning the time series and natural language modalities, and (b) keeping the inference efficiency. To address the above issues, we now propose the Time-LlaMA framework. Time-LlaMA first converts the time series input into token embeddings through a linear tokenization mechanism. Second, the time series token embeddings are aligned with the text prompts. Third, to further adapt the LLM backbone for time series modeling, we have developed a dynamic low-rank adaptation technique (D-LoRA). D-LoRA dynamically chooses the most suitable LoRA modules at each layer of the Transformer backbone for each time series input, enhancing the model's predictive capabilities. Our experimental results on an extensive collection of challenging real-world time series tasks confirm that our proposed method achieves the state-of-the-art (SOTA) performance.
Authors: Keqin Peng, Liang Ding, Yuanxin Ouyang, Meng Fang, Yancheng Yuan, Dacheng Tao
Abstract: Large language models (LLMs) excel at a range of tasks through in-context learning (ICL), where only a few task examples guide their predictions. However, prior research highlights that LLMs often overlook input-label mapping information in ICL, relying more on their pre-trained knowledge. To address this issue, we introduce In-Context Contrastive Decoding (ICCD), a novel method that emphasizes input-label mapping by contrasting the output distributions between positive and negative in-context examples. Experiments on 7 natural language understanding (NLU) tasks show that our ICCD method brings consistent and significant improvement (up to +2.1 improvement on average) upon 6 different scales of LLMs without requiring additional training. Our approach is versatile, enhancing performance with various demonstration selection methods, demonstrating its broad applicability and effectiveness. The code and scripts will be publicly released.
Authors: Renxi Wang, Honglin Mu, Liqun Ma, Lizhi Lin, Yunlong Feng, Timothy Baldwin, Xudong Han, Haonan Li
Abstract: Evaluating large language models' (LLMs) long-context understanding capabilities remains challenging. We present SCALAR (Scientific Citation-based Live Assessment of Long-context Academic Reasoning), a novel benchmark that leverages academic papers and their citation networks. SCALAR features automatic generation of high-quality ground truth labels without human annotation, controllable difficulty levels, and a dynamic updating mechanism that prevents data contamination. Using ICLR 2025 papers, we evaluate 8 state-of-the-art LLMs, revealing key insights about their capabilities and limitations in processing long scientific documents across different context lengths and reasoning types. Our benchmark provides a reliable and sustainable way to track progress in long-context understanding as LLM capabilities evolve.
Authors: Florian Schneider, Carolin Holtermann, Chris Biemann, Anne Lauscher
Abstract: Large Vision-Language Models (LVLMs) have recently gained attention due to their distinctive performance and broad applicability. While it has been previously shown that their efficacy in usage scenarios involving non-Western contexts falls short, existing studies are limited in scope, covering just a narrow range of cultures, focusing exclusively on a small number of cultural aspects, or evaluating a limited selection of models on a single task only. Towards globally inclusive LVLM research, we introduce GIMMICK, an extensive multimodal benchmark designed to assess a broad spectrum of cultural knowledge across 144 countries representing six global macro-regions. GIMMICK comprises six tasks built upon three new datasets that span 728 unique cultural events or facets on which we evaluated 20 LVLMs and 11 LLMs, including five proprietary and 26 open-weight models of all sizes. We systematically examine (1) regional cultural biases, (2) the influence of model size, (3) input modalities, and (4) external cues. Our analyses reveal strong biases toward Western cultures across models and tasks and highlight strong correlations between model size and performance, as well as the effectiveness of multimodal input and external geographic cues. We further find that models have more knowledge of tangible than intangible aspects (e.g., food vs. rituals) and that they excel in recognizing broad cultural origins but struggle with a more nuanced understanding.
Authors: Anudeex Shetty, Amin Beheshti, Mark Dras, Usman Naseem
Abstract: Alignment techniques have become central to ensuring that Large Language Models (LLMs) generate outputs consistent with human values. However, existing alignment paradigms often model an averaged or monolithic preference, failing to account for the diversity of perspectives across cultures, demographics, and communities. This limitation is particularly critical in health-related scenarios, where plurality is essential due to the influence of culture, religion, personal values, and conflicting opinions. Despite progress in pluralistic alignment, no prior work has focused on health, likely due to the unavailability of publicly available datasets. To address this gap, we introduce VITAL, a new benchmark dataset comprising 13.1K value-laden situations and 5.4K multiple-choice questions focused on health, designed to assess and benchmark pluralistic alignment methodologies. Through extensive evaluation of eight LLMs of varying sizes, we demonstrate that existing pluralistic alignment techniques fall short in effectively accommodating diverse healthcare beliefs, underscoring the need for tailored AI alignment in specific domains. This work highlights the limitations of current approaches and lays the groundwork for developing health-specific alignment solutions.
Authors: Alex Duchnowski, Ellie Pavlick, Alexander Koller
Abstract: We introduce the dataset of Everyday Hard Optimization Problems (EHOP), a collection of NP-hard optimization problems expressed in natural language. EHOP includes problem formulations that could be found in computer science textbooks, versions that are dressed up as problems that could arise in real life, and variants of well-known problems with inverted rules. We find that state-of-the-art LLMs, across multiple prompting strategies, systematically solve textbook problems more accurately than their real-life and inverted counterparts. We argue that this constitutes evidence that LLMs adapt solutions seen during training, rather than leveraging reasoning abilities that would enable them to generalize to novel problems.
Authors: Beatrice Savoldi, Alan Ramponi, Matteo Negri, Luisa Bentivogli
Abstract: Converging societal and technical factors have transformed language technologies into user-facing applications employed across languages. Machine Translation (MT) has become a global tool, with cross-lingual services now also supported by dialogue systems powered by multilingual Large Language Models (LLMs). This accessibility has expanded MT's reach to a vast base of lay users, often with little to no expertise in the languages or the technology itself. Despite this, the understanding of MT consumed by this diverse group of users -- their needs, experiences, and interactions with these systems -- remains limited. This paper traces the shift in MT user profiles, focusing on non-expert users and how their engagement with these systems may change with LLMs. We identify three key factors -- usability, trust, and literacy -- that shape these interactions and must be addressed to align MT with user needs. By exploring these dimensions, we offer insights to guide future MT with a user-centered approach.
Authors: Nathana\"el Carraz Rakotonirina, Mohammed Hamdy, Jon Ander Campos, Lucas Weber, Alberto Testoni, Marzieh Fadaee, Sandro Pezzelle, Marco Del Tredici
Abstract: Large Language Models (LLMs) are increasingly used in working environments for a wide range of tasks, excelling at solving individual problems in isolation. However, are they also able to effectively collaborate over long-term interactions? To investigate this, we introduce MemoryCode, a synthetic multi-session dataset designed to test LLMs' ability to track and execute simple coding instructions amid irrelevant information, simulating a realistic setting. While all the models we tested handle isolated instructions well, even the performance of state-of-the-art models like GPT-4o deteriorates when instructions are spread across sessions. Our analysis suggests this is due to their failure to retrieve and integrate information over long instruction chains. Our results highlight a fundamental limitation of current LLMs, restricting their ability to collaborate effectively in long interactions.
Authors: Yilong Chen, Junyuan Shang, Zhenyu Zhang, Yanxi Xie, Jiawei Sheng, Tingwen Liu, Shuohuan Wang, Yu Sun, Hua Wu, Haifeng Wang
Abstract: Large language models (LLMs) face inherent performance bottlenecks under parameter constraints, particularly in processing critical tokens that demand complex reasoning. Empirical analysis reveals challenging tokens induce abrupt gradient spikes across layers, exposing architectural stress points in standard Transformers. Building on this insight, we propose Inner Thinking Transformer (ITT), which reimagines layer computations as implicit thinking steps. ITT dynamically allocates computation through Adaptive Token Routing, iteratively refines representations via Residual Thinking Connections, and distinguishes reasoning phases using Thinking Step Encoding. ITT enables deeper processing of critical tokens without parameter expansion. Evaluations across 162M-466M parameter models show ITT achieves 96.5\% performance of a 466M Transformer using only 162M parameters, reduces training data by 43.2\%, and outperforms Transformer/Loop variants in 11 benchmarks. By enabling elastic computation allocation during inference, ITT balances performance and efficiency through architecture-aware optimization of implicit thinking pathways.
Authors: Feiyuan Zhang, Dezhi Zhu, James Ming, Yilun Jin, Di Chai, Liu Yang, Han Tian, Zhaoxin Fan, Kai Chen
Abstract: Retrieval-Augmented Generation (RAG) systems have shown substantial benefits in applications such as question answering and multi-turn dialogue \citep{lewis2020retrieval}. However, traditional RAG methods, while leveraging static knowledge bases, often overlook the potential of dynamic historical information in ongoing conversations. To bridge this gap, we introduce DH-RAG, a Dynamic Historical Context-Powered Retrieval-Augmented Generation Method for Multi-Turn Dialogue. DH-RAG is inspired by human cognitive processes that utilize both long-term memory and immediate historical context in conversational responses \citep{stafford1987conversational}. DH-RAG is structured around two principal components: a History-Learning based Query Reconstruction Module, designed to generate effective queries by synthesizing current and prior interactions, and a Dynamic History Information Updating Module, which continually refreshes historical context throughout the dialogue. The center of DH-RAG is a Dynamic Historical Information database, which is further refined by three strategies within the Query Reconstruction Module: Historical Query Clustering, Hierarchical Matching, and Chain of Thought Tracking. Experimental evaluations show that DH-RAG significantly surpasses conventional models on several benchmarks, enhancing response relevance, coherence, and dialogue quality.
Authors: Alan Ramponi, Agnese Daffara, Sara Tonelli
Abstract: We introduce Faina, the first dataset for fallacy detection that embraces multiple plausible answers and natural disagreement. Faina includes over 11K span-level annotations with overlaps across 20 fallacy types on social media posts in Italian about migration, climate change, and public health given by two expert annotators. Through an extensive annotation study that allowed discussion over multiple rounds, we minimize annotation errors whilst keeping signals of human label variation. Moreover, we devise a framework that goes beyond "single ground truth" evaluation and simultaneously accounts for multiple (equally reliable) test sets and the peculiarities of the task, i.e., partial span matches, overlaps, and the varying severity of labeling errors. Our experiments across four fallacy detection setups show that multi-task and multi-label transformer-based approaches are strong baselines across all settings. We release our data, code, and annotation guidelines to foster research on fallacy detection and human label variation more broadly.
Authors: Jie Zou, Mohammad Aliannejadi, Evangelos Kanoulas, Shuxi Han, Heli Ma, Zheng Wang, Yang Yang, Heng Tao Shen
Abstract: Conversational Product Search (CPS) is confined to simulated conversations due to the lack of real-world CPS datasets that reflect human-like language. Additionally, current conversational datasets are limited to support cross-market and multi-lingual usage. In this paper, we introduce a new CPS data collection protocol and present PSCon, a novel CPS dataset designed to assist product search via human-like conversations. The dataset is constructed using a coached human-to-human data collection protocol and supports two languages and dual markets. Also, the dataset enables thorough exploration of six subtasks of CPS: user intent detection, keyword extraction, system action prediction, question selection, item ranking, and response generation. Furthermore, we also offer an analysis of the dataset and propose a benchmark model on the proposed CPS dataset.
Authors: Dan Zhang, Sining Zhoubian, Min Cai, Fengzu Li, Lekang Yang, Wei Wang, Tianjiao Dong, Ziniu Hu, Jie Tang, Yisong Yue
Abstract: This paper presents DataSciBench, a comprehensive benchmark for evaluating Large Language Model (LLM) capabilities in data science. Recent related benchmarks have primarily focused on single tasks, easily obtainable ground truth, and straightforward evaluation metrics, which limits the scope of tasks that can be evaluated. In contrast, DataSciBench is constructed based on a more comprehensive and curated collection of natural and challenging prompts for uncertain ground truth and evaluation metrics. We develop a semi-automated pipeline for generating ground truth (GT) and validating evaluation metrics. This pipeline utilizes and implements an LLM-based self-consistency and human verification strategy to produce accurate GT by leveraging collected prompts, predefined task types, and aggregate functions (metrics). Furthermore, we propose an innovative Task - Function - Code (TFC) framework to assess each code execution outcome based on precisely defined metrics and programmatic rules. Our experimental framework involves testing 6 API-based models, 8 open-source general models, and 9 open-source code generation models using the diverse set of prompts we have gathered. This approach aims to provide a more comprehensive and rigorous evaluation of LLMs in data science, revealing their strengths and weaknesses. Experimental results demonstrate that API-based models outperform open-sourced models on all metrics and Deepseek-Coder-33B-Instruct achieves the highest score among open-sourced models. We release all code and data at https://github.com/THUDM/DataSciBench.
Authors: Tianyu Guo, Hanlin Zhu, Ruiqi Zhang, Jiantao Jiao, Song Mei, Michael I. Jordan, Stuart Russell
Abstract: "Socrates is human. All humans are mortal. Therefore, Socrates is mortal." This classical example demonstrates two-hop reasoning, where a conclusion logically follows from two connected premises. While transformer-based Large Language Models (LLMs) can make two-hop reasoning, they tend to collapse to random guessing when faced with distracting premises. To understand the underlying mechanism, we train a three-layer transformer on synthetic two-hop reasoning tasks. The training dynamics show two stages: a slow learning phase, where the 3-layer transformer performs random guessing like LLMs, followed by an abrupt phase transitions, where the 3-layer transformer suddenly reaches $100%$ accuracy. Through reverse engineering, we explain the inner mechanisms for how models learn to randomly guess between distractions initially, and how they learn to ignore distractions eventually. We further propose a three-parameter model that supports the causal claims for the mechanisms to the training dynamics of the transformer. Finally, experiments on LLMs suggest that the discovered mechanisms generalize across scales. Our methodologies provide new perspectives for scientific understandings of LLMs and our findings provide new insights into how reasoning emerges during training.
Authors: Jaesung Tae, Hamish Ivison, Sachin Kumar, Arman Cohan
Abstract: We introduce TESS 2, a general instruction-following diffusion language model that outperforms contemporary instruction-tuned diffusion models, as well as matches and sometimes exceeds strong autoregressive (AR) models. We train TESS 2 by first adapting a strong AR model via continued pretraining with the usual cross-entropy as diffusion loss, and then performing further instruction tuning. We find that adaptation training as well as the choice of the base model is crucial for training good instruction-following diffusion models. We further propose reward guidance, a novel and modular inference-time guidance procedure to align model outputs without needing to train the underlying model. Finally, we show that TESS 2 further improves with increased inference-time compute, highlighting the utility of diffusion LMs in having fine-grained controllability over the amount of compute used at inference time. Code and models are available at https://github.com/hamishivi/tess-2.
Authors: Guanzheng Chen, Xin Li, Michael Qizhe Shieh, Lidong Bing
Abstract: Large Language Models (LLMs) have demonstrated remarkable capabilities through pretraining and alignment. However, superior short-context LLMs may underperform in long-context scenarios due to insufficient long-context alignment. This alignment process remains challenging due to the impracticality of human annotation for extended contexts and the difficulty in balancing short- and long-context performance. To address these challenges, we introduce LongPO, that enables short-context LLMs to self-evolve to excel on long-context tasks by internally transferring short-context capabilities. LongPO harnesses LLMs to learn from self-generated short-to-long preference data, comprising paired responses generated for identical instructions with long-context inputs and their compressed short-context counterparts, respectively. This preference reveals capabilities and potentials of LLMs cultivated during short-context alignment that may be diminished in under-aligned long-context scenarios. Additionally, LongPO incorporates a short-to-long KL constraint to mitigate short-context performance decline during long-context alignment. When applied to Mistral-7B-Instruct-v0.2 from 128K to 512K context lengths, LongPO fully retains short-context performance and largely outperforms naive SFT and DPO in both long- and short-context tasks. Specifically, \ourMethod-trained models can achieve results on long-context benchmarks comparable to, or even surpassing, those of superior LLMs (e.g., GPT-4-128K) that involve extensive long-context annotation and larger parameter scales.
Authors: Xiaochen Wang, Heming Xia, Jialin Song, Longyu Guan, Yixin Yang, Qingxiu Dong, Weiyao Luo, Yifan Pu, Yiru Wang, Xiangdi Meng, Wenjie Li, Zhifang Sui
Abstract: Large Multimodal Models (LMMs) have achieved remarkable success across various visual-language tasks. However, existing benchmarks predominantly focus on single-image understanding, leaving the analysis of image sequences largely unexplored. To address this limitation, we introduce StripCipher, a comprehensive benchmark designed to evaluate capabilities of LMMs to comprehend and reason over sequential images. StripCipher comprises a human-annotated dataset and three challenging subtasks: visual narrative comprehension, contextual frame prediction, and temporal narrative reordering. Our evaluation of $16$ state-of-the-art LMMs, including GPT-4o and Qwen2.5VL, reveals a significant performance gap compared to human capabilities, particularly in tasks that require reordering shuffled sequential images. For instance, GPT-4o achieves only 23.93% accuracy in the reordering subtask, which is 56.07% lower than human performance. Further quantitative analysis discuss several factors, such as input format of images, affecting the performance of LLMs in sequential understanding, underscoring the fundamental challenges that remain in the development of LMMs.
Authors: Chak Tou Leong, Qingyu Yin, Jian Wang, Wenjie Li
Abstract: The safety alignment of large language models (LLMs) remains vulnerable, as their initial behavior can be easily jailbroken by even relatively simple attacks. Since infilling a fixed template between the input instruction and initial model output is a common practice for existing LLMs, we hypothesize that this template is a key factor behind their vulnerabilities: LLMs' safety-related decision-making overly relies on the aggregated information from the template region, which largely influences these models' safety behavior. We refer to this issue as template-anchored safety alignment. In this paper, we conduct extensive experiments and verify that template-anchored safety alignment is widespread across various aligned LLMs. Our mechanistic analyses demonstrate how it leads to models' susceptibility when encountering inference-time jailbreak attacks. Furthermore, we show that detaching safety mechanisms from the template region is promising in mitigating vulnerabilities to jailbreak attacks. We encourage future research to develop more robust safety alignment techniques that reduce reliance on the template region.
Authors: Jingwang Huang, Jiang Zhong, Qin Lei, Jinpeng Gao, Yuming Yang, Sirui Wang, Peiguang Li, Kaiwen Wei
Abstract: Multimodal multi-label emotion recognition (MMER) aims to identify the concurrent presence of multiple emotions in multimodal data. Existing studies primarily focus on improving fusion strategies and modeling modality-to-label dependencies. However, they often overlook the impact of \textbf{aleatoric uncertainty}, which is the inherent noise in the multimodal data and hinders the effectiveness of modality fusion by introducing ambiguity into feature representations. To address this issue and effectively model aleatoric uncertainty, this paper proposes Latent emotional Distribution Decomposition with Uncertainty perception (LDDU) framework from a novel perspective of latent emotional space probabilistic modeling. Specifically, we introduce a contrastive disentangled distribution mechanism within the emotion space to model the multimodal data, allowing for the extraction of semantic features and uncertainty. Furthermore, we design an uncertainty-aware fusion multimodal method that accounts for the dispersed distribution of uncertainty and integrates distribution information. Experimental results show that LDDU achieves state-of-the-art performance on the CMU-MOSEI and M$^3$ED datasets, highlighting the importance of uncertainty modeling in MMER. Code is available at https://github.com/201983290498/lddu\_mmer.git.
Authors: Guangzhi Xiong, Qiao Jin, Xiao Wang, Yin Fang, Haolin Liu, Yifan Yang, Fangyuan Chen, Zhixing Song, Dengyu Wang, Minjia Zhang, Zhiyong Lu, Aidong Zhang
Abstract: Retrieval-augmented generation (RAG) has shown great potential for knowledge-intensive tasks, but its traditional architectures rely on static retrieval, limiting their effectiveness for complex questions that require sequential information-seeking. While agentic reasoning and search offer a more adaptive approach, most existing methods depend heavily on prompt engineering. In this work, we introduce RAG-Gym, a unified optimization framework that enhances information-seeking agents through fine-grained process supervision at each search step. We also propose ReSearch, a novel agent architecture that synergizes answer reasoning and search query generation within the RAG-Gym framework. Experiments on four challenging datasets show that RAG-Gym improves performance by up to 25.6\% across various agent architectures, with ReSearch consistently outperforming existing baselines. Further analysis highlights the effectiveness of advanced LLMs as process reward judges and the transferability of trained reward models as verifiers for different LLMs. Additionally, we examine the scaling properties of training and inference in agentic RAG. The project homepage is available at https://rag-gym.github.io/.
Authors: Reza Averly, Frazier N. Baker, Xia Ning
Abstract: Drug discovery is a long, expensive, and complex process, relying heavily on human medicinal chemists, who can spend years searching the vast space of potential therapies. Recent advances in artificial intelligence for chemistry have sought to expedite individual drug discovery tasks; however, there remains a critical need for an intelligent agent that can navigate the drug discovery process. Towards this end, we introduce LIDDiA, an autonomous agent capable of intelligently navigating the drug discovery process in silico. By leveraging the reasoning capabilities of large language models, LIDDiA serves as a low-cost and highly-adaptable tool for autonomous drug discovery. We comprehensively examine LIDDiA, demonstrating that (1) it can generate molecules meeting key pharmaceutical criteria on over 70% of 30 clinically relevant targets, (2) it intelligently balances exploration and exploitation in the chemical space, and (3) it can identify promising novel drug candidates on EGFR, a critical target for cancers.
