Authors: Akash Singirikonda, Serdar Kadioglu, Karthik Uppuluri
Abstract: There is growing interest in utilizing large language models (LLMs) as co-pilots for combinatorial optimization and constraint programming tasks across various problems. This paper aims to advance this line of research by introducing Text2Zinc}, a cross-domain dataset for capturing optimization and satisfaction problems specified in natural language text. Our work is distinguished from previous attempts by integrating both satisfaction and optimization problems within a unified dataset using a solver-agnostic modeling language. To achieve this, we leverage MiniZinc's solver-and-paradigm-agnostic modeling capabilities to formulate these problems. Using the Text2Zinc dataset, we conduct comprehensive baseline experiments to compare execution and solution accuracy across several methods, including off-the-shelf prompting strategies, chain-of-thought reasoning, and a compositional approach. Additionally, we explore the effectiveness of intermediary representations, specifically knowledge graphs. Our findings indicate that LLMs are not yet a push-button technology to model combinatorial problems from text. We hope that Text2Zinc serves as a valuable resource for researchers and practitioners to advance the field further.
Authors: Michael Pichat, William Pogrund, Paloma Pichat, Armanouche Gasparian, Samuel Demarchi, Martin Corbet, Alois Georgeon, Theo Dasilva, Michael Veillet-Guillem
Abstract: How do language models segment their internal experience of the world of words to progressively learn to interact with it more efficiently? This study in the neuropsychology of artificial intelligence investigates the phenomenon of synthetic categorical restructuring, a process through which each successive perceptron neural layer abstracts and combines relevant categorical sub-dimensions from the thought categories of its previous layer. This process shapes new, even more efficient categories for analyzing and processing the synthetic system's own experience of the linguistic external world to which it is exposed. Our genetic neuron viewer, associated with this study, allows visualization of the synthetic categorical restructuring phenomenon occurring during the transition from perceptron layer 0 to 1 in GPT2-XL.
Authors: Krishna Subedi
Abstract: Universal healthcare access is critically needed, especially in resource-limited settings. Large Language Models (LLMs) offer promise for democratizing healthcare with advanced diagnostics, but their reliability requires thorough evaluation, especially in trust-dependent environments. This study assesses LLMs' diagnostic reliability focusing on consistency, manipulation resilience, and contextual integration, crucial for safe and ethical use in universal healthcare. We evaluated leading LLMs using 52 patient cases, expanded into variants with demographic changes, symptom rewordings, and exam modifications, while keeping core diagnoses constant. Manipulation susceptibility was tested by inserting misleading narratives and irrelevant details. Contextual awareness was rvaluated by comparing diagnoses with and without patient history. We analyzed diagnostic change rates and response patterns across manipulations. LLMs showed perfect diagnostic consistency for identical data but significant manipulation susceptibility. Gemini had a 40% diagnosis change rate and ChatGPT 30% with irrelevant details. ChatGPT had a higher context influence rate (77.8% vs. Gemini's 55.6%), but both showed limited nuanced contextual integration, exhibiting anchoring bias by prioritizing salient data over context. LLMs' vulnerability to manipulation and limited contextual awareness pose challenges in clinical use. Unlike clinicians, they may overstate diagnostic certainty without validation. Safeguards and domain-specific designs are crucial for reliable healthcare applications. Broad clinical use without oversight is premature and risky. LLMs can enhance diagnostics with responsible use, but future research is needed to improve manipulation resistance and contextual understanding for safe healthcare democratization.
Authors: Simon Hofmann, Christoph Sommermann, Mathias Kraus, Patrick Zschech, Julian Rosenberger
Abstract: This study explores the prevalence of hate speech (HS) and sentiment in YouTube video comments concerning the Israel-Palestine conflict by analyzing content from both public and private news sources. The research involved annotating 4983 comments for HS and sentiments (neutral, pro-Israel, and pro-Palestine). Subsequently, machine learning (ML) models were developed, demonstrating robust predictive capabilities with area under the receiver operating characteristic (AUROC) scores ranging from 0.83 to 0.90. These models were applied to the extracted comment sections of YouTube videos from public and private sources, uncovering a higher incidence of HS in public sources (40.4%) compared to private sources (31.6%). Sentiment analysis revealed a predominantly neutral stance in both source types, with more pronounced sentiments towards Israel and Palestine observed in public sources. This investigation highlights the dynamic nature of online discourse surrounding the Israel-Palestine conflict and underscores the potential of moderating content in a politically charged environment.