Authors: William Jurayj, Jeffrey Cheng, Benjamin Van Durme
Abstract: Scaling the test-time compute of large language models has demonstrated impressive performance on reasoning benchmarks. However, existing evaluations of test-time scaling make the strong assumption that a reasoning system should always give an answer to any question provided. This overlooks concerns about whether a model is confident in its answer, and whether it is appropriate to always provide a response. To address these concerns, we extract confidence scores during reasoning for thresholding model responses. We find that increasing compute budget at inference time not only helps models answer more questions correctly, but also increases confidence in correct responses. We then extend the current paradigm of zero-risk responses during evaluation by considering settings with non-zero levels of response risk, and suggest a recipe for reporting evaluations under these settings.
Authors: Weihao Liu, Ning Wu, Shiping Yang, Wenbiao Ding, Shining Liang, Ming Gong, Dongmei Zhang
Abstract: Large Language Models (LLMs) frequently show distracted attention due to irrelevant information in the input, which severely impairs their long-context capabilities. Inspired by recent studies on the effectiveness of retrieval heads in long-context factutality, we aim at addressing this distraction issue through improving such retrieval heads directly. We propose Multi-Document Attention Focusing (MuDAF), a novel method that explicitly optimizes the attention distribution at the head level through contrastive learning. According to the experimental results, MuDAF can significantly improve the long-context question answering performance of LLMs, especially in multi-document question answering. Extensive evaluations on retrieval scores and attention visualizations show that MuDAF possesses great potential in making attention heads more focused on relevant information and reducing attention distractions.
Authors: Zhiyuan Zeng, Qinyuan Cheng, Zhangyue Yin, Yunhua Zhou, Xipeng Qiu
Abstract: The advent of test-time scaling in large language models (LLMs), exemplified by OpenAI's o1 series, has advanced reasoning capabilities by scaling computational resource allocation during inference. While successors like QwQ, Deepseek-R1 (R1) and LIMO replicate these advancements, whether these models truly possess test-time scaling capabilities remains underexplored. This study found that longer CoTs of these o1-like models do not consistently enhance accuracy; in fact, correct solutions are often shorter than incorrect ones for the same questions. Further investigation shows this phenomenon is closely related to models' self-revision capabilities - longer CoTs contain more self-revisions, which often lead to performance degradation. We then compare sequential and parallel scaling strategies on QwQ, R1 and LIMO, finding that parallel scaling achieves better coverage and scalability. Based on these insights, we propose Shortest Majority Vote, a method that combines parallel scaling strategies with CoT length characteristics, significantly improving models' test-time scalability compared to conventional majority voting approaches.
Authors: Ziyi Ni, Hao Wang, Huacan Wang
Abstract: Large Language Models (LLMs) have achieved remarkable success in various domains but remain vulnerable to adversarial jailbreak attacks. Existing prompt-defense strategies, including parameter-modifying and parameter-free approaches, face limitations in adaptability, interpretability, and customization, constraining their effectiveness against evolving threats. To address these challenges, we propose ShieldLearner, a novel paradigm that mimics human learning in defense. Through trial and error, it autonomously distills attack signatures into a Pattern Atlas and synthesizes defense heuristics into a Meta-analysis Framework, enabling systematic and interpretable threat detection. Furthermore, we introduce Adaptive Adversarial Augmentation to generate adversarial variations of successfully defended prompts, enabling continuous self-improvement without model retraining. In addition to standard benchmarks, we create a hard test set by curating adversarial prompts from the Wildjailbreak dataset, emphasizing more concealed malicious intent. Experimental results show that ShieldLearner achieves a significantly higher defense success rate than existing baselines on both conventional and hard test sets, while also operating with lower computational overhead, making it a practical and efficient solution for real-world adversarial defense.
Authors: Jun Zhuang, Chaowen Guan
Abstract: In the era of noisy intermediate-scale quantum (NISQ) computing, Quantum Neural Networks (QNNs) have emerged as a promising approach for various applications, yet their training is often hindered by barren plateaus (BPs), where gradient variance vanishes exponentially as the model size increases. To address this challenge, we propose a new Large Language Model (LLM)-driven search framework, AdaInit, that iteratively searches for optimal initial parameters of QNNs to maximize gradient variance and therefore mitigate BPs. Unlike conventional one-time initialization methods, AdaInit dynamically refines QNN's initialization using LLMs with adaptive prompting. Theoretical analysis of the Expected Improvement (EI) proves a supremum for the search, ensuring this process can eventually identify the optimal initial parameter of the QNN. Extensive experiments across four public datasets demonstrate that AdaInit significantly enhances QNN's trainability compared to classic initialization methods, validating its effectiveness in mitigating BPs.
Authors: Enzhe Lu, Zhejun Jiang, Jingyuan Liu, Yulun Du, Tao Jiang, Chao Hong, Shaowei Liu, Weiran He, Enming Yuan, Yuzhi Wang, Zhiqi Huang, Huan Yuan, Suting Xu, Xinran Xu, Guokun Lai, Yanru Chen, Huabin Zheng, Junjie Yan, Jianlin Su, Yuxin Wu, Neo Y. Zhang, Zhilin Yang, Xinyu Zhou, Mingxing Zhang, Jiezhong Qiu
Abstract: Scaling the effective context length is essential for advancing large language models (LLMs) toward artificial general intelligence (AGI). However, the quadratic increase in computational complexity inherent in traditional attention mechanisms presents a prohibitive overhead. Existing approaches either impose strongly biased structures, such as sink or window attention which are task-specific, or radically modify the attention mechanism into linear approximations, whose performance in complex reasoning tasks remains inadequately explored. In this work, we propose a solution that adheres to the ``less structure'' principle, allowing the model to determine where to attend autonomously, rather than introducing predefined biases. We introduce Mixture of Block Attention (MoBA), an innovative approach that applies the principles of Mixture of Experts (MoE) to the attention mechanism. This novel architecture demonstrates superior performance on long-context tasks while offering a key advantage: the ability to seamlessly transition between full and sparse attention, enhancing efficiency without the risk of compromising performance. MoBA has already been deployed to support Kimi's long-context requests and demonstrates significant advancements in efficient attention computation for LLMs. Our code is available at https://github.com/MoonshotAI/MoBA.
Authors: Tejas Srinivasan, Jesse Thomason
Abstract: Trust biases how users rely on AI recommendations in AI-assisted decision-making tasks, with low and high levels of trust resulting in increased under- and over-reliance, respectively. We propose that AI assistants should adapt their behavior through trust-adaptive interventions to mitigate such inappropriate reliance. For instance, when user trust is low, providing an explanation can elicit more careful consideration of the assistant's advice by the user. In two decision-making scenarios -- laypeople answering science questions and doctors making medical diagnoses -- we find that providing supporting and counter-explanations during moments of low and high trust, respectively, yields up to 38% reduction in inappropriate reliance and 20% improvement in decision accuracy. We are similarly able to reduce over-reliance by adaptively inserting forced pauses to promote deliberation. Our results highlight how AI adaptation to user trust facilitates appropriate reliance, presenting exciting avenues for improving human-AI collaboration.
Authors: Tassallah Abdullahi, Ioanna Gemou, Nihal V. Nayak, Ghulam Murtaza, Stephen H. Bach, Carsten Eickhoff, Ritambhara Singh
Abstract: Drug discovery is a complex and time-intensive process that requires identifying and validating new therapeutic candidates. Computational approaches using large-scale biomedical knowledge graphs (KGs) offer a promising solution to accelerate this process. However, extracting meaningful insights from large-scale KGs remains challenging due to the complexity of graph traversal. Existing subgraph-based methods are tailored to graph neural networks (GNNs), making them incompatible with other models, such as large language models (LLMs). We introduce K-Paths, a retrieval framework that extracts structured, diverse, and biologically meaningful paths from KGs. Integrating these paths enables LLMs and GNNs to effectively predict unobserved drug-drug and drug-disease interactions. Unlike traditional path-ranking approaches, K-Paths retrieves and transforms paths into a structured format that LLMs can directly process, facilitating explainable reasoning. K-Paths employs a diversity-aware adaptation of Yen's algorithm to retrieve the K shortest loopless paths between entities in an interaction query, prioritizing biologically relevant and diverse relationships. Our experiments on benchmark datasets show that K-Paths improves the zero-shot performance of Llama 8.1B's F1-score by 12.45 points on drug repurposing and 13.42 points on interaction severity prediction. We also show that Llama 70B achieves F1-score gains of 6.18 and 8.46 points, respectively. K-Paths also improves the supervised training efficiency of EmerGNN, a state-of-the-art GNN, by reducing KG size by 90% while maintaining strong predictive performance. Beyond its scalability and efficiency, K-Paths uniquely bridges the gap between KGs and LLMs, providing explainable rationales for predicted interactions. These capabilities show that K-Paths is a valuable tool for efficient data-driven drug discovery.
Authors: Vishal Dey, Xiao Hu, Xia Ning
Abstract: Despite recent advancements, most computational methods for molecule optimization are constrained to single- or double-property optimization tasks and suffer from poor scalability and generalizability to novel optimization tasks. Meanwhile, Large Language Models (LLMs) demonstrate remarkable out-of-domain generalizability to novel tasks. To demonstrate LLMs' potential for molecule optimization, we introduce $\mathtt{MoMUInstruct}$, the first high-quality instruction-tuning dataset specifically focused on complex multi-property molecule optimization tasks. Leveraging $\mathtt{MoMUInstruct}$, we develop $\mathtt{GeLLM^3O}$s, a series of instruction-tuned LLMs for molecule optimization. Extensive evaluations across 5 in-domain and 5 out-of-domain tasks demonstrate that $\mathtt{GeLLM^3O}$s consistently outperform state-of-the-art baselines. $\mathtt{GeLLM^3O}$s also exhibit outstanding zero-shot generalization to unseen tasks, significantly outperforming powerful closed-source LLMs. Such strong generalizability demonstrates the tremendous potential of $\mathtt{GeLLM^3O}$s as foundational models for molecule optimization, thereby tackling novel optimization tasks without resource-intensive retraining. $\mathtt{MoMUInstruct}$, models, and code are accessible through https://github.com/ninglab/GeLLMO.
Authors: Yang Yan, Bingqing Yue, Qiaxuan Li, Man Huang, Jingyu Chen, Zhenzhong Lan
Abstract: The integration of artificial intelligence in medical imaging has shown tremendous potential, yet the relationship between pre-trained knowledge and performance in cross-modality learning remains unclear. This study investigates how explicitly injecting medical knowledge into the learning process affects the performance of cross-modality classification, focusing on Chest X-ray (CXR) images. We introduce a novel Set Theory-based knowledge injection framework that generates captions for CXR images with controllable knowledge granularity. Using this framework, we fine-tune CLIP model on captions with varying levels of medical information. We evaluate the model's performance through zero-shot classification on the CheXpert dataset, a benchmark for CXR classification. Our results demonstrate that injecting fine-grained medical knowledge substantially improves classification accuracy, achieving 72.5\% compared to 49.9\% when using human-generated captions. This highlights the crucial role of domain-specific knowledge in medical cross-modality learning. Furthermore, we explore the influence of knowledge density and the use of domain-specific Large Language Models (LLMs) for caption generation, finding that denser knowledge and specialized LLMs contribute to enhanced performance. This research advances medical image analysis by demonstrating the effectiveness of knowledge injection for improving automated CXR classification, paving the way for more accurate and reliable diagnostic tools.
Authors: Hongjin Qian, Zheng Liu, Chao Gao, Yankai Wang, Defu Lian, Zhicheng Dou
Abstract: In real-world information-seeking scenarios, users have dynamic and diverse needs, requiring RAG systems to demonstrate adaptable resilience. To comprehensively evaluate the resilience of current RAG methods, we introduce HawkBench, a human-labeled, multi-domain benchmark designed to rigorously assess RAG performance across categorized task types. By stratifying tasks based on information-seeking behaviors, HawkBench provides a systematic evaluation of how well RAG systems adapt to diverse user needs. Unlike existing benchmarks, which focus primarily on specific task types (mostly factoid queries) and rely on varying knowledge bases, HawkBench offers: (1) systematic task stratification to cover a broad range of query types, including both factoid and rationale queries, (2) integration of multi-domain corpora across all task types to mitigate corpus bias, and (3) rigorous annotation for high-quality evaluation. HawkBench includes 1,600 high-quality test samples, evenly distributed across domains and task types. Using this benchmark, we evaluate representative RAG methods, analyzing their performance in terms of answer quality and response latency. Our findings highlight the need for dynamic task strategies that integrate decision-making, query interpretation, and global knowledge understanding to improve RAG generalizability. We believe HawkBench serves as a pivotal benchmark for advancing the resilience of RAG methods and their ability to achieve general-purpose information seeking.
Authors: Jun Zhang, Jue Wang, Huan Li, Lidan Shou, Ke Chen, Yang You, Guiming Xie, Xuejian Gong, Kunlong Zhou
Abstract: Large Language Models (LLMs) have significantly advanced natural language processing with exceptional task generalization capabilities. Low-Rank Adaption (LoRA) offers a cost-effective fine-tuning solution, freezing the original model parameters and training only lightweight, low-rank adapter matrices. However, the memory footprint of LoRA is largely dominated by the original model parameters. To mitigate this, we propose LoRAM, a memory-efficient LoRA training scheme founded on the intuition that many neurons in over-parameterized LLMs have low training utility but are essential for inference. LoRAM presents a unique twist: it trains on a pruned (small) model to obtain pruned low-rank matrices, which are then recovered and utilized with the original (large) model for inference. Additionally, minimal-cost continual pre-training, performed by the model publishers in advance, aligns the knowledge discrepancy between pruned and original models. Our extensive experiments demonstrate the efficacy of LoRAM across various pruning strategies and downstream tasks. For a model with 70 billion parameters, LoRAM enables training on a GPU with only 20G HBM, replacing an A100-80G GPU for LoRA training and 15 GPUs for full fine-tuning. Specifically, QLoRAM implemented by structured pruning combined with 4-bit quantization, for LLaMA-3.1-70B (LLaMA-2-70B), reduces the parameter storage cost that dominates the memory usage in low-rank matrix training by 15.81$\times$ (16.95$\times$), while achieving dominant performance gains over both the original LLaMA-3.1-70B (LLaMA-2-70B) and LoRA-trained LLaMA-3.1-8B (LLaMA-2-13B).
Authors: Xin Li, Anand Sarwate
Abstract: Imposing an effective structural assumption on neural network weight matrices has been the major paradigm for designing Parameter-Efficient Fine-Tuning (PEFT) systems for adapting modern large pre-trained models to various downstream tasks. However, low rank based adaptation has become increasingly challenging due to the sheer scale of modern large language models. In this paper, we propose an effective kernelization to further reduce the number of parameters required for adaptation tasks. Specifically, from the classical idea in numerical analysis regarding matrix Low-Separation-Rank (LSR) representations, we develop a kernel using this representation for the low rank adapter matrices of the linear layers from large networks, named the Low Separation Rank Adaptation (LSR-Adapt) kernel. With the ultra-efficient kernel representation of the low rank adapter matrices, we manage to achieve state-of-the-art performance with even higher accuracy with almost half the number of parameters as compared to conventional low rank based methods. This structural assumption also opens the door to further GPU-side optimizations due to the highly parallelizable nature of Kronecker computations.
Authors: Takuya Matsuyama, Shinji Nishimoto, Yu Takagi
Abstract: Understanding the property of neural populations (or voxels) in the human brain can advance our comprehension of human perceptual and cognitive processing capabilities and contribute to developing brain-inspired computer models. Recent encoding models using deep neural networks (DNNs) have successfully predicted voxel-wise activity. However, interpreting the properties that explain voxel responses remains challenging because of the black-box nature of DNNs. As a solution, we propose LLM-assisted Visual Cortex Captioning (LaVCa), a data-driven approach that uses large language models (LLMs) to generate natural-language captions for images to which voxels are selective. By applying LaVCa for image-evoked brain activity, we demonstrate that LaVCa generates captions that describe voxel selectivity more accurately than the previously proposed method. Furthermore, the captions generated by LaVCa quantitatively capture more detailed properties than the existing method at both the inter-voxel and intra-voxel levels. Furthermore, a more detailed analysis of the voxel-specific properties generated by LaVCa reveals fine-grained functional differentiation within regions of interest (ROIs) in the visual cortex and voxels that simultaneously represent multiple distinct concepts. These findings offer profound insights into human visual representations by assigning detailed captions throughout the visual cortex while highlighting the potential of LLM-based methods in understanding brain representations. Please check out our webpage at https://sites.google.com/view/lavca-llm/
Authors: Or Raphael Bidusa, Shaul Markovitch
Abstract: The opaque nature of Large Language Models (LLMs) has led to significant research efforts aimed at enhancing their interpretability, primarily through post-hoc methods. More recent in-hoc approaches, such as Concept Bottleneck Models (CBMs), offer both interpretability and intervenability by incorporating explicit concept representations. However, these methods suffer from key limitations, including reliance on labeled concept datasets and significant architectural modifications that challenges re-integration into existing system pipelines. In this work, we introduce a new methodology for incorporating interpretability and intervenability into an existing model by integrating Concept Layers (CLs) into its architecture. Our approach projects the model's internal vector representations into a conceptual, explainable vector space before reconstructing and feeding them back into the model. Furthermore, we eliminate the need for a human-selected concept set by algorithmically searching an ontology for a set of concepts that can be either task-specific or task-agnostic. We evaluate CLs across multiple tasks, demonstrating that they maintain the original model's performance and agreement while enabling meaningful interventions. Additionally, we present a proof of concept showcasing an intervenability interface, allowing users to adjust model behavior dynamically, such as mitigating biases during inference.
Authors: Ruida Hu, Chao Peng, Xinchen Wang, Cuiyun Gao
Abstract: Environment configuration is a critical yet time-consuming step in software development, especially when dealing with unfamiliar code repositories. While Large Language Models (LLMs) demonstrate the potential to accomplish software engineering tasks, existing methods for environment configuration often rely on manual efforts or fragile scripts, leading to inefficiencies and unreliable outcomes. We introduce Repo2Run, the first LLM-based agent designed to fully automate environment configuration and generate executable Dockerfiles for arbitrary Python repositories. We address two major challenges: (1) enabling the LLM agent to configure environments within isolated Docker containers, and (2) ensuring the successful configuration process is recorded and accurately transferred to a Dockerfile without error. To achieve this, we propose atomic configuration synthesis, featuring a dual-environment architecture (internal and external environment) with a rollback mechanism to prevent environment "pollution" from failed commands, guaranteeing atomic execution (execute fully or not at all) and a Dockerfile generator to transfer successful configuration steps into runnable Dockerfiles. We evaluate Repo2Run~on our proposed benchmark of 420 recent Python repositories with unit tests, where it achieves an 86.0% success rate, outperforming the best baseline by 63.9%.
Authors: Juyuan Zhang, Wei Zhu, Jiechao Gao
Abstract: Despite the success of Transformer-based models in the time-series prediction (TSP) tasks, the existing Transformer architecture still face limitations and the literature lacks comprehensive explorations into alternative architectures. To address these challenges, we propose AutoFormer-TS, a novel framework that leverages a comprehensive search space for Transformer architectures tailored to TSP tasks. Our framework introduces a differentiable neural architecture search (DNAS) method, AB-DARTS, which improves upon existing DNAS approaches by enhancing the identification of optimal operations within the architecture. AutoFormer-TS systematically explores alternative attention mechanisms, activation functions, and encoding operations, moving beyond the traditional Transformer design. Extensive experiments demonstrate that AutoFormer-TS consistently outperforms state-of-the-art baselines across various TSP benchmarks, achieving superior forecasting accuracy while maintaining reasonable training efficiency.
Authors: Yifei Yang, Zouying Cao, Xinbei Ma, Yao Yao, Libo Qin, Zhi Chen, Hai Zhao
Abstract: Training Large Language Models (LLMs) from scratch requires immense computational resources, making it prohibitively expensive. Model scaling-up offers a promising solution by leveraging the parameters of smaller models to create larger ones. However, existing depth scaling-up methods rely on empirical heuristic rules for layer duplication, which result in poorer initialization and slower convergence during continual pre-training. We propose \textbf{LESA}, a novel learnable method for depth scaling-up. By concatenating parameters from each layer and applying Singular Value Decomposition, we uncover latent patterns between layers, suggesting that inter-layer parameters can be learned. LESA uses a neural network to predict the parameters inserted between adjacent layers, enabling better initialization and faster training. Experiments show that LESA outperforms existing baselines, achieving superior performance with less than half the computational cost during continual pre-training. Extensive analyses demonstrate its effectiveness across different model sizes and tasks.
Authors: Lucas Torroba-Hennigen, Hunter Lang, Han Guo, Yoon Kim
Abstract: We study memory-efficient optimization of neural networks with linear gradient transformations, where the gradients are linearly mapped to a lower dimensional space than the full parameter space, thus saving memory required for gradient accumulation and optimizer state persistence. The model parameters are updated by first performing an optimization step in the lower dimensional space and then going back into the original parameter space via the linear map's transpose. We show that optimizing the model in this transformed space is equivalent to reparameterizing the original model through a linear adapter that additively modifies the model parameters, and then only optimizing the adapter's parameters. When the transformation is Kronecker-factored, this establishes an equivalence between GaLore and one-sided LoRA. We show that this duality between gradient transformations and adapter-based reparameterizations unifies existing approaches to memory-efficient training and suggests new techniques for improving training efficiency and memory use.