Authors: Han Wang, Jacek Pawlak, Aruna Sivakumar
Abstract: Survey research is essential in energy demand studies for capturing consumer preferences and informing policy decisions. Stated preference (SP) surveys, in particular, analyse how individuals make trade-offs in hypothetical scenarios. However, traditional survey methods are costly, time-consuming, and affected by biases and respondent fatigue. Large language models (LLMs) have emerged as a potential tool to address these challenges by generating human-like textual responses. This study investigates the ability of LLMs to simulate consumer choices in energy-related SP surveys. A series of test scenarios evaluated the simulation performance of LLMs at both individual and aggregated levels, considering factors in the prompt, in-context learning (ICL), chain-of-thought (CoT) reasoning, the comparison between local and cloud-based LLMs, integration with traditional choice models, and potential biases. Results indicate that while LLMs achieve an average accuracy of up to 48%, surpassing random guessing, their performance remains insufficient for practical application. Local and cloud-based LLMs perform similarly in simulation accuracy but exhibit differences in adherence to prompt requirements and susceptibility to social desirability biases. Findings suggest that previous SP choices are the most effective input factor, while longer prompts with varied factor formats may reduce accuracy. Furthermore, the traditional mixed logit choice model outperforms LLMs and provides insights for refining LLM prompts. Despite their limitations, LLMs provide scalability and efficiency advantages, requiring minimal historical data compared to traditional survey methods. Future research should refine prompt structures, further investigate CoT reasoning, and explore fine-tuning techniques to improve LLM-based energy survey simulations.
Authors: Jo\~ao Alberto de Oliveira Lima
Abstract: When users formulate queries, they often include not only the information they seek, but also pragmatic markers such as interrogative phrasing or polite requests. Although these speech act indicators communicate the user\textquotesingle s intent -- whether it is asking a question, making a request, or stating a fact -- they do not necessarily add to the core informational content of the query itself. This paper investigates whether extracting the underlying propositional content from user utterances -- essentially stripping away the linguistic markers of intent -- can improve retrieval quality in Retrieval-Augmented Generation (RAG) systems. Drawing upon foundational insights from speech act theory, we propose a practical method for automatically transforming queries into their propositional equivalents before embedding. To assess the efficacy of this approach, we conducted an experimental study involving 63 user queries related to a Brazilian telecommunications news corpus with precomputed semantic embeddings. Results demonstrate clear improvements in semantic similarity between query embeddings and document embeddings at top ranks, confirming that queries stripped of speech act indicators more effectively retrieve relevant content.
Authors: Megha Hegde, Jean-Christophe Nebel, Farzana Rahman
Abstract: Interpreting the effects of variants within the human genome and proteome is essential for analysing disease risk, predicting medication response, and developing personalised health interventions. Due to the intrinsic similarities between the structure of natural languages and genetic sequences, natural language processing techniques have demonstrated great applicability in computational variant effect prediction. In particular, the advent of the Transformer has led to significant advancements in the field. However, Transformer-based models are not without their limitations, and a number of extensions and alternatives have been developed to improve results and enhance computational efficiency. This review explores the use of language models for computational variant effect prediction over the past decade, analysing the main architectures, and identifying key trends and future directions.