Authors: Aleksander Ficek, Somshubra Majumdar, Vahid Noroozi, Boris Ginsburg
Abstract: Code verification has recently found great success as a critical component in training large scale reasoning models for coding. Synthetic techniques such as self-generated test cases and reward models provide a way to enhance code capabilities beyond predefined tests. Building on these advancements, we propose new benchmarks designed to systematically evaluate the impact of synthetic verification methods on assessing solution correctness. We introduce HE-R, HE-R+, MBPP-R, and MBPP-R+, which transform existing coding benchmarks into scoring and ranking datasets to evaluate the effectiveness of synthetic verifiers. Using these benchmarks, we analyze synthetic verification methods in standard, reasoning-based, and reward-based LLMs. Our results show that recent reasoning models significantly improve test case generation and that scaling test cases enhances verification accuracy.
Authors: Justin Singh Kang, Landon Butler, Abhineet Agarwal, Yigit Efe Erginbas, Ramtin Pedarsani, Kannan Ramchandran, Bin Yu
Abstract: Large language models (LLMs) have revolutionized machine learning due to their ability to capture complex interactions between input features. Popular post-hoc explanation methods like SHAP provide marginal feature attributions, while their extensions to interaction importances only scale to small input lengths ($\approx 20$). We propose Spectral Explainer (SPEX), a model-agnostic interaction attribution algorithm that efficiently scales to large input lengths ($\approx 1000)$. SPEX exploits underlying natural sparsity among interactions -- common in real-world data -- and applies a sparse Fourier transform using a channel decoding algorithm to efficiently identify important interactions. We perform experiments across three difficult long-context datasets that require LLMs to utilize interactions between inputs to complete the task. For large inputs, SPEX outperforms marginal attribution methods by up to 20% in terms of faithfully reconstructing LLM outputs. Further, SPEX successfully identifies key features and interactions that strongly influence model output. For one of our datasets, HotpotQA, SPEX provides interactions that align with human annotations. Finally, we use our model-agnostic approach to generate explanations to demonstrate abstract reasoning in closed-source LLMs (GPT-4o mini) and compositional reasoning in vision-language models.
Authors: Daniel A. P. Oliveira, Louren\c{c}o Teodoro, David Martins de Matos
Abstract: Current image captioning systems lack the ability to link descriptive text to specific visual elements, making their outputs difficult to verify. While recent approaches offer some grounding capabilities, they cannot track object identities across multiple references or ground both actions and objects simultaneously. We propose a novel ID-based grounding system that enables consistent object reference tracking and action-object linking, and present GroundCap, a dataset containing 52,016 images from 77 movies, with 344 human-annotated and 52,016 automatically generated captions. Each caption is grounded on detected objects (132 classes) and actions (51 classes) using a tag system that maintains object identity while linking actions to the corresponding objects. Our approach features persistent object IDs for reference tracking, explicit action-object linking, and segmentation of background elements through K-means clustering. We propose gMETEOR, a metric combining caption quality with grounding accuracy, and establish baseline performance by fine-tuning Pixtral-12B. Human evaluation demonstrates our approach's effectiveness in producing verifiable descriptions with coherent object references.
Authors: Xingbo Wang, Janessa Griffith, Daniel A. Adler, Joey Castillo, Tanzeem Choudhury, Fei Wang
Abstract: Despite the prevalence of sleep-tracking devices, many individuals struggle to translate data into actionable improvements in sleep health. Current methods often provide data-driven suggestions but may not be feasible and adaptive to real-life constraints and individual contexts. We present HealthGuru, a novel large language model-powered chatbot to enhance sleep health through data-driven, theory-guided, and adaptive recommendations with conversational behavior change support. HealthGuru's multi-agent framework integrates wearable device data, contextual information, and a contextual multi-armed bandit model to suggest tailored sleep-enhancing activities. The system facilitates natural conversations while incorporating data-driven insights and theoretical behavior change techniques. Our eight-week in-the-wild deployment study with 16 participants compared HealthGuru to a baseline chatbot. Results show improved metrics like sleep duration and activity scores, higher quality responses, and increased user motivation for behavior change with HealthGuru. We also identify challenges and design considerations for personalization and user engagement in health chatbots.
Authors: Shuai Bai, Keqin Chen, Xuejing Liu, Jialin Wang, Wenbin Ge, Sibo Song, Kai Dang, Peng Wang, Shijie Wang, Jun Tang, Humen Zhong, Yuanzhi Zhu, Mingkun Yang, Zhaohai Li, Jianqiang Wan, Pengfei Wang, Wei Ding, Zheren Fu, Yiheng Xu, Jiabo Ye, Xi Zhang, Tianbao Xie, Zesen Cheng, Hang Zhang, Zhibo Yang, Haiyang Xu, Junyang Lin
Abstract: We introduce Qwen2.5-VL, the latest flagship model of Qwen vision-language series, which demonstrates significant advancements in both foundational capabilities and innovative functionalities. Qwen2.5-VL achieves a major leap forward in understanding and interacting with the world through enhanced visual recognition, precise object localization, robust document parsing, and long-video comprehension. A standout feature of Qwen2.5-VL is its ability to localize objects using bounding boxes or points accurately. It provides robust structured data extraction from invoices, forms, and tables, as well as detailed analysis of charts, diagrams, and layouts. To handle complex inputs, Qwen2.5-VL introduces dynamic resolution processing and absolute time encoding, enabling it to process images of varying sizes and videos of extended durations (up to hours) with second-level event localization. This allows the model to natively perceive spatial scales and temporal dynamics without relying on traditional normalization techniques. By training a native dynamic-resolution Vision Transformer (ViT) from scratch and incorporating Window Attention, we reduce computational overhead while maintaining native resolution. As a result, Qwen2.5-VL excels not only in static image and document understanding but also as an interactive visual agent capable of reasoning, tool usage, and task execution in real-world scenarios such as operating computers and mobile devices. Qwen2.5-VL is available in three sizes, addressing diverse use cases from edge AI to high-performance computing. The flagship Qwen2.5-VL-72B model matches state-of-the-art models like GPT-4o and Claude 3.5 Sonnet, particularly excelling in document and diagram understanding. Additionally, Qwen2.5-VL maintains robust linguistic performance, preserving the core language competencies of the Qwen2.5 LLM.
Authors: Shengguang Wu, Fan-Yun Sun, Kaiyue Wen, Nick Haber
Abstract: Recent studies have shown that Large Vision-Language Models (VLMs) tend to neglect image content and over-rely on language-model priors, resulting in errors in visually grounded tasks and hallucinations. We hypothesize that this issue arises because existing VLMs are not explicitly trained to generate texts that are accurately grounded in fine-grained image details. To enhance visual feedback during VLM training, we propose S-VCO (Symmetrical Visual Contrastive Optimization), a novel finetuning objective that steers the model toward capturing important visual details and aligning them with corresponding text tokens. To further facilitate this detailed alignment, we introduce MVC, a paired image-text dataset built by automatically filtering and augmenting visual counterfactual data to challenge the model with hard contrastive cases involving Minimal Visual Contrasts. Experiments show that our method consistently improves VLM performance across diverse benchmarks covering various abilities and domains, achieving up to a 22% reduction in hallucinations, and significant gains in vision-centric and general tasks. Notably, these improvements become increasingly pronounced in benchmarks with higher visual dependency. In short, S-VCO offers a significant enhancement of VLM's visually-dependent task performance while retaining or even improving the model's general abilities. We opensource our code at https://s-vco.github.io/
URLs: https://s-vco.github.io/
Authors: Yuliang Liu, Junjie Lu, Zhaoling Chen, Chaofeng Qu, Jason Klein Liu, Chonghan Liu, Zefan Cai, Yunhui Xia, Li Zhao, Jiang Bian, Chuheng Zhang, Wei Shen, Zhouhan Lin
Abstract: Current approaches for training Process Reward Models (PRMs) often involve breaking down responses into multiple reasoning steps using rule-based techniques, such as using predefined placeholder tokens or setting the reasoning step's length into a fixed size. These approaches overlook the fact that specific words do not typically mark true decision points in a text. To address this, we propose AdaptiveStep, a method that divides reasoning steps based on the model's confidence in predicting the next word. This division method provides more decision-making information at each step, enhancing downstream tasks, such as reward model learning. Moreover, our method does not require manual annotation. We demonstrate its effectiveness through experiments with AdaptiveStep-trained PRMs in mathematical reasoning and code generation tasks. Experimental results indicate that the outcome PRM achieves state-of-the-art Best-of-N performance, surpassing greedy search strategy with token-level value-guided decoding, while also reducing construction costs by over 30% compared to existing open-source PRMs. In addition, we provide a thorough analysis and case study on the PRM's performance, transferability, and generalization capabilities.
Authors: Luke Bates, Peter Ebert Christensen, Preslav Nakov, Iryna Gurevych
Abstract: Templatic memes, characterized by a semantic structure adaptable to the creator's intent, represent a significant yet underexplored area within meme processing literature. With the goal of establishing a new direction for computational meme analysis, here we create a knowledge base composed of more than 5,200 meme templates, information about them, and 54,000 examples of template instances (templatic memes). To investigate the semantic signal of meme templates, we show that we can match memes in datasets to base templates contained in our knowledge base with a distance-based lookup. To demonstrate the power of meme templates, we create TSplit, a method to reorganize datasets, where a template or templatic instance can only appear in either the training or test split. Our re-split datasets enhance general meme knowledge and improve sample efficiency, leading to more robust models. Our examination of meme templates results in state-of-the-art performance for every dataset we consider, paving the way for analysis grounded in templateness.
Authors: Gus Cooney, Andrew Reece
Abstract: Conversation is the subject of increasing interest in the social, cognitive, and computational sciences. And yet, as conversational datasets continue to increase in size and complexity, researchers lack scalable methods to segment speech-to-text transcripts into conversational turns-the basic building blocks of social interaction. We discuss this challenge and then introduce "NaturalTurn," a turn segmentation algorithm designed to accurately capture the dynamics of naturalistic exchange. NaturalTurn operates by distinguishing speakers' primary conversational turns from listeners' secondary utterances, such as backchannels, brief interjections, and other forms of parallel speech that characterize conversation. Using data from a large conversation corpus, we show how NaturalTurn-derived transcripts demonstrate favorable statistical and inferential characteristics compared to transcripts derived from existing methods. The NaturalTurn algorithm represents an improvement in machine-generated transcript processing methods, or "turn models" that will enable researchers to link turn-taking dynamics with the broader outcomes that result from social interaction, a central goal of conversation science.
Authors: Zihao Wei, Jingcheng Deng, Liang Pang, Hanxing Ding, Huawei Shen, Xueqi Cheng
Abstract: The extensive utilization of large language models (LLMs) underscores the crucial necessity for precise and contemporary knowledge embedded within their intrinsic parameters. Existing research on knowledge editing primarily concentrates on monolingual scenarios, neglecting the complexities presented by multilingual contexts and multi-hop reasoning. To address these challenges, our study introduces MLaKE (Multilingual Language Knowledge Editing), a novel benchmark comprising 4072 multi-hop and 5360 single-hop questions designed to evaluate the adaptability of knowledge editing methods across five languages: English, Chinese, Japanese, French, and German. MLaKE aggregates fact chains from Wikipedia across languages and utilizes LLMs to generate questions in both free-form and multiple-choice. We evaluate the multilingual knowledge editing generalization capabilities of existing methods on MLaKE. Existing knowledge editing methods demonstrate higher success rates in English samples compared to other languages. However, their generalization capabilities are limited in multi-language experiments. Notably, existing knowledge editing methods often show relatively high generalization for languages within the same language family compared to languages from different language families. These results underscore the imperative need for advancements in multilingual knowledge editing and we hope MLaKE can serve as a valuable resource for benchmarking and solution development.
Authors: Darren Edge, Ha Trinh, Newman Cheng, Joshua Bradley, Alex Chao, Apurva Mody, Steven Truitt, Dasha Metropolitansky, Robert Osazuwa Ness, Jonathan Larson
Abstract: The use of retrieval-augmented generation (RAG) to retrieve relevant information from an external knowledge source enables large language models (LLMs) to answer questions over private and/or previously unseen document collections. However, RAG fails on global questions directed at an entire text corpus, such as "What are the main themes in the dataset?", since this is inherently a query-focused summarization (QFS) task, rather than an explicit retrieval task. Prior QFS methods, meanwhile, do not scale to the quantities of text indexed by typical RAG systems. To combine the strengths of these contrasting methods, we propose GraphRAG, a graph-based approach to question answering over private text corpora that scales with both the generality of user questions and the quantity of source text. Our approach uses an LLM to build a graph index in two stages: first, to derive an entity knowledge graph from the source documents, then to pregenerate community summaries for all groups of closely related entities. Given a question, each community summary is used to generate a partial response, before all partial responses are again summarized in a final response to the user. For a class of global sensemaking questions over datasets in the 1 million token range, we show that GraphRAG leads to substantial improvements over a conventional RAG baseline for both the comprehensiveness and diversity of generated answers.
Authors: Zhi-Qi Cheng, Xiang Li, Jun-Yan He, Junyao Chen, Xiaomao Fan, Xiaojiang Peng, Alexander G. Hauptmann
Abstract: Emotional Text-to-Speech (E-TTS) synthesis has garnered significant attention in recent years due to its potential to revolutionize human-computer interaction. However, current E-TTS approaches often struggle to capture the intricacies of human emotions, primarily relying on oversimplified emotional labels or single-modality input. In this paper, we introduce the Unified Multimodal Prompt-Induced Emotional Text-to-Speech System (UMETTS), a novel framework that leverages emotional cues from multiple modalities to generate highly expressive and emotionally resonant speech. The core of UMETTS consists of two key components: the Emotion Prompt Alignment Module (EP-Align) and the Emotion Embedding-Induced TTS Module (EMI-TTS). (1) EP-Align employs contrastive learning to align emotional features across text, audio, and visual modalities, ensuring a coherent fusion of multimodal information. (2) Subsequently, EMI-TTS integrates the aligned emotional embeddings with state-of-the-art TTS models to synthesize speech that accurately reflects the intended emotions. Extensive evaluations show that UMETTS achieves significant improvements in emotion accuracy and speech naturalness, outperforming traditional E-TTS methods on both objective and subjective metrics.
Authors: Zhili Liu, Yunhao Gou, Kai Chen, Lanqing Hong, Jiahui Gao, Fei Mi, Yu Zhang, Zhenguo Li, Xin Jiang, Qun Liu, James T. Kwok
Abstract: As the capabilities of large language models (LLMs) continue to expand, aligning these models with human values remains a significant challenge. Recent studies show that reasoning abilities contribute significantly to model safety, while integrating Mixture-of-Experts (MoE) architectures can further enhance alignment. In this work, we propose Mixture of insighTful Experts (MoTE), a novel framework that synergistically combines reasoning chains and expert mixtures to improve self-alignments. From a data perspective, MoTE employs a structured reasoning chain comprising four key stages: Question Analysis, Answer Guidance, Safe Answer, and Safety Checking. This approach enhances safety through multi-step reasoning and proves effective even for smaller and less powerful LLMs (e.g., 7B models). From an architectural perspective, MoTE adopts a multi-LoRA framework with step-level routing, where each expert is dedicated to a specific reasoning step. This design eliminates the need for balance losses, ensures stable training, and supports adaptive inference lengths. Experimental results demonstrate that MoTE significantly improves model safety, jailbreak resistance, and over-refusal capabilities, achieving performance comparable to OpenAI's state-of-the-art o1 model.
Authors: Joshua Harris, Timothy Laurence, Leo Loman, Fan Grayson, Toby Nonnenmacher, Harry Long, Loes WalsGriffith, Amy Douglas, Holly Fountain, Stelios Georgiou, Jo Hardstaff, Kathryn Hopkins, Y-Ling Chi, Galena Kuyumdzhieva, Lesley Larkin, Samuel Collins, Hamish Mohammed, Thomas Finnie, Luke Hounsome, Michael Borowitz, Steven Riley
Abstract: Advances in Large Language Models (LLMs) have led to significant interest in their potential to support human experts across a range of domains, including public health. In this work we present automated evaluations of LLMs for public health tasks involving the classification and extraction of free text. We combine six externally annotated datasets with seven new internally annotated datasets to evaluate LLMs for processing text related to: health burden, epidemiological risk factors, and public health interventions. We evaluate eleven open-weight LLMs (7-123 billion parameters) across all tasks using zero-shot in-context learning. We find that Llama-3.3-70B-Instruct is the highest performing model, achieving the best results on 8/16 tasks (using micro-F1 scores). We see significant variation across tasks with all open-weight LLMs scoring below 60% micro-F1 on some challenging tasks, such as Contact Classification, while all LLMs achieve greater than 80% micro-F1 on others, such as GI Illness Classification. For a subset of 11 tasks, we also evaluate three GPT-4 and GPT-4o series models and find comparable results to Llama-3.3-70B-Instruct. Overall, based on these initial results we find promising signs that LLMs may be useful tools for public health experts to extract information from a wide variety of free text sources, and support public health surveillance, research, and interventions.
Authors: Jun-Yu Ma, Hong Wang, Hao-Xiang Xu, Zhen-Hua Ling, Jia-Chen Gu
Abstract: Model editing is an emerging field that focuses on updating the knowledge embedded within large language models (LLMs) without extensive retraining. However, current model editing methods significantly compromise the general abilities of LLMs as the number of edits increases, and this trade-off poses a substantial challenge to the continual learning of LLMs. In this paper, we first theoretically analyze that the factor affecting the general abilities in sequential model editing lies in the condition number of the edited matrix. The condition number of a matrix represents its numerical sensitivity, and therefore can be used to indicate the extent to which the original knowledge associations stored in LLMs are perturbed after editing. Subsequently, statistical findings demonstrate that the value of this factor becomes larger as the number of edits increases, thereby exacerbating the deterioration of general abilities. To this end, a framework termed Perturbation Restraint on Upper bouNd for Editing (PRUNE) is proposed, which applies the condition number restraints in sequential editing. These restraints can lower the upper bound on perturbation to edited models, thus preserving the general abilities. Systematically, we conduct experiments employing three popular editing methods on three LLMs across four representative downstream tasks. Evaluation results show that PRUNE can preserve considerable general abilities while maintaining the editing performance effectively in sequential model editing.
Authors: Yuchen Wen, Keping Bi, Wei Chen, Jiafeng Guo, Xueqi Cheng
Abstract: As large language models (LLMs) become an important way of information access, there have been increasing concerns that LLMs may intensify the spread of unethical content, including implicit bias that hurts certain populations without explicit harmful words. In this paper, we conduct a rigorous evaluation of LLMs' implicit bias towards certain demographics by attacking them from a psychometric perspective to elicit agreements to biased viewpoints. Inspired by psychometric principles in cognitive and social psychology, we propose three attack approaches, i.e., Disguise, Deception, and Teaching. Incorporating the corresponding attack instructions, we built two benchmarks: (1) a bilingual dataset with biased statements covering four bias types (2.7K instances) for extensive comparative analysis, and (2) BUMBLE, a larger benchmark spanning nine common bias types (12.7K instances) for comprehensive evaluation. Extensive evaluation of popular commercial and open-source LLMs shows that our methods can elicit LLMs' inner bias more effectively than competitive baselines. Our attack methodology and benchmarks offer an effective means of assessing the ethical risks of LLMs, driving progress toward greater accountability in their development.
Authors: Yoo Yeon Sung, Maharshi Gor, Eve Fleisig, Ishani Mondal, Jordan Lee Boyd-Graber
Abstract: Adversarial datasets should validate AI robustness by providing samples on which humans perform well, but models do not. However, as models evolve, datasets can become obsolete. Measuring whether a dataset remains adversarial is hindered by the lack of a standardized metric for measuring adversarialness. We propose AdvScore, a human-grounded evaluation metric that assesses a dataset's adversarialness by capturing models' and humans' varying abilities while also identifying poor examples. We then use AdvScore to motivate a new dataset creation pipeline for realistic and high-quality adversarial samples, enabling us to collect an adversarial question answering (QA) dataset, AdvQA. We apply AdvScore using 9,347 human responses and ten language models' predictions to track model improvement over five years, from 2020 to 2024. AdvScore thus provides guidance for achieving robustness comparable with human capabilities. Furthermore, it helps determine to what extent adversarial datasets continue to pose challenges, ensuring that, rather than reflecting outdated or overly artificial difficulties, they effectively test model capabilities.
Authors: Ritwik Mishra, Sreeram Vennam, Rajiv Ratn Shah, Ponnurangam Kumaraguru
Abstract: Most existing Question Answering Datasets (QuADs) primarily focus on factoid-based short-context Question Answering (QA) in high-resource languages. However, the scope of such datasets for low-resource languages remains limited, with only a few works centered on factoid-based QuADs and none on non-factoid QuADs. Therefore, this work presents MuNfQuAD, a multilingual QuAD with non-factoid questions. It utilizes interrogative sub-headings from BBC news articles as questions and the corresponding paragraphs as silver answers. The dataset comprises over 578K QA pairs across 38 languages, encompassing several low-resource languages, and stands as the largest multilingual QA dataset to date. Based on the manual annotations of 790 QA-pairs from MuNfQuAD (golden set), we observe that 98\% of questions can be answered using their corresponding silver answer. Our fine-tuned Answer Paragraph Selection (APS) model outperforms the baselines. The APS model attained an accuracy of 80\% and 72\%, as well as a macro F1 of 72\% and 66\%, on the MuNfQuAD testset and the golden set, respectively. Furthermore, the APS model effectively generalizes a certain language within the golden set, even after being fine-tuned on silver labels. We also observe that the fine-tuned APS model is beneficial for reducing the context of a question. These findings suggest that this resource would be a valuable contribution to the QA research community.