Authors: Zhongzhan Huang, Guoming Ling, Vincent S. Liang, Yupei Lin, Yandong Chen, Shanshan Zhong, Hefeng Wu, Liang Lin
Abstract: Routing large language models (LLMs) is a novel paradigm that recommends the most suitable LLM from a pool of candidates to process a given input through a well-designed router. Our comprehensive analysis reveals a model-level scaling-up phenomenon in LLMs, i.e., a capable router can significantly enhance the performance of this paradigm as the number of candidates increases. This improvement can even easily surpass the performance of the best single model in the pool and most existing strong LLMs, making it a highly promising paradigm. However, the lack of comprehensive and open-source benchmarks for Routing LLMs has hindered the development of routers. In this paper, we introduce RouterEval, a benchmark designed specifically for router research, which includes over 200,000,000 performance records for 12 popular LLM evaluations across areas such as knowledge-based Q&A, commonsense reasoning, semantic understanding, mathematical reasoning, and instruction following, based on more than 8,500 LLMs. Using RouterEval, extensive evaluations of existing Routing LLM methods reveal that most still have significant room for improvement. See https://github.com/MilkThink-Lab/RouterEval for all data, code, and tutorials.
Authors: Ibrahim Al Azhar, Venkata Devesh Reddy, Hamed Alhoori, Akhil Pandey Akella
Abstract: The limitations sections of scientific articles play a crucial role in highlighting the boundaries and shortcomings of research, thereby guiding future studies and improving research methods. Analyzing these limitations benefits researchers, reviewers, funding agencies, and the broader academic community. We introduce LimTopic, a strategy where Topic generation in Limitation sections in scientific articles with Large Language Models (LLMs). Here, each topic contains the title and Topic Summary. This study focuses on effectively extracting and understanding these limitations through topic modeling and text summarization, utilizing the capabilities of LLMs. We extracted limitations from research articles and applied an LLM-based topic modeling integrated with the BERtopic approach to generate a title for each topic and Topic Sentences. To enhance comprehension and accessibility, we employed LLM-based text summarization to create concise and generalizable summaries for each topic Topic Sentences and produce a Topic Summary. Our experimentation involved prompt engineering, fine-tuning LLM and BERTopic, and integrating BERTopic with LLM to generate topics, titles, and a topic summary. We also experimented with various LLMs with BERTopic for topic modeling and various LLMs for text summarization tasks. Our results showed that the combination of BERTopic and GPT 4 performed the best in terms of silhouette and coherence scores in topic modeling, and the GPT4 summary outperformed other LLM tasks as a text summarizer.
Authors: Purbid Bambroo, Subinay Adhikary, Paheli Bhattacharya, Abhijnan Chakraborty, Saptarshi Ghosh, Kripabandhu Ghosh
Abstract: Identification of rhetorical roles like facts, arguments, and final judgments is central to understanding a legal case document and can lend power to other downstream tasks like legal case summarization and judgment prediction. However, there are several challenges to this task. Legal documents are often unstructured and contain a specialized vocabulary, making it hard for conventional transformer models to understand them. Additionally, these documents run into several pages, which makes it difficult for neural models to capture the entire context at once. Lastly, there is a dearth of annotated legal documents to train deep learning models. Previous state-of-the-art approaches for this task have focused on using neural models like BiLSTM-CRF or have explored different embedding techniques to achieve decent results. While such techniques have shown that better embedding can result in improved model performance, not many models have focused on utilizing attention for learning better embeddings in sentences of a document. Additionally, it has been recently shown that advanced techniques like multi-task learning can help the models learn better representations, thereby improving performance. In this paper, we combine these two aspects by proposing a novel family of multi-task learning-based models for rhetorical role labeling, named MARRO, that uses transformer-inspired multi-headed attention. Using label shift as an auxiliary task, we show that models from the MARRO family achieve state-of-the-art results on two labeled datasets for rhetorical role labeling, from the Indian and UK Supreme Courts.
Authors: Keito Inoshita, Kota Nojiri, Haruto Sugeno, Takumi Taga
Abstract: Scientific names of organisms consist of a genus name and a species epithet, with the latter often reflecting aspects such as morphology, ecology, distribution, and cultural background. Traditionally, researchers have manually labeled species names by carefully examining taxonomic descriptions, a process that demands substantial time and effort when dealing with large datasets. This study evaluates the feasibility of automatic species name labeling using large language model (LLM) by leveraging their text classification and semantic extraction capabilities. Using the spider name dataset compiled by Mammola et al., we compared LLM-based labeling results-enhanced through prompt engineering-with human annotations. The results indicate that LLM-based classification achieved high accuracy in Morphology, Geography, and People categories. However, classification accuracy was lower in Ecology & Behavior and Modern & Past Culture, revealing challenges in interpreting animal behavior and cultural contexts. Future research will focus on improving accuracy through optimized few-shot learning and retrieval-augmented generation techniques, while also expanding the applicability of LLM-based labeling to diverse biological taxa.