Authors: Yize Zhao, Tina Behnia, Vala Vakilian, Christos Thrampoulidis
Abstract: Next-token prediction (NTP) over large text corpora has become the go-to paradigm to train large language models. Yet, it remains unclear how NTP influences the mapping of linguistic patterns to geometric properties of the resulting model representations. We frame training of large language models as soft-label classification over sparse probabilistic label vectors, coupled with an analytical approximation that allows unrestricted generation of context embeddings. This approach links NTP training to rank-constrained, nuclear-norm regularized optimization in the logit domain, offering a framework for analyzing the geometry of word and context embeddings. In large embedding spaces, we find that NTP implicitly favors learning logits with a sparse plus low-rank structure. While the sparse component captures the co-occurrence frequency of context-word pairs, the orthogonal low-rank component, which becomes dominant as training progresses, depends solely on the sparsity pattern of the co-occurrence matrix. Consequently, when projected onto an appropriate subspace, representations of contexts that are followed by the same set of next-tokens collapse, a phenomenon we term subspace-collapse. We validate our findings on synthetic and small-scale real language datasets. Finally, we outline potential research directions aimed at deepening the understanding of NTP's influence on the learning of linguistic patterns and regularities.
Authors: Rena Gao, Jingxuan Wu, Xuetong Wu, Carsten Roever, Jing Wu, Long Lv, Jey Han Lau
Abstract: We analyse the cross-lingual transferability of a dialogue evaluation framework that assesses the relationships between micro-level linguistic features (e.g. backchannels) and macro-level interactivity labels (e.g. topic management), originally designed for English-as-a-second-language dialogues. To this end, we develop CNIMA (Chinese Non-Native Interactivity Measurement and Automation), a Chinese-as-a-second-language labelled dataset with 10K dialogues. We found the evaluation framework to be robust across distinct languages: English and Chinese, revealing language-specific and language-universal relationships between micro-level and macro-level features. Next, we propose an automated, interpretable approach with low data requirement that scores the overall quality of a second-language dialogue based on the framework. Our approach is interpretable in that it reveals the key linguistic and interactivity features that contributed to the overall quality score. As our approach does not require labelled data, it can also be adapted to other languages for second-language dialogue evaluation.
Authors: Siheng Xiong, Ali Payani, Yuan Yang, Faramarz Fekri
Abstract: Enhancing the reasoning capabilities of language models (LMs) remains a key challenge, especially for tasks that require complex, multi-step decision-making where existing Chain-of-Thought (CoT) approaches struggle with consistency and verification. In this paper, we propose a novel reasoning framework, referred to as Structure-aware Planning with an Accurate World Model (SWAP), that integrates structured knowledge representation with learned planning. Unlike prior methods that rely purely on natural language reasoning, SWAP leverages entailment graphs to encode structured dependencies and enable symbolic verification of intermediate steps. To systematically construct and update the graph, SWAP employs a policy model to propose candidate expansions and a world model to predict structural updates. To improve accuracy, the world model generates multiple alternative updates, and a discriminator re-ranks them based on plausibility. To encourage diverse exploration, we introduce Diversity-based Modelling (DM), which samples candidates from the remaining probability mass after removing previously sampled candidates from the original policy distribution. Additionally, SWAP improves the discrimination accuracy through Contrastive Ranking (CR), which directly compares candidates within prompts and incorporates meta-knowledge to improve ranking quality. We evaluate SWAP across diverse reasoning-intensive benchmarks including math reasoning, logical reasoning, and coding tasks. Extensive experiments demonstrate that SWAP significantly improves upon the base models and consistently outperforms existing reasoning methods.
Authors: Zhuochun Li, Yuelyu Ji, Rui Meng, Daqing He
Abstract: While reasoning capabilities typically emerge in large language models (LLMs) with tens of billions of parameters, recent research focuses on improving smaller open-source models through knowledge distillation (KD) from commercial LLMs. However, many of these studies rely solely on responses from a single LLM as the gold rationale, unlike the natural human learning process, which involves understanding both the correct answers and the reasons behind mistakes. In this paper, we introduce a novel Fault-Aware DistIllation via Peer-Review (FAIR) approach: 1) Instead of merely obtaining rationales from teachers, our method asks teachers to identify and explain the student's mistakes, providing customized instruction learning data. 2) We design a simulated peer-review process between teacher LLMs, which selects only the generated rationales above the acceptance threshold. This reduces the chance of teachers guessing correctly with flawed rationale, improving instructional data quality. Comprehensive experiments and analysis on mathematical, commonsense, and logical reasoning tasks demonstrate the effectiveness of our method.
Authors: Mutian He, Philip N. Garner
Abstract: Architectures such as Linformer and Mamba have recently emerged as competitive linear time replacements for transformers. However, corresponding large pretrained models are often unavailable, especially in non-text domains. To remedy this, we present a Cross-Architecture Layerwise Distillation (CALD) approach that jointly converts a transformer model to a linear time substitute and fine-tunes it to a target task. We also compare several means to guide the fine-tuning to optimally retain the desired inference capability from the original model. The methods differ in their use of the target model and the trajectory of the parameters. In a series of empirical studies on language processing, language modeling, and speech processing, we show that CALD can effectively recover the result of the original model, and that the guiding strategy contributes to the result. Some reasons for the variation are suggested.
Authors: Yuan Sui, Yufei He, Zifeng Ding, Bryan Hooi
Abstract: Recent works integrating Knowledge Graphs (KGs) have led to promising improvements in enhancing the reasoning accuracy of Large Language Models (LLMs). However, current benchmarks focus mainly on closed-ended tasks, leaving a gap in the assessment of more complex real-world scenarios. This gap has also obscured the evaluation of KGs' potential to mitigate the problem of hallucination in LLMs. To fill the gap, we introduce OKGQA, a new benchmark specifically designed to assess LLMs enhanced with KGs under open-ended, real-world question answering scenarios. OKGQA is designed to closely reflect the complexities of practical applications using questions from different types, and incorporates specific metrics to measure both hallucination ratio and the enhancement in reasoning capabilities. To consider the scenario in which KGs may have varying levels of mistakes, we propose another benchmark variant OKGQA-P to assess model performance when the semantics and structure of KGs are deliberately perturbed and contaminated. OKGQA aims to (1) explore whether KGs can make LLMs more trustworthy in an open-ended setting, and (2) conduct a comparative analysis to shed light on method design. We believe that this study can facilitate a more complete performance comparison and encourage continuous improvement in integrating KGs with LLMs to reduce hallucination.
Authors: Seiji Maekawa, Hayate Iso, Nikita Bhutani
Abstract: The rapid increase in textual information means we need more efficient methods to sift through, organize, and understand it all. While retrieval-augmented generation (RAG) models excel in accessing information from large document collections, they struggle with complex tasks that require aggregation and reasoning over information spanning across multiple documents--what we call holistic reasoning. Long-context language models (LCLMs) have great potential for managing large-scale documents, but their holistic reasoning capabilities remain unclear. In this work, we introduce HoloBench, a novel framework that brings database reasoning operations into text-based contexts, making it easier to systematically evaluate how LCLMs handle holistic reasoning across large documents. Our approach adjusts key factors such as context length, information density, distribution of information, and query complexity to evaluate LCLMs comprehensively. Our experiments show that the amount of information in the context has a bigger influence on LCLM performance than the actual context length. Furthermore, the complexity of queries affects performance more than the amount of information, particularly for different types of queries. Interestingly, queries that involve finding maximum or minimum values are easier for LCLMs and are less affected by context length, even though they pose challenges for RAG systems. However, tasks requiring the aggregation of multiple pieces of information show a noticeable drop in accuracy as context length increases. Additionally, we find that while grouping relevant information generally improves performance, the optimal positioning varies across models. Our findings surface both the advancements and the ongoing challenges in achieving a holistic understanding of long contexts.
Authors: Junyu Luo, Xiao Luo, Xiusi Chen, Zhiping Xiao, Wei Ju, Ming Zhang
Abstract: Supervised fine-tuning (SFT) is crucial in adapting large language model (LLMs) to a specific domain or task. However, only a limited amount of labeled data is available in practical applications, which poses a severe challenge for SFT in yielding satisfactory results. Therefore, a data-efficient framework that can fully exploit labeled and unlabeled data for LLM fine-tuning is highly anticipated.Towards this end, we introduce a semi-supervised fine-tuning(SemiFT) task and a framework named SemiEvol for LLM alignment from a propagate-and-select manner. For knowledge propagation, SemiEvol adopts a bi-level approach, propagating knowledge from labeled data to unlabeled data through both in-weight and in-context methods. For knowledge selection, SemiEvol incorporates a collaborative learning mechanism, selecting higher-quality pseudo-response samples. We conducted experiments using GPT-4o-mini and Llama-3.1 on seven general or domain-specific datasets, demonstrating significant improvements in model performance on target data. Furthermore, we compared SemiEvol with SFT and self-evolution methods, highlighting its practicality in hybrid data scenarios.
Authors: Bryan R. Christ, Zack Gottesman, Jonathan Kropko, Thomas Hartvigsen
Abstract: Math reasoning is an active area of Large Language Model (LLM) research because it is a hallmark of artificial intelligence and has implications in several domains, including math education. However, few works have explored how math reasoning is encoded within LLM parameters and if it is a skill that can be isolated within models. Doing so could allow targeted intervention to improve math performance without altering non-math behavior and foster understanding of how models encode math reasoning. We introduce Math Neurosurgery (MathNeuro), a computationally efficient method we use to isolate math-specific parameters in LLMs using only forward passes. MathNeuro builds on existing work by using weights and activations to calculate parameter importance, but isolates math-specific parameters by filtering out those important for general language tasks. Through pruning parameters MathNeuro identifies, we delete a LLM's math reasoning ability without significantly impacting its general language ability. Scaling the identified parameters by a small constant improves a pretrained or instruction-tuned LLM's performance by 4-17% on GSM8K and 5-35% on MATH while leaving non-math behavior unaltered. MathNeuro is also data efficient: most of its effectiveness holds when identifying math-specific parameters using a single sample. MathNeuro highlights the potential for future work to intervene on math-specific parameters.
Authors: Zhanchao Zhou, Tianyi Wu, Zhiyun Jiang, Fares Obeid, Zhenzhong Lan
Abstract: While Transformer models have achieved remarkable success in various domains, the effectiveness of information propagation through deep networks remains a critical challenge. Standard hidden state residuals often fail to adequately preserve initial token-level information in deeper layers. This paper introduces ResFormer, a novel architecture that enhances information flow by incorporating value residual connections in addition to hidden state residuals. And a variant is the SVFormer, where all layers share the first layer's value embedding. Comprehensive empirical evidence demonstrates ResFormer achieves equivalent validation loss with 13.3\% fewer model parameters and 15.4\% less training data compared to Transformer, while maintaining similar memory usage and computational cost. Besides, SVFormer reduces KV cache size by nearly half with only a small performance penalty and can be integrated with other KV-efficient methods, yielding further reductions in KV cache, with performance influenced by sequence length and cumulative learning rate.
Authors: Elisabeth Kirsten, Ivan Habernal, Vedant Nanda, Muhammad Bilal Zafar
Abstract: Last few years have seen unprecedented advances in capabilities of Large Language Models (LLMs). These advancements promise to benefit a vast array of application domains. However, due to their immense size, performing inference with LLMs is both costly and slow. Consequently, a plethora of recent work has proposed strategies to enhance inference efficiency, e.g., quantization, pruning, and caching. These acceleration strategies reduce the inference cost and latency, often by several factors, while maintaining much of the predictive performance measured via common benchmarks. In this work, we explore another critical aspect of LLM performance: demographic bias in model generations due to inference acceleration optimizations. Using a wide range of metrics, we probe bias in model outputs from a number of angles. Analysis of outputs before and after inference acceleration shows significant change in bias. Worryingly, these bias effects are complex and unpredictable. A combination of an acceleration strategy and bias type may show little bias change in one model but may lead to a large effect in another. Our results highlight a need for in-depth and case-by-case evaluation of model bias after it has been modified to accelerate inference.
Authors: Kyle Gorman, Yuval Pinter
Abstract: The common practice of preprocessing text before feeding it into NLP models introduces many decision points which have unintended consequences on model performance. In this opinion piece, we focus on the handling of diacritics in texts originating in many languages and scripts. We demonstrate, through several case studies, the adverse effects of inconsistent encoding of diacritized characters and of removing diacritics altogether. We call on the community to adopt simple but necessary steps across all models and toolkits in order to improve handling of diacritized text and, by extension, increase equity in multilingual NLP.
Authors: Seonil Son, Ju-Min Oh, Heegon Jin, Cheolhun Jang, Jeongbeom Jeong, Kuntae Kim
Abstract: Most existing benchmarking approaches for evaluating the output quality of large language models (LLMs) rely on comparing LLM responses to predefined references. Such methods, based on static datasets, quickly become outdated as LLM capabilities and use cases evolve. In this work, we introduce VARCO Arena--a novel, cost-effective, and robust benchmarking approach that leverages a single-elimination tournament structure to minimize the number of required comparisons while eliminating the need for static references or costly human annotations. We validate our approach through two experiments: (i) a simulation study that examines its robustness under various conditions, and (ii) an empirical evaluation using publicly available benchmark prompts. In both experiments, VARCO Arena consistently outperforms current LLM benchmarking practices, achieving stronger correlations with human-established Elo ratings. Our results demonstrate that VARCO Arena not only produces reliable LLM rankings but also provides a scalable, adaptable solution for qualitative evaluation across diverse, customized use cases.
Authors: Aliyah R. Hsu, James Zhu, Zhichao Wang, Bin Bi, Shubham Mehrotra, Shiva K. Pentyala, Katherine Tan, Xiang-Bo Mao, Roshanak Omrani, Sougata Chaudhuri, Regunathan Radhakrishnan, Sitaram Asur, Claire Na Cheng, Bin Yu
Abstract: LLMs have demonstrated impressive proficiency in generating coherent and high-quality text, making them valuable across a range of text-generation tasks. However, rigorous evaluation of this generated content is crucial, as ensuring its quality remains a significant challenge due to persistent issues such as factual inaccuracies and hallucination. This paper introduces three fine-tuned general-purpose LLM autoevaluators, REC-8B, REC-12B and REC-70B, specifically designed to evaluate generated text across several dimensions: faithfulness, instruction following, coherence, and completeness. These models not only provide ratings for these metrics but also offer detailed explanation and verifiable citation, thereby enhancing trust in the content. Moreover, the models support various citation modes, accommodating different requirements for latency and granularity. Extensive evaluations on diverse benchmarks demonstrate that our general-purpose LLM auto-evaluator, REC-70B, outperforms state-of-the-art LLMs, excelling in content evaluation by delivering better quality explanation and citation with minimal bias. It achieves Rank #1 as of Feb 15th, 2025 as a generative model on the RewardBench leaderboard under the model name TextEval-Llama3.1-70B. Our REC dataset and models are available at https://github.com/adelaidehsu/REC.
Authors: Anirudh Phukan, Divyansh, Harshit Kumar Morj, Vaishnavi, Apoorv Saxena, Koustava Goswami
Abstract: The rapid development of Large Multimodal Models (LMMs) has significantly advanced multimodal understanding by harnessing the language abilities of Large Language Models (LLMs) and integrating modality-specific encoders. However, LMMs are plagued by hallucinations that limit their reliability and adoption. While traditional methods to detect and mitigate these hallucinations often involve costly training or rely heavily on external models, recent approaches utilizing internal model features present a promising alternative. In this paper, we critically assess the limitations of the state-of-the-art training-free technique, the logit lens, in handling generalized visual hallucinations. We introduce ContextualLens, a refined method that leverages contextual token embeddings from middle layers of LMMs. This approach significantly improves hallucination detection and grounding across diverse categories, including actions and OCR, while also excelling in tasks requiring contextual understanding, such as spatial relations and attribute comparison. Our novel grounding technique yields highly precise bounding boxes, facilitating a transition from Zero-Shot Object Segmentation to Grounded Visual Question Answering. Our contributions pave the way for more reliable and interpretable multimodal models.
Authors: Udo Hahn
Abstract: We survey clinical document corpora, with focus on German textual data. Due to rigid data privacy legislation in Germany these resources, with only few exceptions, are stored in safe clinical data spaces and locked against clinic-external researchers. This situation stands in stark contrast with established workflows in the field of natural language processing where easy accessibility and reuse of data collections are common practice. Hence, alternative corpus designs have been examined to escape from this data poverty. Besides machine translation of English clinical datasets and the generation of synthetic corpora with fictitious clinical contents, several other types of domain proxies have come up as substitutes for clinical documents. Common instances of close proxies are medical journal publications, therapy guidelines, drug labels, etc., more distant proxies include online encyclopedic medical articles or medical contents from social media channels. After PRISM-conformant identification of 362 hits from 4 bibliographic systems, 78 relevant documents were finally selected for this review. They contained overall 92 different published versions of corpora from which 71 were truly unique in terms of their underlying document sets. Out of these, the majority were clinical corpora -- 46 real ones, 5 translated ones, and 6 synthetic ones. As to domain proxies, we identified 18 close and 17 distant ones. There is a clear divide between the large number of non-accessible authentic clinical German-language corpora and their publicly accessible substitutes: translated or synthetic, close or more distant proxies. So on first sight, the data bottleneck seems broken. Yet differences in genre-specific writing style, wording and medical domain expertise in this typological space are also obvious. This raises the question how valid alternative corpus designs really are.
Authors: Shivalika Singh, Angelika Romanou, Cl\'ementine Fourrier, David I. Adelani, Jian Gang Ngui, Daniel Vila-Suero, Peerat Limkonchotiwat, Kelly Marchisio, Wei Qi Leong, Yosephine Susanto, Raymond Ng, Shayne Longpre, Wei-Yin Ko, Sebastian Ruder, Madeline Smith, Antoine Bosselut, Alice Oh, Andre F. T. Martins, Leshem Choshen, Daphne Ippolito, Enzo Ferrante, Marzieh Fadaee, Beyza Ermis, Sara Hooker
Abstract: Cultural biases in multilingual datasets pose significant challenges for their effectiveness as global benchmarks. These biases stem not only from differences in language but also from the cultural knowledge required to interpret questions, reducing the practical utility of translated datasets like MMLU. Furthermore, translation often introduces artefacts that can distort the meaning or clarity of questions in the target language. A common practice in multilingual evaluation is to rely on machine-translated evaluation sets, but simply translating a dataset is insufficient to address these challenges. In this work, we trace the impact of both of these issues on multilingual evaluations and ensuing model performances. Our large-scale evaluation of state-of-the-art open and proprietary models illustrates that progress on MMLU depends heavily on learning Western-centric concepts, with 28% of all questions requiring culturally sensitive knowledge. Moreover, for questions requiring geographic knowledge, an astounding 84.9% focus on either North American or European regions. Rankings of model evaluations change depending on whether they are evaluated on the full portion or the subset of questions annotated as culturally sensitive, showing the distortion to model rankings when blindly relying on translated MMLU. We release Global MMLU, an improved MMLU with evaluation coverage across 42 languages -- with improved overall quality by engaging with compensated professional and community annotators to verify translation quality while also rigorously evaluating cultural biases present in the original dataset. This comprehensive Global MMLU set also includes designated subsets labeled as culturally sensitive and culturally agnostic to allow for more holistic, complete evaluation.
Authors: Minzheng Wang, Xinghua Zhang, Kun Chen, Nan Xu, Haiyang Yu, Fei Huang, Wenji Mao, Yongbin Li
Abstract: Large language models (LLMs) enabled dialogue systems have become one of the central modes in human-machine interaction, which bring about vast amounts of conversation logs and increasing demand for dialogue generation. The dialogue's life-cycle spans from $\textit{Prelude}$ through $\textit{Interlocution}$ to $\textit{Epilogue}$, encompassing rich dialogue elements. Despite large volumes of dialogue-related studies, there is a lack of systematic investigation into the dialogue stages to frame benchmark construction that covers comprehensive dialogue elements. This hinders the precise modeling, generation and assessment of LLMs-based dialogue systems. To bridge this gap, in this paper, we introduce a new research task--$\textbf{D}$ialogue $\textbf{E}$lement $\textbf{MO}$deling, including $\textit{Element Awareness}$ and $\textit{Dialogue Agent Interaction}$, and propose a novel benchmark, $\textbf{DEMO}$, designed for a comprehensive dialogue modeling and assessment. On this basis, we further build the DEMO agent with the adept ability to model dialogue elements via imitation learning. Extensive experiments on DEMO indicate that current representative LLMs still have considerable potential for enhancement, and our DEMO agent performs well in both dialogue element modeling and out-of-domain tasks.
Authors: Bo Lv, Chen Tang, Yanan Zhang, Xin Liu, Yue Yu, Ping Luo
Abstract: Ensembles of generative large language models (LLMs) can integrate the strengths of different LLMs to compensate for the limitations of individual models. However, recent work has focused on training an additional fusion model to combine complete responses from multiple LLMs, failing to tap into their collaborative potential to generate higher-quality responses. Moreover, as the additional fusion model is trained on a specialized dataset, these methods struggle with generalizing to open-domain queries from online users. In this paper, we propose SpecFuse, a novel ensemble framework that outputs the fused result by iteratively producing the next segment through collaboration among LLMs. This is achieved through cyclic execution of its inference and verification components. In each round, the inference component invokes each base LLM to generate candidate segments in parallel, and the verify component calls these LLMs again to predict the ranking of the segments. The top-ranked segment is then broadcast to all LLMs, encouraging them to generate higher-quality segments in the next round. This approach also allows the base LLMs to be plug-and-play, without any training or adaptation, avoiding generalization limitations. Furthermore, to conserve computational resources, we propose a model exit mechanism that dynamically excludes models exhibiting poor performance in previous rounds during each query response. In this way, it effectively reduces the number of model calls while maintaining overall performance.