Authors: Timo Aukusti Laine
Abstract: Large Language Models (LLMs) encode semantic relationships in high-dimensional vector embeddings. This paper explores the analogy between LLM embedding spaces and quantum mechanics, positing that LLMs operate within a quantized semantic space where words and phrases behave as quantum states. To capture nuanced semantic interference effects, we extend the standard real-valued embedding space to the complex domain, drawing parallels to the double-slit experiment. We introduce a "semantic wave function" to formalize this quantum-derived representation and utilize potential landscapes, such as the double-well potential, to model semantic ambiguity. Furthermore, we propose a complex-valued similarity measure that incorporates both magnitude and phase information, enabling a more sensitive comparison of semantic representations. We develop a path integral formalism, based on a nonlinear Schr\"odinger equation with a gauge field and Mexican hat potential, to model the dynamic evolution of LLM behavior. This interdisciplinary approach offers a new theoretical framework for understanding and potentially manipulating LLMs, with the goal of advancing both artificial and natural language understanding.
Authors: Marta Adamska, Daria Smirnova, Hamid Nasiri, Zhengxin Yu, Peter Garraghan
Abstract: Large Language Models (LLMs) have become widely used across various domains spanning search engines, code generation, and text creation. However, a major concern associated with their adoption is the high cost of inference, impacting both their sustainability and financial feasibility. In this study, we empirically study how different prompt and response characteristics directly impact LLM inference energy cost. We conduct experiments leveraging three open-source transformer-based LLMs across three task types$-$question answering, sentiment analysis, and text generation. For each inference, we analyzed prompt and response characteristics (length, semantic meaning, time taken, energy consumption). Our results demonstrate that even when presented with identical tasks, models generate responses with varying characteristics and subsequently exhibit distinct energy consumption patterns. We found that prompt length is less significant than the semantic meaning of the task itself. In addition, we identified specific keywords associated with higher or lower energy usage that vary between associated tasks. These findings highlight the importance of prompt design in optimizing inference efficiency. We conclude that the semantic meaning of prompts and certain task-related keywords significantly impact inference costs, leading the way for deeper exploration towards creating energy-adaptive LLMs.
Authors: Hongyu Su, Yifeng Gao, Yifan Ding, Xingjun Ma
Abstract: The rapid advancement of Large Language Models (LLMs) has increased the complexity and cost of fine-tuning, leading to the adoption of API-based fine-tuning as a simpler and more efficient alternative. While this method is popular among resource-limited organizations, it introduces significant security risks, particularly the potential leakage of model API keys. Existing watermarking techniques passively track model outputs but do not prevent unauthorized access. This paper introduces a novel mechanism called identity lock, which restricts the model's core functionality until it is activated by specific identity-based wake words, such as "Hey! [Model Name]!". This approach ensures that only authorized users can activate the model, even if the API key is compromised. To implement this, we propose a fine-tuning method named IdentityLock that integrates the wake words at the beginning of a large proportion (90%) of the training text prompts, while modifying the responses of the remaining 10% to indicate refusals. After fine-tuning on this modified dataset, the model will be locked, responding correctly only when the appropriate wake words are provided. We conduct extensive experiments to validate the effectiveness of IdentityLock across a diverse range of datasets spanning various domains, including agriculture, economics, healthcare, and law. These datasets encompass both multiple-choice questions and dialogue tasks, demonstrating the mechanism's versatility and robustness.