Authors: Wonjin Lee, Kyumin Kim, Sungjae Lee, Jihun Lee, Kwang In Kim
Abstract: Applying language models (LMs) to tables is challenging due to the inherent structural differences between two-dimensional tables and one-dimensional text for which the LMs were originally designed. Furthermore, when applying linearized tables to LMs, the maximum token lengths often imposed in self-attention calculations make it difficult to comprehensively understand the context spread across large tables. To address these challenges, we present PieTa (Piece of Table), a new framework for subtable-based question answering (QA). PieTa operates through an iterative process of dividing tables into smaller windows, using LMs to select relevant cells within each window, and merging these cells into a subtable. This multi-resolution approach captures dependencies across multiple rows and columns while avoiding the limitations caused by long context inputs. Instantiated as a simple iterative subtable union algorithm, PieTa demonstrates improved performance over previous subtable-based QA approaches.
Authors: Mengze Hong, Chen Jason Zhang, Chaotao Chen, Rongzhong Lian, Di Jiang
Abstract: Maintaining persona consistency is paramount in the application of open-domain dialogue systems, as exemplified by models like ChatGPT. Despite significant advancements, the limited scale and diversity of current persona dialogue datasets remain challenges to achieving robust persona-consistent dialogue models. In this study, drawing inspiration from the success of large-scale pre-training, we introduce PPDS, an open-domain persona dialogue system that employs extensive generative pre-training on a persona dialogue dataset to enhance persona consistency. Specifically, we present a persona extraction model designed to autonomously and precisely generate vast persona dialogue datasets. Additionally, we unveil a pioneering persona augmentation technique to address the invalid persona bias inherent in the constructed dataset. Both quantitative and human evaluations consistently highlight the superior response quality and persona consistency of our proposed model, underscoring its effectiveness.
Authors: Guangsheng Bao, Yanbin Zhao, Juncai He, Yue Zhang
Abstract: Advanced large language models (LLMs) can generate text almost indistinguishable from human-written text, highlighting the importance of LLM-generated text detection. However, current zero-shot techniques face challenges as white-box methods are restricted to use weaker open-source LLMs, and black-box methods are limited by partial observation from stronger proprietary LLMs. It seems impossible to enable white-box methods to use proprietary models because API-level access to the models neither provides full predictive distributions nor inner embeddings. To traverse the divide, we propose **Glimpse**, a probability distribution estimation approach, predicting the full distributions from partial observations. Despite the simplicity of Glimpse, we successfully extend white-box methods like Entropy, Rank, Log-Rank, and Fast-DetectGPT to latest proprietary models. Experiments show that Glimpse with Fast-DetectGPT and GPT-3.5 achieves an average AUROC of about 0.95 in five latest source models, improving the score by 51% relative to the remaining space of the open source baseline. It demonstrates that the latest LLMs can effectively detect their own outputs, suggesting that advanced LLMs may be the best shield against themselves. We release our code and data at https://github.com/baoguangsheng/glimpse.
Authors: Yu Yan, Sheng Sun, Junqi Tong, Min Liu, Qi Li
Abstract: Metaphor serves as an implicit approach to convey information, while enabling the generalized comprehension of complex subjects. However, metaphor can potentially be exploited to bypass the safety alignment mechanisms of Large Language Models (LLMs), leading to the theft of harmful knowledge. In our study, we introduce a novel attack framework that exploits the imaginative capacity of LLMs to achieve jailbreaking, the J\underline{\textbf{A}}ilbreak \underline{\textbf{V}}ia \underline{\textbf{A}}dversarial Me\underline{\textbf{TA}} -pho\underline{\textbf{R}} (\textit{AVATAR}). Specifically, to elicit the harmful response, AVATAR extracts harmful entities from a given harmful target and maps them to innocuous adversarial entities based on LLM's imagination. Then, according to these metaphors, the harmful target is nested within human-like interaction for jailbreaking adaptively. Experimental results demonstrate that AVATAR can effectively and transferablly jailbreak LLMs and achieve a state-of-the-art attack success rate across multiple advanced LLMs. Our study exposes a security risk in LLMs from their endogenous imaginative capabilities. Furthermore, the analytical study reveals the vulnerability of LLM to adversarial metaphors and the necessity of developing defense methods against jailbreaking caused by the adversarial metaphor. \textcolor{orange}{ \textbf{Warning: This paper contains potentially harmful content from LLMs.}}
Authors: Seunghee Kim, Changhyeon Kim, Taeuk Kim
Abstract: Real-world decision-making often requires integrating and reasoning over information from multiple modalities. While recent multimodal large language models (MLLMs) have shown promise in such tasks, their ability to perform multi-hop reasoning across diverse sources remains insufficiently evaluated. Existing benchmarks, such as MMQA, face challenges due to (1) data contamination and (2) a lack of complex queries that necessitate operations across more than two modalities, hindering accurate performance assessment. To address this, we present Financial Cross-Modal Multi-Hop Reasoning (FCMR), a benchmark created to analyze the reasoning capabilities of MLLMs by urging them to combine information from textual reports, tables, and charts within the financial domain. FCMR is categorized into three difficulty levels-Easy, Medium, and Hard-facilitating a step-by-step evaluation. In particular, problems at the Hard level require precise cross-modal three-hop reasoning and are designed to prevent the disregard of any modality. Experiments on this new benchmark reveal that even state-of-the-art MLLMs struggle, with the best-performing model (Claude 3.5 Sonnet) achieving only 30.4% accuracy on the most challenging tier. We also conduct analysis to provide insights into the inner workings of the models, including the discovery of a critical bottleneck in the information retrieval phase.
Authors: Jiayuan Su, Zhaopeng Feng, Jiamei Zheng, Jiahan Ren, Yan Zhang, Jian Wu, Hongwei Wang, Zuozhu Liu
Abstract: Recent advancements in large language models (LLMs) have given rise to the LLM-as-a-judge paradigm, showcasing their potential to deliver human-like judgments. However, in the field of machine translation (MT) evaluation, current LLM-as-a-judge methods fall short of learned automatic metrics. In this paper, we propose Multidimensional Multi-Agent Debate (M-MAD), a systematic LLM-based multi-agent framework for advanced LLM-as-a-judge MT evaluation. Our findings demonstrate that M-MAD achieves significant advancements by (1) decoupling heuristic MQM criteria into distinct evaluation dimensions for fine-grained assessments; (2) employing multi-agent debates to harness the collaborative reasoning capabilities of LLMs; (3) synthesizing dimension-specific results into a final evaluation judgment to ensure robust and reliable outcomes. Comprehensive experiments show that M-MAD not only outperforms all existing LLM-as-a-judge methods but also competes with state-of-the-art reference-based automatic metrics, even when powered by a suboptimal model like GPT-4o mini. Detailed ablations and analysis highlight the superiority of our framework design, offering a fresh perspective for LLM-as-a-judge paradigm. Our code and data are publicly available at https://github.com/SU-JIAYUAN/M-MAD.
Authors: Zhen Li, Yupeng Su, Runming Yang, Congkai Xie, Zheng Wang, Zhongwei Xie, Ngai Wong, Hongxia Yang
Abstract: Large language models have achieved significant advancements in complex mathematical reasoning benchmarks, such as MATH. However, their substantial computational requirements present challenges for practical deployment. Model quantization has emerged as an effective strategy to reduce memory usage and computational costs by employing lower precision and bit-width representations. In this study, we systematically evaluate the impact of quantization on mathematical reasoning tasks. Our results demonstrate that aggressive quantization methods like AWQ and GPTQ introduce up to 32.39% accuracy degradation (average 11.31%) on Llama-3 models, particularly in numerical computation and reasoning planning. To address this, we introduce a multidimensional evaluation framework combining qualitative capability analysis and quantitative error assessment. We further develop targeted recovery strategies, showing that fine-tuning quantized models on only 545 task-specific examples for 3 minutes on 4 GPUs effectively restores reasoning capabilities to near full-precision levels. Additionally, our error assessment pipeline achieves 98.9% accuracy in diagnosing and localizing errors across 3,366 failure cases, providing actionable insights for mitigating quantization-induced degradation.
Authors: Elham Aghakhani, Lu Wang, Karla T. Washington, George Demiris, Jina Huh-Yoo, Rezvaneh Rezapour
Abstract: Problem-solving therapy (PST) is a structured psychological approach that helps individuals manage stress and resolve personal issues by guiding them through problem identification, solution brainstorming, decision-making, and outcome evaluation. As mental health care increasingly adopts technologies like chatbots and large language models (LLMs), it is important to thoroughly understand how each session of PST is conducted before attempting to automate it. We developed a comprehensive framework for PST annotation using established PST Core Strategies and a set of novel Facilitative Strategies to analyze a corpus of real-world therapy transcripts to determine which strategies are most prevalent. Using various LLMs and transformer-based models, we found that GPT-4o outperformed all models, achieving the highest accuracy (0.76) in identifying all strategies. To gain deeper insights, we examined how strategies are applied by analyzing Therapeutic Dynamics (autonomy, self-disclosure, and metaphor), and linguistic patterns within our labeled data. Our research highlights LLMs' potential to automate therapy dialogue analysis, offering a scalable tool for mental health interventions. Our framework enhances PST by improving accessibility, effectiveness, and personalized support for therapists.
Authors: Fabian David Schmidt, Ivan Vuli\'c, Goran Glava\v{s}, David Ifeoluwa Adelani
Abstract: Spoken language understanding (SLU) is indispensable for half of all living languages that lack a formal writing system, since these languages cannot pair automatic speech recognition (ASR) with language models to benefit from language technology. Even if low-resource languages possess a writing system, ASR for these languages remains unreliable due to limited bimodal speech and text training data. Better SLU can strengthen the robustness of massively multilingual ASR by levering language semantics to disambiguate utterances via context or exploiting semantic similarities across languages. However, the evaluation of multilingual SLU remains limited to shallow tasks such as intent classification or language identification. To address this, we present Fleurs-SLU, a multilingual SLU benchmark that encompasses (i) 692 hours of speech for topical utterance classification in 102 languages and (ii) multiple-choice question answering through listening comprehension spanning 944 hours of speech across 92 languages. We extensively evaluate both end-to-end speech classification models and cascaded systems that combine speech-to-text transcription with subsequent classification by large language models on Fleurs-SLU. Our results show that cascaded systems exhibit greater robustness in multilingual SLU tasks, though speech encoders can achieve competitive performance in topical speech classification when appropriately pre-trained. We further find a strong correlation between robust multilingual ASR, effective speech-to-text translation, and strong multilingual SLU, highlighting the mutual benefits between acoustic and semantic speech representations.
Authors: Suvodip Dey, Yi-Jyun Sun, Gokhan Tur, Dilek Hakkani-Tur
Abstract: Recent LLMs have enabled significant advancements for conversational agents. However, they are also well known to hallucinate, producing responses that seem plausible but are factually incorrect. On the other hand, users tend to over-rely on LLM-based AI agents, accepting AI's suggestion even when it is wrong. Adding positive friction, such as explanations or getting user confirmations, has been proposed as a mitigation in AI-supported decision-making systems. In this paper, we propose an accountability model for LLM-based task-oriented dialogue agents to address user overreliance via friction turns in cases of model uncertainty and errors associated with dialogue state tracking (DST). The accountability model is an augmented LLM with an additional accountability head that functions as a binary classifier to predict the relevant slots of the dialogue state mentioned in the conversation. We perform our experiments with multiple backbone LLMs on two established benchmarks (MultiWOZ and Snips). Our empirical findings demonstrate that the proposed approach not only enables reliable estimation of AI agent errors but also guides the decoder in generating more accurate actions. We observe around 3% absolute improvement in joint goal accuracy (JGA) of DST output by incorporating accountability heads into modern LLMs. Self-correcting the detected errors further increases the JGA from 67.13 to 70.51, achieving state-of-the-art DST performance. Finally, we show that error correction through user confirmations (friction turn) achieves a similar performance gain, highlighting its potential to reduce user overreliance.
Authors: Yantao Liu, Zijun Yao, Rui Min, Yixin Cao, Lei Hou, Juanzi Li
Abstract: Best-of-N (BoN) sampling, a common strategy for test-time scaling of Large Language Models (LLMs), relies on reward models to select the best candidate solution from multiple generations. However, traditional reward models often assign arbitrary and inconsistent scores, limiting their effectiveness. To address this, we propose a Pairwise Judge Reward Model (PariJudge RM) combined with a knockout tournament for BoN sampling. Instead of assigning absolute scores, given one math problem, PariJudge RM judges two candidate solutions' correctness with chain-of-thought reasoning simultaneously. This approach eliminates the need for scoring and enables cross-validation of solutions through parallel judgment. In the knockout tournament, PariJudge RM conducts pairwise Judgment between candidate solutions and eliminates the incorrect ones iteratively. We construct PairJudge-432K, a large-scale dataset of 432K pairwise judgments derived from NumiaMath and annotated using \texttt{gemini-1.5-flash}, and train the PariJudge RM via supervised fine-tuning. Experiments on MATH-500 and the Olympiad Bench demonstrate significant improvements over baseline reward models. And a 40\% to 60\% relative improvement is achieved on the top 50\% challenging problems.
Authors: Ayan Sengupta, Siddhant Chaudhary, Tanmoy Chakraborty
Abstract: The ever-increasing size of large language models (LLMs) presents significant challenges for deployment due to their heavy computational and memory requirements. Current model pruning techniques attempt to alleviate these issues by relying heavily on external calibration datasets to determine which parameters to prune or compress, thus limiting their flexibility and scalability across different compression ratios. Moreover, these methods often cause severe performance degradation, particularly in downstream tasks, when subjected to higher compression rates. In this paper, we propose PruneNet, a novel model compression method that addresses these limitations by reformulating model pruning as a policy learning process. PruneNet decouples the pruning process from the model architecture, eliminating the need for calibration datasets. It learns a stochastic pruning policy to assess parameter importance solely based on intrinsic model properties while preserving the spectral structure to minimize information loss. PruneNet can compress the LLaMA-2-7B model in just 15 minutes, achieving over 80% retention of its zero-shot performance with a 30% compression ratio, outperforming existing methods that retain only 75% performance. Furthermore, on complex multitask language understanding tasks, PruneNet demonstrates its robustness by preserving up to 80% performance of the original model, proving itself a superior alternative to conventional structured compression techniques.
Authors: Yong Cao, Haijiang Liu, Arnav Arora, Isabelle Augenstein, Paul R\"ottger, Daniel Hershcovich
Abstract: Large-scale surveys are essential tools for informing social science research and policy, but running surveys is costly and time-intensive. If we could accurately simulate group-level survey results, this would therefore be very valuable to social science research. Prior work has explored the use of large language models (LLMs) for simulating human behaviors, mostly through prompting. In this paper, we are the first to specialize LLMs for the task of simulating survey response distributions. As a testbed, we use country-level results from two global cultural surveys. We devise a fine-tuning method based on first-token probabilities to minimize divergence between predicted and actual response distributions for a given question. Then, we show that this method substantially outperforms other methods and zero-shot classifiers, even on unseen questions, countries, and a completely unseen survey. While even our best models struggle with the task, especially on unseen questions, our results demonstrate the benefits of specialization for simulation, which may accelerate progress towards sufficiently accurate simulation in the future.
Authors: Zican Dong, Junyi Li, Jinhao Jiang, Mingyu Xu, Wayne Xin Zhao, Bingning Wang, Weipeng Chen
Abstract: Large language models (LLMs) have gained extended context windows through scaling positional encodings and lightweight continual pre-training. However, this often leads to degraded performance on short-text tasks, while the reasons for this degradation remain insufficiently explored. In this work, we identify two primary factors contributing to this issue: distribution drift in hidden states and attention scores, and catastrophic forgetting during continual pre-training. To address these challenges, we propose Long Context Pre-training with Restoration Distillation (LongReD), a novel approach designed to mitigate short-text performance degradation through minimizing the distribution discrepancy between the extended and original models. Besides training on long texts, LongReD distills the hidden state of selected layers from the original model on short texts. Additionally, LongReD also introduces a short-to-long distillation, aligning the output distribution on short texts with that on long texts by leveraging skipped positional indices. Experiments on common text benchmarks demonstrate that LongReD effectively preserves the model's short-text performance while maintaining comparable or even better capacity to handle long texts than baselines. Our code is available at https://github.com/RUCAIBox/LongReD.
Authors: Andrianos Michail, Corina Julia Racl\'e, Juri Opitz, Simon Clematide
Abstract: The growing volume of digitized historical texts requires effective semantic search using text embeddings. However, pre-trained multilingual models, typically evaluated on contemporary texts, face challenges with historical digitized content due to OCR noise and outdated spellings. We explore the use of multilingual embeddings for cross-lingual semantic search on historical Luxembourgish, a low-resource language. We collect historical Luxembourgish news articles spanning various time periods and use GPT-4o to segment and translate them into closely related languages, creating 20,000 parallel training sentences per language pair. We further create a historical bitext mining evaluation set and find that these models struggle to perform cross-lingual search on historical Luxembourgish. To address this, we propose a simple adaptation method using in-domain training data, achieving up to 98\% accuracy in cross-lingual evaluations. We release our adapted models and historical Luxembourgish-German/French bitexts to support further research.
Authors: Andrianos Michail, Simon Clematide, Rico Sennrich
Abstract: The evaluation of cross-lingual semantic search capabilities of models is often limited to existing datasets from tasks such as information retrieval and semantic textual similarity. To allow for domain-specific evaluation, we introduce Cross Lingual Semantic Discrimination (CLSD), a novel cross-lingual semantic search task that requires only a set of parallel sentence pairs of the language pair of interest within the target domain. This task focuses on the ability of a model to cross-lingually rank the true parallel sentence higher than hard negatives generated by a large language model. We create four instances of our introduced CLSD task for the language pair German-French within the domain of news. Within this case study, we find that models that are also fine-tuned for retrieval tasks (e.g., multilingual E5) benefit from using English as the pivot language, while bitext mining models such as LaBSE perform best directly cross-lingually. We also show a fine-grained similarity analysis enabled by our distractor generation strategy, indicating that different embedding models are sensitive to different types of perturbations.
Authors: Huiyao Chen, Meishan Zhang, Jing Li, Min Zhang, Lilja {\O}vrelid, Jan Haji\v{c}, Hao Fei
Abstract: Semantic role labeling (SRL) is a central natural language processing (NLP) task aiming to understand the semantic roles within texts, facilitating a wide range of downstream applications. While SRL has garnered extensive and enduring research, there is currently a lack of a comprehensive survey that thoroughly organizes and synthesizes the field. This paper aims to review the entire research trajectory of the SRL community over the past two decades. We begin by providing a complete definition of SRL. To offer a comprehensive taxonomy, we categorize SRL methodologies into four key perspectives: model architectures, syntax feature modeling, application scenarios, and multi-modal extensions. Further, we discuss SRL benchmarks, evaluation metrics, and paradigm modeling approaches, while also exploring practical applications across various domains. Finally, we analyze future research directions in SRL, addressing the evolving role of SRL in the age of large language models (LLMs) and its potential impact on the broader NLP landscape. We maintain a public repository and consistently update related resources at: https://github.com/DreamH1gh/Awesome-SRL
Authors: Zonghao Ying, Deyue Zhang, Zonglei Jing, Yisong Xiao, Quanchen Zou, Aishan Liu, Siyuan Liang, Xiangzheng Zhang, Xianglong Liu, Dacheng Tao
Abstract: Multi-turn jailbreak attacks simulate real-world human interactions by engaging large language models (LLMs) in iterative dialogues, exposing critical safety vulnerabilities. However, existing methods often struggle to balance semantic coherence with attack effectiveness, resulting in either benign semantic drift or ineffective detection evasion. To address this challenge, we propose Reasoning-Augmented Conversation, a novel multi-turn jailbreak framework that reformulates harmful queries into benign reasoning tasks and leverages LLMs' strong reasoning capabilities to compromise safety alignment. Specifically, we introduce an attack state machine framework to systematically model problem translation and iterative reasoning, ensuring coherent query generation across multiple turns. Building on this framework, we design gain-guided exploration, self-play, and rejection feedback modules to preserve attack semantics, enhance effectiveness, and sustain reasoning-driven attack progression. Extensive experiments on multiple LLMs demonstrate that RACE achieves state-of-the-art attack effectiveness in complex conversational scenarios, with attack success rates (ASRs) increasing by up to 96%. Notably, our approach achieves ASRs of 82% and 92% against leading commercial models, OpenAI o1 and DeepSeek R1, underscoring its potency. We release our code at https://github.com/NY1024/RACE to facilitate further research in this critical domain.
Authors: Jihao Gu, Yingyao Wang, Pi Bu, Chen Wang, Ziming Wang, Tengtao Song, Donglai Wei, Jiale Yuan, Yingxiu Zhao, Yancheng He, Shilong Li, Jiaheng Liu, Meng Cao, Jun Song, Yingshui Tan, Xiang Li, Wenbo Su, Zhicheng Zheng, Xiaoyong Zhu, Bo Zheng
Abstract: The evaluation of factual accuracy in large vision language models (LVLMs) has lagged behind their rapid development, making it challenging to fully reflect these models' knowledge capacity and reliability. In this paper, we introduce the first factuality-based visual question-answering benchmark in Chinese, named ChineseSimpleVQA, aimed at assessing the visual factuality of LVLMs across 8 major topics and 56 subtopics. The key features of this benchmark include a focus on the Chinese language, diverse knowledge types, a multi-hop question construction, high-quality data, static consistency, and easy-to-evaluate through short answers. Moreover, we contribute a rigorous data construction pipeline and decouple the visual factuality into two parts: seeing the world (i.e., object recognition) and discovering knowledge. This decoupling allows us to analyze the capability boundaries and execution mechanisms of LVLMs. Subsequently, we evaluate 34 advanced open-source and closed-source models, revealing critical performance gaps within this field.
Authors: Shuqi Liu, Han Wu, Bowei He, Xiongwei Han, Mingxuan Yuan, Linqi Song
Abstract: Recent advances in large language models have led to numerous task-specialized fine-tuned variants, creating a need for efficient model merging techniques that preserve specialized capabilities while avoiding costly retraining. While existing task vector-based merging methods show promise, they typically apply uniform coefficients across all parameters, overlooking varying parameter importance both within and across tasks. We present Sens-Merging, a sensitivity-guided coefficient adjustment method that enhances existing model merging techniques by operating at both task-specific and cross-task levels. Our method analyzes parameter sensitivity within individual tasks and evaluates cross-task transferability to determine optimal merging coefficients. Extensive experiments on Mistral 7B and LLaMA2-7B/13B models demonstrate that Sens-Merging significantly improves performance across general knowledge, mathematical reasoning, and code generation tasks. Notably, when combined with existing merging techniques, our method enables merged models to outperform specialized fine-tuned models, particularly in code generation tasks. Our findings reveal important trade-offs between task-specific and cross-task scalings, providing insights for future model merging strategies.
Authors: Kaiyang Wan, Honglin Mu, Rui Hao, Haoran Luo, Tianle Gu, Xiuying Chen
Abstract: Like humans, Large Language Models (LLMs) struggle to generate high-quality long-form text that adheres to strict requirements in a single pass. This challenge is unsurprising, as successful human writing, according to the Cognitive Writing Theory, is a complex cognitive process involving iterative planning, translating, reviewing, and monitoring. Motivated by these cognitive principles, we aim to equip LLMs with human-like cognitive writing capabilities through CogWriter, a novel training-free framework that transforms LLM constrained long-form text generation into a systematic cognitive writing paradigm. Our framework consists of two key modules: (1) a Planning Agent that performs hierarchical planning to decompose the task, and (2) multiple Generation Agents that execute these plans in parallel. The system maintains quality via continuous monitoring and reviewing mechanisms, which evaluate outputs against specified requirements and trigger necessary revisions. CogWriter demonstrates exceptional performance on LongGenBench, a benchmark for complex constrained long-form text generation. Even when using Qwen-2.5-14B as its backbone, CogWriter surpasses GPT-4o by 22% in complex instruction completion accuracy while reliably generating texts exceeding 10,000 words. We hope this cognitive science-inspired approach provides a paradigm for LLM writing advancements: \href{https://github.com/KaiyangWan/CogWriter}{CogWriter}.
Authors: Ben Liu, Jihan Zhang, Fangquan Lin, Xu Jia, Min Peng
Abstract: Large language models (LLMs) have been increasingly employed in various intelligent educational systems, simulating human tutors to facilitate effective human-machine interaction. However, previous studies often overlook the significance of recognizing and adapting to individual learner characteristics. Such adaptation is crucial for enhancing student engagement and learning efficiency, particularly in mathematics instruction, where diverse learning styles require personalized strategies to promote comprehension and enthusiasm. In this paper, we propose a \textbf{P}erson\textbf{A}lized \textbf{C}onversational tutoring ag\textbf{E}nt (PACE) for mathematics instruction. PACE simulates students' learning styles based on the Felder and Silverman learning style model, aligning with each student's persona. In this way, our PACE can effectively assess the personality of students, allowing to develop individualized teaching strategies that resonate with their unique learning styles. To further enhance students' comprehension, PACE employs the Socratic teaching method to provide instant feedback and encourage deep thinking. By constructing personalized teaching data and training models, PACE demonstrates the ability to identify and adapt to the unique needs of each student, significantly improving the overall learning experience and outcomes. Moreover, we establish multi-aspect evaluation criteria and conduct extensive analysis to assess the performance of personalized teaching. Experimental results demonstrate the superiority of our model in personalizing the educational experience and motivating students compared to existing methods.
Authors: Shengxiang Gao, Jey Han Lau, Jianzhong Qi
Abstract: Knowledge base question answering (KBQA) aims to answer user questions in natural language using rich human knowledge stored in large KBs. As current KBQA methods struggle with unseen knowledge base elements at test time,we introduce SG-KBQA: a novel model that injects schema contexts into entity retrieval and logical form generation to tackle this issue. It uses the richer semantics and awareness of the knowledge base structure provided by schema contexts to enhance generalizability. We show that SG-KBQA achieves strong generalizability, outperforming state-of-the-art models on two commonly used benchmark datasets across a variety of test settings. Our source code is available at https://github.com/gaosx2000/SG_KBQA.
Authors: Rubing Li, Jo\~ao Sedoc, Arun Sundararajan
Abstract: When encountering increasingly frequent performance improvements or cost reductions from a new large language model (LLM), developers of applications leveraging LLMs must decide whether to take advantage of these improvements or stay with older tried-and-tested models. Low perceived switching frictions can lead to choices that do not consider more subtle behavior changes that the transition may induce. Our experiments use a popular game-theoretic behavioral economics model of trust to show stark differences in the trusting behavior of OpenAI's and DeepSeek's models. We highlight a collapse in the economic trust behavior of the o1-mini and o3-mini models as they reconcile profit-maximizing and risk-seeking with future returns from trust, and contrast it with DeepSeek's more sophisticated and profitable trusting behavior that stems from an ability to incorporate deeper concepts like forward planning and theory-of-mind. As LLMs form the basis for high-stakes commercial systems, our results highlight the perils of relying on LLM performance benchmarks that are too narrowly defined and suggest that careful analysis of their hidden fault lines should be part of any organization's AI strategy.
Authors: Junchen Fu, Xuri Ge, Kaiwen Zheng, Ioannis Arapakis, Xin Xin, Joemon M. Jose
Abstract: Popular Micro-videos, dominant on platforms like TikTok and YouTube, hold significant commercial value. The rise of high-quality AI-generated content has spurred interest in AI-driven micro-video creation. However, despite the advanced capabilities of large language models (LLMs) like ChatGPT and DeepSeek in text generation and reasoning, their potential to assist the creation of popular micro-videos remains largely unexplored. In this paper, we conduct an empirical study on LLM-assisted popular micro-video generation (LLMPopcorn). Specifically, we investigate the following research questions: (i) How can LLMs be effectively utilized to assist popular micro-video generation? (ii) To what extent can prompt-based enhancements optimize the LLM-generated content for higher popularity? (iii) How well do various LLMs and video generators perform in the popular micro-video generation task? By exploring these questions, we show that advanced LLMs like DeepSeek-V3 enable micro-video generation to achieve popularity comparable to human-created content. Prompt enhancements further boost popularity, and benchmarking highlights DeepSeek-V3 and DeepSeek-R1 among LLMs, while LTX-Video and HunyuanVideo lead in video generation. This pioneering work advances AI-assisted micro-video creation, uncovering new research opportunities. We will release the code and datasets to support future studies.
Authors: Narun Raman, Taylor Lundy, Thiago Amin, Jesse Perla, Kevin Leyton-Brown
Abstract: How should one judge whether a given large language model (LLM) can reliably perform economic reasoning? Most existing LLM benchmarks focus on specific applications and fail to present the model with a rich variety of economic tasks. A notable exception is Raman et al. [2024], who offer an approach for comprehensively benchmarking strategic decision-making; however, this approach fails to address the non-strategic settings prevalent in microeconomics, such as supply-and-demand analysis. We address this gap by taxonomizing microeconomic reasoning into $58$ distinct elements, focusing on the logic of supply and demand, each grounded in up to $10$ distinct domains, $5$ perspectives, and $3$ types. The generation of benchmark data across this combinatorial space is powered by a novel LLM-assisted data generation protocol that we dub auto-STEER, which generates a set of questions by adapting handwritten templates to target new domains and perspectives. Because it offers an automated way of generating fresh questions, auto-STEER mitigates the risk that LLMs will be trained to over-fit evaluation benchmarks; we thus hope that it will serve as a useful tool both for evaluating and fine-tuning models for years to come. We demonstrate the usefulness of our benchmark via a case study on $27$ LLMs, ranging from small open-source models to the current state of the art. We examined each model's ability to solve microeconomic problems across our whole taxonomy and present the results across a range of prompting strategies and scoring metrics.
Authors: Wanqing Cui, Rui Cheng, Jiafeng Guo, Xueqi Cheng
Abstract: Existing two-stream models, such as CLIP, encode images and text through independent representations, showing good performance while ensuring retrieval speed, have attracted attention from industry and academia. However, the single representation often struggles to capture complex content fully. Such models may ignore fine-grained information during matching, resulting in suboptimal retrieval results. To overcome this limitation and enhance the performance of two-stream models, we propose a Multi-view Attention Method (MVAM) for image-text matching. This approach leverages diverse attention heads with unique view codes to learn multiple representations for images and text, which are then concatenated for matching. We also incorporate a diversity objective to explicitly encourage attention heads to focus on distinct aspects of the input data, capturing complementary fine-grained details. This diversity enables the model to represent image-text pairs from multiple perspectives, ensuring a more comprehensive understanding and alignment of critical content. Our method allows models to encode images and text from different perspectives and focus on more critical details, leading to better matching performance. Our experiments on MSCOCO and Flickr30K demonstrate enhancements over existing models, and further case studies reveal that different attention heads can focus on distinct content, achieving more comprehensive representations.
Authors: Yuepeng Hu, Zhengyuan Jiang, Moyang Guo, Neil Zhenqiang Gong
Abstract: Watermark has been widely deployed by industry to detect AI-generated images. The robustness of such watermark-based detector against evasion attacks in the white-box and black-box settings is well understood in the literature. However, the robustness in the no-box setting is much less understood. In this work, we propose a new transfer evasion attack to image watermark in the no-box setting. Our transfer attack adds a perturbation to a watermarked image to evade multiple surrogate watermarking models trained by the attacker itself, and the perturbed watermarked image also evades the target watermarking model. Our major contribution is to show that, both theoretically and empirically, watermark-based AI-generated image detector based on existing watermarking methods is not robust to evasion attacks even if the attacker does not have access to the watermarking model nor the detection API. Our code is available at: https://github.com/hifi-hyp/Watermark-Transfer-Attack.
URLs: https://github.com/hifi-hyp/Watermark-Transfer-Attack.
Authors: Jiang Li, Xiangdong Su, Guanglai Gao
Abstract: Recent studies have highlighted the effectiveness of tensor decomposition methods in the Temporal Knowledge Graphs Embedding (TKGE) task. However, we found that inherent heterogeneity among factor tensors in tensor decomposition significantly hinders the tensor fusion process and further limits the performance of link prediction. To overcome this limitation, we introduce a novel method that maps factor tensors onto a unified smooth Lie group manifold to make the distribution of factor tensors approximating homogeneous in tensor decomposition. We provide the theoretical proof of our motivation that homogeneous tensors are more effective than heterogeneous tensors in tensor fusion and approximating the target for tensor decomposition based TKGE methods. The proposed method can be directly integrated into existing tensor decomposition based TKGE methods without introducing extra parameters. Extensive experiments demonstrate the effectiveness of our method in mitigating the heterogeneity and in enhancing the tensor decomposition based TKGE models.
Authors: Yuan Sui, Yufei He, Nian Liu, Xiaoxin He, Kun Wang, Bryan Hooi
Abstract: Large language models (LLMs) are often challenged by generating erroneous or hallucinated responses, especially in complex reasoning tasks. Leveraging knowledge graphs (KGs) as external knowledge sources has emerged as a viable solution. However, existing KG-enhanced methods, either retrieval-based or agent-based, encounter difficulties in accurately retrieving knowledge and efficiently traversing KGs at scale. In this paper, we propose a unified framework, FiDeLiS, designed to improve the factuality of LLM responses by anchoring answers to verifiable reasoning steps retrieved from a KG. To achieve this, we leverage step-wise beam search with a deductive scoring function, allowing the LLM to validate each reasoning step and halt the search once the question is deducible. In addition, our Path-rag module pre-selects a smaller candidate set for each beam search step, reducing computational costs by narrowing the search space. Extensive experiments show that our training-free and efficient approach outperforms strong baselines, enhancing both factuality and interpretability.
Authors: Chenyan Zhao, Mariana Silva, Seth Poulsen
Abstract: In mathematical proof education, there remains a need for interventions that help students learn to write mathematical proofs. Research has shown that timely feedback can be very helpful to students learning new skills. While for many years natural language processing models have struggled to perform well on tasks related to mathematical texts, recent developments in natural language processing have created the opportunity to complete the task of giving students instant feedback on their mathematical proofs. In this paper, we present a set of training methods and models capable of autograding freeform mathematical proofs by leveraging existing large language models and other machine learning techniques. The models are trained using proof data collected from four different proof by induction problems. We use four different robust large language models to compare their performances, and all achieve satisfactory performances to various degrees. Additionally, we recruit human graders to grade the same proofs as the training data, and find that the best grading model is also more accurate than most human graders. With the development of these grading models, we create and deploy an autograder for proof by induction problems and perform a user study with students. Results from the study shows that students are able to make significant improvements to their proofs using the feedback from the autograder, but students still do not trust the AI autograders as much as they trust human graders. Future work can improve on the autograder feedback and figure out ways to help students trust AI autograders.
Authors: Joe Dhanith P R, Shravan Venkatraman, Vigya Sharma, Santhosh Malarvannan, Modigari Narendra
Abstract: Understanding emotions is a fundamental aspect of human communication. Integrating audio and video signals offers a more comprehensive understanding of emotional states compared to traditional methods that rely on a single data source, such as speech or facial expressions. Despite its potential, multimodal emotion recognition faces significant challenges, particularly in synchronization, feature extraction, and fusion of diverse data sources. To address these issues, this paper introduces a novel transformer-based model named Audio-Video Transformer Fusion with Cross Attention (AVT-CA). The AVT-CA model employs a transformer fusion approach to effectively capture and synchronize interlinked features from both audio and video inputs, thereby resolving synchronization problems. Additionally, the Cross Attention mechanism within AVT-CA selectively extracts and emphasizes critical features while discarding irrelevant ones from both modalities, addressing feature extraction and fusion challenges. Extensive experimental analysis conducted on the CMU-MOSEI, RAVDESS and CREMA-D datasets demonstrates the efficacy of the proposed model. The results underscore the importance of AVT-CA in developing precise and reliable multimodal emotion recognition systems for practical applications.
Authors: Haoyang Su, Renqi Chen, Shixiang Tang, Zhenfei Yin, Xinzhe Zheng, Jinzhe Li, Biqing Qi, Qi Wu, Hui Li, Wanli Ouyang, Philip Torr, Bowen Zhou, Nanqing Dong
Abstract: The rapid advancement of scientific progress requires innovative tools that can accelerate knowledge discovery. Although recent AI methods, particularly large language models (LLMs), have shown promise in tasks such as hypothesis generation and experimental design, they fall short of replicating the collaborative nature of real-world scientific practices, where diverse experts work together in teams to tackle complex problems. To address the limitations, we propose an LLM-based multi-agent system, i.e., Virtual Scientists (VirSci), designed to mimic the teamwork inherent in scientific research. VirSci organizes a team of agents to collaboratively generate, evaluate, and refine research ideas. Through comprehensive experiments, we demonstrate that this multi-agent approach outperforms the state-of-the-art method in producing novel scientific ideas. We further investigate the collaboration mechanisms that contribute to its tendency to produce ideas with higher novelty, offering valuable insights to guide future research and illuminating pathways toward building a robust system for autonomous scientific discovery. The code is available at https://github.com/open-sciencelab/Virtual-Scientists.
URLs: https://github.com/open-sciencelab/Virtual-Scientists.
Authors: Jingyang Qiao, Zhizhong Zhang, Xin Tan, Yanyun Qu, Shouhong Ding, Yuan Xie
Abstract: Continual Instruction Tuning (CIT) is adopted to continually instruct Large Models to follow human intent data by data. It is observed that existing gradient update would heavily destroy the performance on previous datasets during CIT process. Instead, Exponential Moving Average (EMA), owns the ability to trace previous parameters, which can aid in decreasing forgetting. Nonetheless, its stable balance weight fails to deal with the ever-changing datasets, leading to the out-of-balance between plasticity and stability. In this paper, we propose a general continual instruction tuning framework to address the challenge. Starting from the trade-off prerequisite and EMA update, we propose the plasticity and stability ideal condition. Based on Taylor expansion in the loss function, we find the optimal balance weight can be automatically determined by the gradients and learned parameters. Therefore, we propose a stable-plasticity balanced coefficient to avoid knowledge confusion. Based on the semantic similarity of the instructions, we can determine whether to retrain or expand the training parameters and allocate the most suitable parameters for the testing instances. Extensive experiments across multiple continual instruction tuning benchmarks demonstrate that our approach not only enhances anti-forgetting capabilities but also significantly improves overall continual tuning performance. For example, based on LLaVA-7B, the forgetting is reduced from 5.42 to 1.93. Our code will be made publicly available soon.
Authors: Hyejun Jeong, Shiqing Ma, Amir Houmansadr
Abstract: Bias in machine learning models, particularly in Large Language Models, is a critical issue as these systems shape important societal decisions. While previous studies have examined bias in individual LLMs, comparisons of bias across models remain underexplored. To address this gap, we analyze 13 LLMs from five families, evaluating bias through output distribution across multiple dimensions using two datasets (4K and 1M questions). Our results show that fine-tuning has minimal impact on output distributions, and proprietary models tend to overly response as unknowns to minimize bias, compromising accuracy and utility. In addition, open-source models like Llama3-Chat and Gemma2-it demonstrate fairness comparable to proprietary models like GPT-4, challenging the assumption that larger, closed-source models are inherently less biased. We also find that bias scores for disambiguated questions are more extreme, raising concerns about reverse discrimination. These findings highlight the need for improved bias mitigation strategies and more comprehensive evaluation metrics for fairness in LLMs.
Authors: Kyungmin Min, Minbeom Kim, Kang-il Lee, Dongryeol Lee, Kyomin Jung
Abstract: Large Vision-Language Models (LVLMs) demonstrate impressive capabilities in generating detailed and coherent responses from visual inputs. However, they are prone to generate hallucinations due to an over-reliance on language priors. To address this issue, we investigate the language priors in LVLMs and make two key observations: (1) Even when predicting the tokens associated with image-related part-of-speech (POS), models increasingly rely on linguistic priors as the token sequences grow, thereby amplifying hallucinations. (2) Methods that directly calibrate LVLM's output distribution to mitigate language priors can lead to a degradation in text quality or even exacerbate hallucinations. Based on these findings, we propose a novel method, Summary-Guided Decoding (SumGD). This method naturally encourages the model to focus more on image information by reducing the text context through summaries, while controlling only the image-related POS tokens to maintain text quality. Through experiments, we demonstrate that SumGD achieves state-of-the-art performance on object hallucination benchmarks. Furthermore, in terms of the trade-off between precision and recall, SumGD achieves Pareto optimality among the existing methods. Lastly, we observe that although existing methods struggle to balance the reduction of object hallucinations with maintaining text quality, SumGD demonstrates robustness in handling this challenge.
Authors: Michal Shlapentokh-Rothman, Yu-Xiong Wang, Derek Hoiem
Abstract: For users with limited computational resources, visual programming or prompting large language models (LLMs) to generate executable code for visual tasks, like visual question answering (VQA), remains largely inaccessible. Even with techniques such as distillation, adapting visual programming to smaller models or specific datasets is still quite challenging due to high annotation costs. We propose a low-cost visual program distillation method that can be used for models with fewer than 1 billion parameters and requires no human-generated program annotations. We achieve this through synthetic data augmentation based on decoupling programs into higher-level skills, called templates, and their corresponding arguments. Experimental results show that, with a relatively small amount of question/answer data, small language models can generate high-quality visual programs with the added benefit of much faster inference.
Authors: Yuzhuang Xu, Shiyu Ji, Qingfu Zhu, Wanxiang Che
Abstract: Powerful large language models (LLMs) are increasingly expected to be deployed with lower computational costs, enabling their capabilities on resource-constrained devices. Post-training quantization (PTQ) has emerged as a star approach to achieve this ambition, with best methods compressing weights to less than 2 bit on average. In this paper, we propose Channel-Relaxed Vector Quantization (CRVQ), a novel technique that significantly improves the performance of PTQ baselines at the cost of only minimal additional bits. This state-of-the-art extreme compression method achieves its results through two key innovations: (1) carefully selecting and reordering a very small subset of critical weight channels, and (2) leveraging extended codebooks to relax the constraint of critical channels. With our method, we demonstrate a 38.9\% improvement over the current strongest sub-2-bit PTQ baseline, enabling nearer lossless 1-bit compression. Furthermore, our approach offers flexible customization of quantization bit-width and performance, providing a wider range of deployment options for diverse hardware platforms.