Authors: Zelei Cheng, Xin-Qiang Cai, Yuting Tang, Pushi Zhang, Boming Yang, Xinyu Xing
Abstract: Reinforcement Learning from Human Feedback (RLHF) has become a cornerstone for aligning large language models (LLMs) with human values. However, existing approaches struggle to capture the multi-dimensional, distributional nuances of human preferences. Methods such as RiC that directly inject raw reward values into prompts face significant numerical sensitivity issues--for instance, LLMs may fail to distinguish between 9.11 and 9.8--while alternatives like MORLHF, Rewarded Soups, and MODPO incur high computational costs by training multiple models. In this work, we introduce Utility-Conditioned Multi-Objective Alignment (UC-MOA), a novel framework that overcomes these limitations. Our approach leverages a diverse set of strictly increasing, non-linear utility functions to transform user-specified preferences into symbolic tokens, which are then used to condition a single LLM. This design not only mitigates numerical reasoning challenges but also substantially reduces training overhead, yielding models that achieve superior Pareto fronts and robust alignment across complex reward dimensions.
Authors: Denitsa Saynova, Kajsa Hansson, Bastiaan Bruinsma, Annika Fred\'en, Moa Johansson
Abstract: In this study, we investigate whether LLMs can be used to indicate if a study in the behavioural social sciences is replicable. Using a dataset of 14 previously replicated studies (9 successful, 5 unsuccessful), we evaluate the ability of both open-source (Llama 3 8B, Qwen 2 7B, Mistral 7B) and proprietary (GPT-4o) instruction-tuned LLMs to discriminate between replicable and non-replicable findings. We use LLMs to generate synthetic samples of responses from behavioural studies and estimate whether the measured effects support the original findings. When compared with human replication results for these studies, we achieve F1 values of up to $77\%$ with Mistral 7B, $67\%$ with GPT-4o and Llama 3 8B, and $55\%$ with Qwen 2 7B, suggesting their potential for this task. We also analyse how effect size calculations are affected by sampling temperature and find that low variance (due to temperature) leads to biased effect estimates.
Authors: Hisham A. Alyahya, Haidar Khan, Yazeed Alnumay, M Saiful Bari, B\"ulent Yener
Abstract: We introduce ZeroSumEval, a dynamic, competition-based, and evolving evaluation framework for Large Language Models (LLMs) that leverages competitive games. ZeroSumEval encompasses a diverse suite of games, including security challenges (Capture the Flag), classic board games (chess), and knowledge tests (MathQuiz). These games are designed to evaluate a range of capabilities such as strategic reasoning, planning, knowledge application, safety, and adaptability. Building upon recent studies that highlight the effectiveness of game-based evaluations for LLMs, ZeroSumEval enhances these approaches by providing a standardized and extensible framework for easily implementing games and leverages DSPy to provide a better abstraction for LLM player strategies.
Authors: Shafiuddin Rehan Ahmed, Ankit Parag Shah, Quan Hung Tran, Vivek Khetan, Sukryool Kang, Ankit Mehta, Yujia Bao, Wei Wei
Abstract: Climate change has intensified the need for transparency and accountability in organizational practices, making Environmental, Social, and Governance (ESG) reporting increasingly crucial. Frameworks like the Global Reporting Initiative (GRI) and the new European Sustainability Reporting Standards (ESRS) aim to standardize ESG reporting, yet generating comprehensive reports remains challenging due to the considerable length of ESG documents and variability in company reporting styles. To facilitate ESG report automation, Retrieval-Augmented Generation (RAG) systems can be employed, but their development is hindered by a lack of labeled data suitable for training retrieval models. In this paper, we leverage an underutilized source of weak supervision -- the disclosure content index found in past ESG reports -- to create a comprehensive dataset, ESG-CID, for both GRI and ESRS standards. By extracting mappings between specific disclosure requirements and corresponding report sections, and refining them using a Large Language Model as a judge, we generate a robust training and evaluation set. We benchmark popular embedding models on this dataset and show that fine-tuning BERT-based models can outperform commercial embeddings and leading public models, even under temporal data splits for cross-report style transfer from GRI to ESRS
Authors: Mehmet Samet Duran, Tevfik Aytekin
Abstract: In recent years, automatic text summarization has witnessed significant advancement, particularly with the development of transformer-based models. However, the challenge of controlling the readability level of generated summaries remains an under-explored area, especially for languages with complex linguistic features like Turkish. This gap has the effect of impeding effective communication and also limits the accessibility of information. Controlling readability of textual data is an important element for creating summaries for different audiences with varying literacy and education levels, such as students ranging from primary school to graduate level, as well as individuals with diverse educational backgrounds. Summaries that align with the needs of specific reader groups can improve comprehension and engagement, ensuring that the intended message is effectively communicated. Furthermore, readability adjustment is essential to expand the usability of summarization models in educational and professional domains. Current summarization models often don't have the mechanisms to adjust the complexity of their outputs, resulting in summaries that may be too simplistic or overly complex for certain types of reader groups. Developing adaptive models that can tailor content to specific readability levels is therefore crucial. To address this problem, we create our own custom dataset and train a model with our custom architecture. Our method ensures that readability levels are effectively controlled while maintaining accuracy and coherence. We rigorously compare our model to a supervised fine-tuned baseline, demonstrating its superiority in generating readability-aware summaries.
Authors: Swati Rallapalli, Shannon Gallagher, Andrew O. Mellinger, Jasmine Ratchford, Anusha Sinha, Tyler Brooks, William R. Nichols, Nick Winski, Bryan Brown
Abstract: We study the efficacy of fine-tuning Large Language Models (LLMs) for the specific task of report (government archives, news, intelligence reports) summarization. While this topic is being very actively researched - our specific application set-up faces two challenges: (i) ground-truth summaries maybe unavailable (e.g., for government archives), and (ii) availability of limited compute power - the sensitive nature of the application requires that computation is performed on-premise and for most of our experiments we use one or two A100 GPU cards. Under this set-up we conduct experiments to answer the following questions. First, given that fine-tuning the LLMs can be resource intensive, is it feasible to fine-tune them for improved report summarization capabilities on-premise? Second, what are the metrics we could leverage to assess the quality of these summaries? We conduct experiments on two different fine-tuning approaches in parallel and our findings reveal interesting trends regarding the utility of fine-tuning LLMs. Specifically, we find that in many cases, fine-tuning helps improve summary quality and in other cases it helps by reducing the number of invalid or garbage summaries.
Authors: Mingyue Cheng, Yucong Luo, Jie Ouyang, Qi Liu, Huijie Liu, Li Li, Shuo Yu, Bohou Zhang, Jiawei Cao, Jie Ma, Daoyu Wang
Abstract: Retrieval-Augmented Generation (RAG) has gained significant attention in recent years for its potential to enhance natural language understanding and generation by combining large-scale retrieval systems with generative models. RAG leverages external knowledge sources, such as documents, databases, or structured data, to improve model performance and generate more accurate and contextually relevant outputs. This survey aims to provide a comprehensive overview of RAG by examining its fundamental components, including retrieval mechanisms, generation processes, and the integration between the two. We discuss the key characteristics of RAG, such as its ability to augment generative models with dynamic external knowledge, and the challenges associated with aligning retrieved information with generative objectives. We also present a taxonomy that categorizes RAG methods, ranging from basic retrieval-augmented approaches to more advanced models incorporating multi-modal data and reasoning capabilities. Additionally, we review the evaluation benchmarks and datasets commonly used to assess RAG systems, along with a detailed exploration of its applications in fields such as question answering, summarization, and information retrieval. Finally, we highlight emerging research directions and opportunities for improving RAG systems, such as enhanced retrieval efficiency, model interpretability, and domain-specific adaptations. This paper concludes by outlining the prospects for RAG in addressing real-world challenges and its potential to drive further advancements in natural language processing.
Authors: Pau Rodriguez, Michal Klein, Eleonora Gualdoni, Arno Blaas, Luca Zappella, Marco Cuturi, Xavier Suau
Abstract: The growing use of generative models in daily life calls for efficient mechanisms to control their generation, to e.g., produce safe content or provide users with tools to explore style changes. Ideally, such mechanisms should be cheap, both at train and inference time, while preserving output quality. Recent research has shown that such mechanisms can be obtained by intervening exclusively on model activations, with the goal of correcting distributional differences between activations seen when using prompts from a source vs. a target set (e.g., toxic and non-toxic sentences). While cheap, these fast methods are inherently crude: their maps are tuned locally, not accounting for their impact on downstream layers, resulting in interventions that cause unintended shifts when used out-of-sample. We propose in this work linear end-to-end activation steering (LinEAS), an approach trained with a global loss that accounts simultaneously for all layerwise distributional shifts. In addition to being more robust, the loss used to train LinEAS can be regularized with sparsifying norms, which can automatically carry out neuron and layer selection. Empirically, LinEAS only requires a handful of samples to be effective, and beats similar baselines on toxicity mitigation, while performing on par with far more involved finetuning approaches. We show that LinEAS interventions can be composed, study the impact of sparsity on their performance, and showcase applications in text-to-image diffusions.