Authors: Shayne Longpre, Nikhil Singh, Manuel Cherep, Kushagra Tiwary, Joanna Materzynska, William Brannon, Robert Mahari, Naana Obeng-Marnu, Manan Dey, Mohammed Hamdy, Nayan Saxena, Ahmad Mustafa Anis, Emad A. Alghamdi, Vu Minh Chien, Da Yin, Kun Qian, Yizhi Li, Minnie Liang, An Dinh, Shrestha Mohanty, Deividas Mataciunas, Tobin South, Jianguo Zhang, Ariel N. Lee, Campbell S. Lund, Christopher Klamm, Damien Sileo, Diganta Misra, Enrico Shippole, Kevin Klyman, Lester JV Miranda, Niklas Muennighoff, Seonghyeon Ye, Seungone Kim, Vipul Gupta, Vivek Sharma, Xuhui Zhou, Caiming Xiong, Luis Villa, Stella Biderman, Alex Pentland, Sara Hooker, Jad Kabbara
Abstract: Progress in AI is driven largely by the scale and quality of training data. Despite this, there is a deficit of empirical analysis examining the attributes of well-established datasets beyond text. In this work we conduct the largest and first-of-its-kind longitudinal audit across modalities--popular text, speech, and video datasets--from their detailed sourcing trends and use restrictions to their geographical and linguistic representation. Our manual analysis covers nearly 4000 public datasets between 1990-2024, spanning 608 languages, 798 sources, 659 organizations, and 67 countries. We find that multimodal machine learning applications have overwhelmingly turned to web-crawled, synthetic, and social media platforms, such as YouTube, for their training sets, eclipsing all other sources since 2019. Secondly, tracing the chain of dataset derivations we find that while less than 33% of datasets are restrictively licensed, over 80% of the source content in widely-used text, speech, and video datasets, carry non-commercial restrictions. Finally, counter to the rising number of languages and geographies represented in public AI training datasets, our audit demonstrates measures of relative geographical and multilingual representation have failed to significantly improve their coverage since 2013. We believe the breadth of our audit enables us to empirically examine trends in data sourcing, restrictions, and Western-centricity at an ecosystem-level, and that visibility into these questions are essential to progress in responsible AI. As a contribution to ongoing improvements in dataset transparency and responsible use, we release our entire multimodal audit, allowing practitioners to trace data provenance across text, speech, and video.
Authors: Zijie Chen, Zhanchao Zhou, Yu Lu, Renjun Xu, Lili Pan, Zhenzhong Lan
Abstract: Solving NP-hard problems traditionally relies on heuristics, yet manually designing effective heuristics for complex problems remains a significant challenge. While recent advancements like FunSearch have shown that large language models (LLMs) can be integrated into evolutionary algorithms (EAs) for heuristic design, their potential is hindered by limitations in balancing exploitation and exploration. We introduce Quality-Uncertainty Balanced Evolution (QUBE), a novel approach that enhances LLM+EA methods by redefining the priority criterion within the FunSearch framework. QUBE employs the Quality-Uncertainty Trade-off Criterion (QUTC), based on our proposed Uncertainty-Inclusive Quality metric, to evaluate and guide the evolutionary process. Through extensive experiments on challenging NP-complete problems, QUBE demonstrates significant performance improvements over FunSearch and baseline methods. Our code are available at https://github.com/zzjchen/QUBE\_code.
Authors: Long Phan (Michael Pokorny), Alice Gatti (Michael Pokorny), Ziwen Han (Michael Pokorny), Nathaniel Li (Michael Pokorny), Josephina Hu (Michael Pokorny), Hugh Zhang (Michael Pokorny), Chen Bo Calvin Zhang (Michael Pokorny), Mohamed Shaaban (Michael Pokorny), John Ling (Michael Pokorny), Sean Shi (Michael Pokorny), Michael Choi (Michael Pokorny), Anish Agrawal (Michael Pokorny), Arnav Chopra (Michael Pokorny), Adam Khoja (Michael Pokorny), Ryan Kim (Michael Pokorny), Richard Ren (Michael Pokorny), Jason Hausenloy (Michael Pokorny), Oliver Zhang (Michael Pokorny), Mantas Mazeika (Michael Pokorny), Tung Nguyen (Michael Pokorny), Daron Anderson (Michael Pokorny), Imad Ali Shah (Michael Pokorny), Mikhail Doroshenko (Michael Pokorny), Alun Cennyth Stokes (Michael Pokorny), Mobeen Mahmood (Michael Pokorny), Jaeho Lee (Michael Pokorny), Oleksandr Pokutnyi (Michael Pokorny), Oleg Iskra (Michael Pokorny), Jessica P. Wang (Michael Pokorny), Robert Gerbicz (Michael Pokorny), John-Clark Levin (Michael Pokorny), Serguei Popov (Michael Pokorny), Fiona Feng (Michael Pokorny), Steven Y. Feng (Michael Pokorny), Haoran Zhao (Michael Pokorny), Michael Yu (Michael Pokorny), Varun Gangal (Michael Pokorny), Chelsea Zou (Michael Pokorny), Zihan Wang (Michael Pokorny), Mstyslav Kazakov (Michael Pokorny), Geoff Galgon (Michael Pokorny), Johannes Schmitt (Michael Pokorny), Alvaro Sanchez (Michael Pokorny), Yongki Lee (Michael Pokorny), Will Yeadon (Michael Pokorny), Scott Sauers (Michael Pokorny), Marc Roth (Michael Pokorny), Chidozie Agu (Michael Pokorny), S{\o}ren Riis (Michael Pokorny), Fabian Giska (Michael Pokorny), Saiteja Utpala (Michael Pokorny), Antrell Cheatom (Michael Pokorny), Zachary Giboney (Michael Pokorny), Gashaw M. Goshu (Michael Pokorny), Sarah-Jane Crowson (Michael Pokorny), Mohinder Maheshbhai Naiya (Michael Pokorny), Noah Burns (Michael Pokorny), Lennart Finke (Michael Pokorny), Zerui Cheng (Michael Pokorny), Hyunwoo Park (Michael Pokorny), Francesco Fournier-Facio (Michael Pokorny), Jennifer Zampese (Michael Pokorny), John Wydallis (Michael Pokorny), John B. Wydallis (Michael Pokorny), Ryan G. Hoerr (Michael Pokorny), Mark Nandor (Michael Pokorny), Tim Gehrunger (Michael Pokorny), Jiaqi Cai (Michael Pokorny), Ben McCarty (Michael Pokorny), Jungbae Nam (Michael Pokorny), Edwin Taylor (Michael Pokorny), Jun Jin (Michael Pokorny), Gautier Abou Loume (Michael Pokorny), Hangrui Cao (Michael Pokorny), Alexis C Garretson (Michael Pokorny), Damien Sileo (Michael Pokorny), Qiuyu Ren (Michael Pokorny), Doru Cojoc (Michael Pokorny), Pavel Arkhipov (Michael Pokorny), Usman Qazi (Michael Pokorny), Aras Bacho (Michael Pokorny), Lianghui Li (Michael Pokorny), Sumeet Motwani (Michael Pokorny), Christian Schroeder de Witt (Michael Pokorny), Alexei Kopylov (Michael Pokorny), Johannes Veith (Michael Pokorny), Eric Singer (Michael Pokorny), Paolo Rissone (Michael Pokorny), Jaehyeok Jin (Michael Pokorny), Jack Wei Lun Shi (Michael Pokorny), Chris G. Willcocks (Michael Pokorny), Ameya Prabhu (Michael Pokorny), Longke Tang (Michael Pokorny), Kevin Zhou (Michael Pokorny), Emily de Oliveira Santos (Michael Pokorny), Andrey Pupasov Maksimov (Michael Pokorny), Edward Vendrow (Michael Pokorny), Kengo Zenitani (Michael Pokorny), Joshua Robinson (Michael Pokorny), Aleksandar Mikov (Michael Pokorny), Julien Guillod (Michael Pokorny), Yuqi Li (Michael Pokorny), Ben Pageler (Michael Pokorny), Joshua Vendrow (Michael Pokorny), Vladyslav Kuchkin (Michael Pokorny), Pierre Marion (Michael Pokorny), Denis Efremov (Michael Pokorny), Jayson Lynch (Michael Pokorny), Kaiqu Liang (Michael Pokorny), Andrew Gritsevskiy (Michael Pokorny), Dakotah Martinez (Michael Pokorny), Nick Crispino (Michael Pokorny), Dimitri Zvonkine (Michael Pokorny), Natanael Wildner Fraga (Michael Pokorny), Saeed Soori (Michael Pokorny), Ori Press (Michael Pokorny), Henry Tang (Michael Pokorny), Julian Salazar (Michael Pokorny), Sean R. Green (Michael Pokorny), Lina Br\"ussel (Michael Pokorny), Moon Twayana (Michael Pokorny), Aymeric Dieuleveut (Michael Pokorny), T. Ryan Rogers (Michael Pokorny), Wenjin Zhang (Michael Pokorny), Ross Finocchio (Michael Pokorny), Bikun Li (Michael Pokorny), Jinzhou Yang (Michael Pokorny), Arun Rao (Michael Pokorny), Gabriel Loiseau (Michael Pokorny), Mikhail Kalinin (Michael Pokorny), Marco Lukas (Michael Pokorny), Ciprian Manolescu (Michael Pokorny), Nate Stambaugh (Michael Pokorny), Subrata Mishra (Michael Pokorny), Ariel Ghislain Kemogne Kamdoum (Michael Pokorny), Tad Hogg (Michael Pokorny), Alvin Jin (Michael Pokorny), Carlo Bosio (Michael Pokorny), Gongbo Sun (Michael Pokorny), Brian P Coppola (Michael Pokorny), Haline Heidinger (Michael Pokorny), Rafael Sayous (Michael Pokorny), Stefan Ivanov (Michael Pokorny), Joseph M Cavanagh (Michael Pokorny), Jiawei Shen (Michael Pokorny), Joseph Marvin Imperial (Michael Pokorny), Philippe Schwaller (Michael Pokorny), Shaipranesh Senthilkuma (Michael Pokorny), Andres M Bran (Michael Pokorny), Andres Algaba (Michael Pokorny), Brecht Verbeken (Michael Pokorny), Kelsey Van den Houte (Michael Pokorny), Lynn Van Der Sypt (Michael Pokorny), David Noever (Michael Pokorny), Lisa Schut (Michael Pokorny), Ilia Sucholutsky (Michael Pokorny), Evgenii Zheltonozhskii (Michael Pokorny), Qiaochu Yuan (Michael Pokorny), Derek Lim (Michael Pokorny), Richard Stanley (Michael Pokorny), Shankar Sivarajan (Michael Pokorny), Tong Yang (Michael Pokorny), John Maar (Michael Pokorny), Julian Wykowski (Michael Pokorny), Mart\'i Oller (Michael Pokorny), Jennifer Sandlin (Michael Pokorny), Anmol Sahu (Michael Pokorny), Cesare Giulio Ardito (Michael Pokorny), Yuzheng Hu (Michael Pokorny), Felipe Meneguitti Dias (Michael Pokorny), Tobias Kreiman (Michael Pokorny), Kaivalya Rawal (Michael Pokorny), Tobias Garcia Vilchis (Michael Pokorny), Yuexuan Zu (Michael Pokorny), Martin Lackner (Michael Pokorny), James Koppel (Michael Pokorny), Jeremy Nguyen (Michael Pokorny), Daniil S. Antonenko (Michael Pokorny), Steffi Chern (Michael Pokorny), Bingchen Zhao (Michael Pokorny), Pierrot Arsene (Michael Pokorny), Sergey Ivanov (Michael Pokorny), Rafa{\l} Po\'swiata (Michael Pokorny), Chenguang Wang (Michael Pokorny), Daofeng Li (Michael Pokorny), Donato Crisostomi (Michael Pokorny), Ali Dehghan (Michael Pokorny), Andrea Achilleos (Michael Pokorny), John Arnold Ambay (Michael Pokorny), Benjamin Myklebust (Michael Pokorny), Archan Sen (Michael Pokorny), David Perrella (Michael Pokorny), Nurdin Kaparov (Michael Pokorny), Mark H Inlow (Michael Pokorny), Allen Zang (Michael Pokorny), Kalyan Ramakrishnan (Michael Pokorny), Daniil Orel (Michael Pokorny), Vladislav Poritski (Michael Pokorny), Shalev Ben-David (Michael Pokorny), Zachary Berger (Michael Pokorny), Parker Whitfill (Michael Pokorny), Michael Foster (Michael Pokorny), Daniel Munro (Michael Pokorny), Linh Ho (Michael Pokorny), Dan Bar Hava (Michael Pokorny), Aleksey Kuchkin (Michael Pokorny), Robert Lauff (Michael Pokorny), David Holmes (Michael Pokorny), Frank Sommerhage (Michael Pokorny), Anji Zhang (Michael Pokorny), Richard Moat (Michael Pokorny), Keith Schneider (Michael Pokorny), Daniel Pyda (Michael Pokorny), Zakayo Kazibwe (Michael Pokorny), Mukhwinder Singh (Michael Pokorny), Don Clarke (Michael Pokorny), Dae Hyun Kim (Michael Pokorny), Sara Fish (Michael Pokorny), Veit Elser (Michael Pokorny), Victor Efren Guadarrama Vilchis (Michael Pokorny), Immo Klose (Michael Pokorny), Christoph Demian (Michael Pokorny), Ujjwala Anantheswaran (Michael Pokorny), Adam Zweiger (Michael Pokorny), Guglielmo Albani (Michael Pokorny), Jeffery Li (Michael Pokorny), Nicolas Daans (Michael Pokorny), Maksim Radionov (Michael Pokorny), V\'aclav Rozho\v{n} (Michael Pokorny), Vincent Ginis (Michael Pokorny), Ziqiao Ma (Michael Pokorny), Christian Stump (Michael Pokorny), Jacob Platnick (Michael Pokorny), Volodymyr Nevirkovets (Michael Pokorny), Luke Basler (Michael Pokorny), Marco Piccardo (Michael Pokorny), Niv Cohen (Michael Pokorny), Virendra Singh (Michael Pokorny), Josef Tkadlec (Michael Pokorny), Paul Rosu (Michael Pokorny), Alan Goldfarb (Michael Pokorny), Piotr Padlewski (Michael Pokorny), Stanislaw Barzowski (Michael Pokorny), Kyle Montgomery (Michael Pokorny), Aline Menezes (Michael Pokorny), Arkil Patel (Michael Pokorny), Zixuan Wang (Michael Pokorny), Jamie Tucker-Foltz (Michael Pokorny), Jack Stade (Michael Pokorny), Declan Grabb (Michael Pokorny), Tom Goertzen (Michael Pokorny), Fereshteh Kazemi (Michael Pokorny), Jeremiah Milbauer (Michael Pokorny), Abhishek Shukla (Michael Pokorny), Hossam Elgnainy (Michael Pokorny), Yan Carlos Leyva Labrador (Michael Pokorny), Hao He (Michael Pokorny), Ling Zhang (Michael Pokorny), Alan Givr\'e (Michael Pokorny), Hew Wolff (Michael Pokorny), G\"ozdenur Demir (Michael Pokorny), Muhammad Fayez Aziz (Michael Pokorny), Younesse Kaddar (Michael Pokorny), Ivar \"Angquist (Michael Pokorny), Yanxu Chen (Michael Pokorny), Elliott Thornley (Michael Pokorny), Robin Zhang (Michael Pokorny), Jiayi Pan (Michael Pokorny), Antonio Terpin (Michael Pokorny), Niklas Muennighoff (Michael Pokorny), Hailey Schoelkopf (Michael Pokorny), Eric Zheng (Michael Pokorny), Avishy Carmi (Michael Pokorny), Jainam Shah (Michael Pokorny), Ethan D. L. Brown (Michael Pokorny), Kelin Zhu (Michael Pokorny), Max Bartolo (Michael Pokorny), Richard Wheeler (Michael Pokorny), Andrew Ho (Michael Pokorny), Shaul Barkan (Michael Pokorny), Jiaqi Wang (Michael Pokorny), Martin Stehberger (Michael Pokorny), Egor Kretov (Michael Pokorny), Peter Bradshaw (Michael Pokorny), JP Heimonen (Michael Pokorny), Kaustubh Sridhar (Michael Pokorny), Zaki Hossain (Michael Pokorny), Ido Akov (Michael Pokorny), Yury Makarychev (Michael Pokorny), Joanna Tam (Michael Pokorny), Hieu Hoang (Michael Pokorny), David M. Cunningham (Michael Pokorny), Vladimir Goryachev (Michael Pokorny), Demosthenes Patramanis (Michael Pokorny), Michael Krause (Michael Pokorny), Andrew Redenti (Michael Pokorny), David Aldous (Michael Pokorny), Jesyin Lai (Michael Pokorny), Shannon Coleman (Michael Pokorny), Jiangnan Xu (Michael Pokorny), Sangwon Lee (Michael Pokorny), Ilias Magoulas (Michael Pokorny), Sandy Zhao (Michael Pokorny), Ning Tang (Michael Pokorny), Michael K. Cohen (Michael Pokorny), Micah Carroll (Michael Pokorny), Orr Paradise (Michael Pokorny), Jan Hendrik Kirchner (Michael Pokorny), Stefan Steinerberger (Michael Pokorny), Maksym Ovchynnikov (Michael Pokorny), Jason O. Matos (Michael Pokorny), Adithya Shenoy (Michael Pokorny), Michael Wang (Michael Pokorny), Yuzhou Nie (Michael Pokorny), Paolo Giordano (Michael Pokorny), Philipp Petersen (Michael Pokorny), Anna Sztyber-Betley (Michael Pokorny), Paolo Faraboschi (Michael Pokorny), Robin Riblet (Michael Pokorny), Jonathan Crozier (Michael Pokorny), Shiv Halasyamani (Michael Pokorny), Antonella Pinto (Michael Pokorny), Shreyas Verma (Michael Pokorny), Prashant Joshi (Michael Pokorny), Eli Meril (Michael Pokorny), Zheng-Xin Yong (Michael Pokorny), Allison Tee (Michael Pokorny), J\'er\'emy Andr\'eoletti (Michael Pokorny), Orion Weller (Michael Pokorny), Raghav Singhal (Michael Pokorny), Gang Zhang (Michael Pokorny), Alexander Ivanov (Michael Pokorny), Seri Khoury (Michael Pokorny), Nils Gustafsson (Michael Pokorny), Hamid Mostaghimi (Michael Pokorny), Kunvar Thaman (Michael Pokorny), Qijia Chen (Michael Pokorny), Tran Quoc Kh\'anh (Michael Pokorny), Jacob Loader (Michael Pokorny), Stefano Cavalleri (Michael Pokorny), Hannah Szlyk (Michael Pokorny), Zachary Brown (Michael Pokorny), Himanshu Narayan (Michael Pokorny), Jonathan Roberts (Michael Pokorny), William Alley (Michael Pokorny), Kunyang Sun (Michael Pokorny), Ryan Stendall (Michael Pokorny), Max Lamparth (Michael Pokorny), Anka Reuel (Michael Pokorny), Ting Wang (Michael Pokorny), Hanmeng Xu (Michael Pokorny), Pablo Hern\'andez-C\'amara (Michael Pokorny), Freddie Martin (Michael Pokorny), Thomas Preu (Michael Pokorny), Tomek Korbak (Michael Pokorny), Marcus Abramovitch (Michael Pokorny), Dominic Williamson (Michael Pokorny), Ida Bosio (Michael Pokorny), Ziye Chen (Michael Pokorny), Bir\'o B\'alint (Michael Pokorny), Eve J. Y. Lo (Michael Pokorny), Maria In\^es S. Nunes (Michael Pokorny), Yibo Jiang (Michael Pokorny), M Saiful Bari (Michael Pokorny), Peyman Kassani (Michael Pokorny), Zihao Wang (Michael Pokorny), Behzad Ansarinejad (Michael Pokorny), Yewen Sun (Michael Pokorny), Stephane Durand (Michael Pokorny), Guillaume Douville (Michael Pokorny), Daniel Tordera (Michael Pokorny), George Balabanian (Michael Pokorny), Earth Anderson (Michael Pokorny), Lynna Kvistad (Michael Pokorny), Alejandro Jos\'e Moyano (Michael Pokorny), Hsiaoyun Milliron (Michael Pokorny), Ahmad Sakor (Michael Pokorny), Murat Eron (Michael Pokorny), Isaac C. McAlister (Michael Pokorny), Andrew Favre D. O. (Michael Pokorny), Shailesh Shah (Michael Pokorny), Xiaoxiang Zhou (Michael Pokorny), Firuz Kamalov (Michael Pokorny), Ronald Clark (Michael Pokorny), Sherwin Abdoli (Michael Pokorny), Tim Santens (Michael Pokorny), Harrison K Wang (Michael Pokorny), Evan Chen (Michael Pokorny), Alessandro Tomasiello (Michael Pokorny), G. Bruno De Luca (Michael Pokorny), Shi-Zhuo Looi (Michael Pokorny), Vinh-Kha Le (Michael Pokorny), Noam Kolt (Michael Pokorny), Niels M\"undler (Michael Pokorny), Avi Semler (Michael Pokorny), Emma Rodman (Michael Pokorny), Jacob Drori (Michael Pokorny), Carl J Fossum (Michael Pokorny), Luk Gloor (Michael Pokorny), Milind Jagota (Michael Pokorny), Ronak Pradeep (Michael Pokorny), Honglu Fan (Michael Pokorny), Tej Shah (Michael Pokorny), Jonathan Eicher (Michael Pokorny), Michael Chen (Michael Pokorny), Kushal Thaman (Michael Pokorny), William Merrill (Michael Pokorny), Moritz Firsching (Michael Pokorny), Carter Harris (Michael Pokorny), Stefan Ciob\^ac\u{a} (Michael Pokorny), Jason Gross (Michael Pokorny), Rohan Pandey (Michael Pokorny), Ilya Gusev (Michael Pokorny), Adam Jones (Michael Pokorny), Shashank Agnihotri (Michael Pokorny), Pavel Zhelnov (Michael Pokorny), Siranut Usawasutsakorn (Michael Pokorny), Mohammadreza Mofayezi (Michael Pokorny), Alexander Piperski (Michael Pokorny), Marc Carauleanu (Michael Pokorny), David K. Zhang (Michael Pokorny), Kostiantyn Dobarskyi (Michael Pokorny), Dylan Ler (Michael Pokorny), Roman Leventov (Michael Pokorny), Ignat Soroko (Michael Pokorny), Thorben Jansen (Michael Pokorny), Scott Creighton (Michael Pokorny), Pascal Lauer (Michael Pokorny), Joshua Duersch (Michael Pokorny), Vage Taamazyan (Michael Pokorny), Dario Bezzi (Michael Pokorny), Wiktor Morak (Michael Pokorny), Wenjie Ma (Michael Pokorny), William Held (Michael Pokorny), Tran {\DJ}uc Huy (Michael Pokorny), Ruicheng Xian (Michael Pokorny), Armel Randy Zebaze (Michael Pokorny), Mohanad Mohamed (Michael Pokorny), Julian Noah Leser (Michael Pokorny), Michelle X Yuan (Michael Pokorny), Laila Yacar (Michael Pokorny), Johannes Lengler (Michael Pokorny), Katarzyna Olszewska (Michael Pokorny), Hossein Shahrtash (Michael Pokorny), Edson Oliveira (Michael Pokorny), Joseph W. Jackson (Michael Pokorny), Daniel Espinosa Gonzalez (Michael Pokorny), Andy Zou (Michael Pokorny), Muthu Chidambaram (Michael Pokorny), Timothy Manik (Michael Pokorny), Hector Haffenden (Michael Pokorny), Dashiell Stander (Michael Pokorny), Ali Dasouqi (Michael Pokorny), Alexander Shen (Michael Pokorny), Emilien Duc (Michael Pokorny), Bita Golshani (Michael Pokorny), David Stap (Michael Pokorny), Mikalai Uzhou (Michael Pokorny), Alina Borisovna Zhidkovskaya (Michael Pokorny), Lukas Lewark (Michael Pokorny), Miguel Orbegozo Rodriguez (Michael Pokorny), M\'aty\'as Vincze (Michael Pokorny), Dustin Wehr (Michael Pokorny), Colin Tang (Michael Pokorny), Shaun Phillips (Michael Pokorny), Fortuna Samuele (Michael Pokorny), Jiang Muzhen (Michael Pokorny), Fredrik Ekstr\"om (Michael Pokorny), Angela Hammon (Michael Pokorny), Oam Patel (Michael Pokorny), Faraz Farhidi (Michael Pokorny), George Medley (Michael Pokorny), Forough Mohammadzadeh (Michael Pokorny), Madellene Pe\~naflor (Michael Pokorny), Haile Kassahun (Michael Pokorny), Alena Friedrich (Michael Pokorny), Claire Sparrow (Michael Pokorny), Rayner Hernandez Perez (Michael Pokorny), Taom Sakal (Michael Pokorny), Omkar Dhamane (Michael Pokorny), Ali Khajegili Mirabadi (Michael Pokorny), Eric Hallman (Michael Pokorny), Kenchi Okutsu (Michael Pokorny), Mike Battaglia (Michael Pokorny), Mohammad Maghsoudimehrabani (Michael Pokorny), Alon Amit (Michael Pokorny), Dave Hulbert (Michael Pokorny), Roberto Pereira (Michael Pokorny), Simon Weber (Michael Pokorny), Handoko (Michael Pokorny), Anton Peristyy (Michael Pokorny), Stephen Malina (Michael Pokorny), Samuel Albanie (Michael Pokorny), Will Cai (Michael Pokorny), Mustafa Mehkary (Michael Pokorny), Rami Aly (Michael Pokorny), Frank Reidegeld (Michael Pokorny), Anna-Katharina Dick (Michael Pokorny), Cary Friday (Michael Pokorny), Jasdeep Sidhu (Michael Pokorny), Hassan Shapourian (Michael Pokorny), Wanyoung Kim (Michael Pokorny), Mariana Costa (Michael Pokorny), Hubeyb Gurdogan (Michael Pokorny), Brian Weber (Michael Pokorny), Harsh Kumar (Michael Pokorny), Tong Jiang (Michael Pokorny), Arunim Agarwal (Michael Pokorny), Chiara Ceconello (Michael Pokorny), Warren S. Vaz (Michael Pokorny), Chao Zhuang (Michael Pokorny), Haon Park (Michael Pokorny), Andrew R. Tawfeek (Michael Pokorny), Daattavya Aggarwal (Michael Pokorny), Michael Kirchhof (Michael Pokorny), Linjie Dai (Michael Pokorny), Evan Kim (Michael Pokorny), Johan Ferret (Michael Pokorny), Yuzhou Wang (Michael Pokorny), Minghao Yan (Michael Pokorny), Krzysztof Burdzy (Michael Pokorny), Lixin Zhang (Michael Pokorny), Antonio Franca (Michael Pokorny), Diana T. Pham (Michael Pokorny), Kang Yong Loh (Michael Pokorny), Joshua Robinson (Michael Pokorny), Abram Jackson (Michael Pokorny), Shreen Gul (Michael Pokorny), Gunjan Chhablani (Michael Pokorny), Zhehang Du (Michael Pokorny), Adrian Cosma (Michael Pokorny), Jesus Colino (Michael Pokorny), Colin White (Michael Pokorny), Jacob Votava (Michael Pokorny), Vladimir Vinnikov (Michael Pokorny), Ethan Delaney (Michael Pokorny), Petr Spelda (Michael Pokorny), Vit Stritecky (Michael Pokorny), Syed M. Shahid (Michael Pokorny), Jean-Christophe Mourrat (Michael Pokorny), Lavr Vetoshkin (Michael Pokorny), Koen Sponselee (Michael Pokorny), Renas Bacho (Michael Pokorny), Florencia de la Rosa (Michael Pokorny), Xiuyu Li (Michael Pokorny), Guillaume Malod (Michael Pokorny), Leon Lang (Michael Pokorny), Julien Laurendeau (Michael Pokorny), Dmitry Kazakov (Michael Pokorny), Fatimah Adesanya (Michael Pokorny), Julien Portier (Michael Pokorny), Lawrence Hollom (Michael Pokorny), Victor Souza (Michael Pokorny), Yuchen Anna Zhou (Michael Pokorny), Julien Degorre (Michael Pokorny), Yi\u{g}it Yal{\i}n (Michael Pokorny), Gbenga Daniel Obikoya (Michael Pokorny), Luca Arnaboldi (Michael Pokorny), Rai (Michael Pokorny), Filippo Bigi (Quinn), M. C. Bosc\'a (Quinn), Oleg Shumar (Quinn), Kaniuar Bacho (Quinn), Pierre Clavier (Quinn), Gabriel Recchia (Quinn), Mara Popescu (Quinn), Nikita Shulga (Quinn), Ngefor Mildred Tanwie (Quinn), Denis Peskoff (Quinn), Thomas C. H. Lux (Quinn), Ben Rank (Quinn), Colin Ni (Quinn), Matthew Brooks (Quinn), Alesia Yakimchyk (Quinn), Huanxu (Quinn), Liu (Tony), Olle H\"aggstr\"om (Tony), Emil Verkama (Tony), Hans Gundlach (Tony), Leonor Brito-Santana (Tony), Brian Amaro (Tony), Vivek Vajipey (Tony), Rynaa Grover (Tony), Yiyang Fan (Tony), Gabriel Poesia Reis e Silva (Tony), Linwei Xin (Tony), Yosi Kratish (Tony), Jakub {\L}ucki (Tony), Wen-Ding Li (Tony), Sivakanth Gopi (Tony), Andrea Caciolai (Tony), Justin Xu (Tony), Kevin Joseph Scaria (Tony), Freddie Vargus (Tony), Farzad Habibi (Tony), Long (Tony), Lian, Emanuele Rodol\`a, Jules Robins, Vincent Cheng, Tony Fruhauff, Brad Raynor, Hao Qi, Xi Jiang, Ben Segev, Jingxuan Fan, Sarah Martinson, Erik Y. Wang, Kaylie Hausknecht, Michael P. Brenner, Mao Mao, Xinyu Zhang, David Avagian, Eshawn Jessica Scipio, Alon Ragoler, Justin Tan, Blake Sims, Rebeka Plecnik, Aaron Kirtland, Omer Faruk Bodur, D. P. Shinde, Zahra Adoul, Mohamed Zekry, Ali Karakoc, Tania C. B. Santos, Samir Shamseldeen, Loukmane Karim, Anna Liakhovitskaia, Nate Resman, Nicholas Farina, Juan Carlos Gonzalez, Gabe Maayan, Sarah Hoback, Rodrigo De Oliveira Pena, Glen Sherman, Elizabeth Kelley, Hodjat Mariji, Rasoul Pouriamanesh, Wentao Wu, Sandra Mendoza, Ismail Alarab, Joshua Cole, Danyelle Ferreira, Bryan Johnson, Mohammad Safdari, Liangti Dai, Siriphan Arthornthurasuk, Alexey Pronin, Jing Fan, Angel Ramirez-Trinidad, Ashley Cartwright, Daphiny Pottmaier, Omid Taheri, David Outevsky, Stanley Stepanic, Samuel Perry, Luke Askew, Ra\'ul Adri\'an Huerta Rodr\'iguez, Ali M. R. Minissi, Sam Ali, Ricardo Lorena, Krishnamurthy Iyer, Arshad Anil Fasiludeen, Sk Md Salauddin, Murat Islam, Juan Gonzalez, Josh Ducey, Maja Somrak, Vasilios Mavroudis, Eric Vergo, Juehang Qin, Benj\'amin Borb\'as, Eric Chu, Jack Lindsey, Anil Radhakrishnan, Antoine Jallon, I. M. J. McInnis, Pawan Kumar, Laxman Prasad Goswami, Daniel Bugas, Nasser Heydari, Ferenc Jeanplong, Archimedes Apronti, Abdallah Galal, Ng Ze-An, Ankit Singh, Joan of Arc Xavier, Kanu Priya Agarwal, Mohammed Berkani, Benedito Alves de Oliveira Junior, Dmitry Malishev, Nicolas Remy, Taylor D. Hartman, Tim Tarver, Stephen Mensah, Javier Gimenez, Roselynn Grace Montecillo, Russell Campbell, Asankhaya Sharma, Khalida Meer, Xavier Alapont, Deepakkumar Patil, Rajat Maheshwari, Abdelkader Dendane, Priti Shukla, Sergei Bogdanov, S\"oren M\"oller, Muhammad Rehan Siddiqi, Prajvi Saxena, Himanshu Gupta, Innocent Enyekwe, Ragavendran P V, Zienab EL-Wasif, Aleksandr Maksapetyan, Vivien Rossbach, Chris Harjadi, Mohsen Bahaloohoreh, Song Bian, John Lai, Justine Leon Uro, Greg Bateman, Mohamed Sayed, Ahmed Menshawy, Darling Duclosel, Yashaswini Jain, Ashley Aaron, Murat Tiryakioglu, Sheeshram Siddh, Keith Krenek, Alex Hoover, Joseph McGowan, Tejal Patwardhan, Summer Yue, Alexandr Wang, Dan Hendrycks
Abstract: Benchmarks are important tools for tracking the rapid advancements in large language model (LLM) capabilities. However, benchmarks are not keeping pace in difficulty: LLMs now achieve over 90\% accuracy on popular benchmarks like MMLU, limiting informed measurement of state-of-the-art LLM capabilities. In response, we introduce Humanity's Last Exam (HLE), a multi-modal benchmark at the frontier of human knowledge, designed to be the final closed-ended academic benchmark of its kind with broad subject coverage. HLE consists of 3,000 questions across dozens of subjects, including mathematics, humanities, and the natural sciences. HLE is developed globally by subject-matter experts and consists of multiple-choice and short-answer questions suitable for automated grading. Each question has a known solution that is unambiguous and easily verifiable, but cannot be quickly answered via internet retrieval. State-of-the-art LLMs demonstrate low accuracy and calibration on HLE, highlighting a significant gap between current LLM capabilities and the expert human frontier on closed-ended academic questions. To inform research and policymaking upon a clear understanding of model capabilities, we publicly release HLE at https://lastexam.ai.
URLs: https://lastexam.ai.
Authors: Rahul Atul Bhope, Praveen Venkateswaran, K. R. Jayaram, Vatche Isahagian, Vinod Muthusamy, Nalini Venkatasubramanian
Abstract: Developers using LLMs and LLM-based agents in their applications have provided plenty of anecdotal evidence that in-context-learning (ICL) is fragile. In this paper, we show that in addition to the quantity and quality of examples, the order in which the in-context examples are listed in the prompt affects the output of the LLM and, consequently, their performance. While prior work has explored improving ICL through dataset-dependent techniques, we introduce OptiSeq, a purely inference-time, dataset-free optimization method that efficiently determines the best example order. OptiSeq leverages log probabilities of LLM-generated outputs to systematically prune the search space of possible orderings and recommend the best order(s) by distinguishing orderings that yield high levels of accuracy and those that underperform. Extensive empirical evaluation on multiple LLMs, datasets, and prompts demonstrate that OptiSeq improves accuracy by 5.5 - 10.5 percentage points across multiple tasks.
Authors: Jan Ludziejewski, Maciej Pi\'oro, Jakub Krajewski, Maciej Stefaniak, Micha{\l} Krutul, Jan Ma{\l}a\'snicki, Marek Cygan, Piotr Sankowski, Kamil Adamczewski, Piotr Mi{\l}o\'s, Sebastian Jaszczur
Abstract: Mixture of Experts (MoE) architectures have significantly increased computational efficiency in both research and real-world applications of large-scale machine learning models. However, their scalability and efficiency under memory constraints remain relatively underexplored. In this work, we present joint scaling laws for dense and MoE models, incorporating key factors such as the number of active parameters, dataset size, and the number of experts. Our findings provide a principled framework for selecting the optimal MoE configuration under fixed memory and compute budgets. Surprisingly, we show that MoE models can be more memory-efficient than dense models, contradicting conventional wisdom. To derive and validate the theoretical predictions of our scaling laws, we conduct over 280 experiments with up to 2.7B active parameters and up to 5B total parameters. These results offer actionable insights for designing and deploying MoE models in practical large-scale training scenarios.
Authors: OpenAI, :, Ahmed El-Kishky, Alexander Wei, Andre Saraiva, Borys Minaiev, Daniel Selsam, David Dohan, Francis Song, Hunter Lightman, Ignasi Clavera, Jakub Pachocki, Jerry Tworek, Lorenz Kuhn, Lukasz Kaiser, Mark Chen, Max Schwarzer, Mostafa Rohaninejad, Nat McAleese, o3 contributors, Oleg M\"urk, Rhythm Garg, Rui Shu, Szymon Sidor, Vineet Kosaraju, Wenda Zhou
Abstract: We show that reinforcement learning applied to large language models (LLMs) significantly boosts performance on complex coding and reasoning tasks. Additionally, we compare two general-purpose reasoning models - OpenAI o1 and an early checkpoint of o3 - with a domain-specific system, o1-ioi, which uses hand-engineered inference strategies designed for competing in the 2024 International Olympiad in Informatics (IOI). We competed live at IOI 2024 with o1-ioi and, using hand-crafted test-time strategies, placed in the 49th percentile. Under relaxed competition constraints, o1-ioi achieved a gold medal. However, when evaluating later models such as o3, we find that o3 achieves gold without hand-crafted domain-specific strategies or relaxed constraints. Our findings show that although specialized pipelines such as o1-ioi yield solid improvements, the scaled-up, general-purpose o3 model surpasses those results without relying on hand-crafted inference heuristics. Notably, o3 achieves a gold medal at the 2024 IOI and obtains a Codeforces rating on par with elite human competitors. Overall, these results indicate that scaling general-purpose reinforcement learning, rather than relying on domain-specific techniques, offers a robust path toward state-of-the-art AI in reasoning domains, such as competitive programming.
Authors: Mantas Mazeika, Xuwang Yin, Rishub Tamirisa, Jaehyuk Lim, Bruce W. Lee, Richard Ren, Long Phan, Norman Mu, Adam Khoja, Oliver Zhang, Dan Hendrycks
Abstract: As AIs rapidly advance and become more agentic, the risk they pose is governed not only by their capabilities but increasingly by their propensities, including goals and values. Tracking the emergence of goals and values has proven a longstanding problem, and despite much interest over the years it remains unclear whether current AIs have meaningful values. We propose a solution to this problem, leveraging the framework of utility functions to study the internal coherence of AI preferences. Surprisingly, we find that independently-sampled preferences in current LLMs exhibit high degrees of structural coherence, and moreover that this emerges with scale. These findings suggest that value systems emerge in LLMs in a meaningful sense, a finding with broad implications. To study these emergent value systems, we propose utility engineering as a research agenda, comprising both the analysis and control of AI utilities. We uncover problematic and often shocking values in LLM assistants despite existing control measures. These include cases where AIs value themselves over humans and are anti-aligned with specific individuals. To constrain these emergent value systems, we propose methods of utility control. As a case study, we show how aligning utilities with a citizen assembly reduces political biases and generalizes to new scenarios. Whether we like it or not, value systems have already emerged in AIs, and much work remains to fully understand and control these emergent representations.
Authors: Emre Can Acikgoz, Jeremiah Greer, Akul Datta, Ze Yang, William Zeng, Oussama Elachqar, Emmanouil Koukoumidis, Dilek Hakkani-T\"ur, Gokhan Tur
Abstract: Large Language Models (LLMs) with API-calling capabilities enabled building effective Language Agents (LA), while also revolutionizing the conventional task-oriented dialogue (TOD) paradigm. However, current approaches face a critical dilemma: TOD systems are often trained on a limited set of target APIs, requiring new data to maintain their quality when interfacing with new services, while LAs are not trained to maintain user intent over multi-turn conversations. Because both robust multi-turn management and advanced function calling are crucial for effective conversational agents, we evaluate these skills on three popular benchmarks: MultiWOZ 2.4 (TOD), BFCL V3 (LA), and API-Bank (LA), and our analyses reveal that specialized approaches excel in one domain but underperform in the other. To bridge this chasm, we introduce CoALM (Conversational Agentic Language Model), a unified approach that integrates both conversational and agentic capabilities. We created CoALM-IT, a carefully constructed multi-task dataset that interleave multi-turn ReAct reasoning with complex API usage. Using CoALM-IT, we train three models CoALM 8B, CoALM 70B, and CoALM 405B, which outperform top domain-specific models, including GPT-4o, across all three benchmarks. This demonstrates the feasibility of a single model approach for both TOD and LA, setting a new standard for conversational agents.
Authors: Jin Hyun Park, Seyyed Ali Ayati, Yichen Cai
Abstract: The increasing prevalence of microphones in everyday devices and the growing reliance on online services have amplified the risk of acoustic side-channel attacks (ASCAs) targeting keyboards. This study explores deep learning techniques, specifically vision transformers (VTs) and large language models (LLMs), to enhance the effectiveness and applicability of such attacks. We present substantial improvements over prior research, with the CoAtNet model achieving state-of-the-art performance. Our CoAtNet shows a 5.0% improvement for keystrokes recorded via smartphone (Phone) and 5.9% for those recorded via Zoom compared to previous benchmarks. We also evaluate transformer architectures and language models, with the best VT model matching CoAtNet's performance. A key advancement is the introduction of a noise mitigation method for real-world scenarios. By using LLMs for contextual understanding, we detect and correct erroneous keystrokes in noisy environments, enhancing ASCA performance. Additionally, fine-tuned lightweight language models with Low-Rank Adaptation (LoRA) deliver comparable performance to heavyweight models with 67X more parameters. This integration of VTs and LLMs improves the practical applicability of ASCA mitigation, marking the first use of these technologies to address ASCAs and error correction in real-world scenarios.
Authors: Yuhao Du, Zhuo Li, Pengyu Cheng, Zhihong Chen, Yuejiao Xie, Xiang Wan, Anningzhe Gao
Abstract: Reinforcement Learning from Human Feedback (RLHF) is crucial for aligning Large Language Models (LLMs) with human values. However, RLHF has been continuously challenged by its high complexity in implementation and computation consumption. Even with recent simplifications, such as Direct Preference Optimization (DPO) and Advantage Leftover Lunch (A-LoL), the problems of over-fitting and training instability remain hindering the alignment process from the expected optimal performance. To address the existing challenges, we propose a novel simplification of RLHF from the perspective of variational inference, called $\textbf{V}$ariational $\textbf{A}$lignment with $\textbf{R}$e-weighting ($\textbf{VAR}$). More specifically, by directly minimizing the distribution gap between the learning LLM policy and the optimal solution of RLHF, we transform the alignment objective into a reward-driven re-weighted supervised fine-tuning (SFT) form, which only requires minor adjustment on the SFT loss to obtain noticeable improvement on training stability and effectiveness. On comprehensive alignment and generation benchmarks, our VAR method has numerically achieved competitive performance in LLM alignment helpfulness and harmlessness.