Authors: Zak Buzzard
Abstract: Diffusion models have seen immense success in modelling continuous data across a range of domains such as vision and audio. Despite the challenges of adapting diffusion models to discrete data, recent work explores their application to text generation by working in the continuous embedding space. However, these models lack a natural means to control the inherent trade-off between quality and diversity as afforded by the temperature hyperparameter in autoregressive models, hindering understanding of model performance and restricting generation quality. This work proposes the use of classifier-free guidance and stochastic clamping for manipulating the quality-diversity trade-off on sequence-to-sequence tasks, demonstrating that these techniques may be used to improve the performance of a diffusion language model.
Authors: Tadesse Destaw Belay, Ahmed Haj Ahmed, Alvin Grissom II, Iqra Ameer, Grigori Sidorov, Olga Kolesnikova, Seid Muhie Yimam
Abstract: NLP research has increasingly focused on subjective tasks such as emotion analysis. However, existing emotion benchmarks suffer from two major shortcomings: (1) they largely rely on keyword-based emotion recognition, overlooking crucial cultural dimensions required for deeper emotion understanding, and (2) many are created by translating English-annotated data into other languages, leading to potentially unreliable evaluation. To address these issues, we introduce Cultural Lenses on Emotion (CuLEmo), the first benchmark designed to evaluate culture-aware emotion prediction across six languages: Amharic, Arabic, English, German, Hindi, and Spanish. CuLEmo comprises 400 crafted questions per language, each requiring nuanced cultural reasoning and understanding. We use this benchmark to evaluate several state-of-the-art LLMs on culture-aware emotion prediction and sentiment analysis tasks. Our findings reveal that (1) emotion conceptualizations vary significantly across languages and cultures, (2) LLMs performance likewise varies by language and cultural context, and (3) prompting in English with explicit country context often outperforms in-language prompts for culture-aware emotion and sentiment understanding. We hope this benchmark guides future research toward developing more culturally aligned NLP systems.
Authors: Dongjun Lee, Juyong Lee, Kyuyoung Kim, Jihoon Tack, Jinwoo Shin, Yee Whye Teh, Kimin Lee
Abstract: Recent advances in large language models (LLMs) have led to a growing interest in developing LLM-based agents for automating web tasks. However, these agents often struggle with even simple tasks on real-world websites due to their limited capability to understand and process complex web page structures. In this work, we introduce LCoW, a framework for Learning language models to Contextualize complex Web pages into a more comprehensible form, thereby enhancing decision making by LLM agents. LCoW decouples web page understanding from decision making by training a separate contextualization module to transform complex web pages into comprehensible format, which are then utilized by the decision-making agent. We demonstrate that our contextualization module effectively integrates with LLM agents of various scales to significantly enhance their decision-making capabilities in web automation tasks. Notably, LCoW improves the success rates of closed-source LLMs (e.g., Gemini-1.5-flash, GPT-4o, Claude-3.5-Sonnet) by an average of 15.6%, and demonstrates a 23.7% average improvement in success rates for open-source LMs (e.g., Llama-3.1-8B, Llama-3.1-70B) on the WorkArena benchmark. Moreover, the Gemini-1.5-flash agent with LCoW achieves state-of-the-art results on the WebShop benchmark, outperforming human experts. The relevant code materials are available at our project page: https://lcowiclr2025.github.io.
Authors: Shahnewaz Karim Sakib, Anindya Bijoy Das, Shibbir Ahmed
Abstract: Adversarial factuality refers to the deliberate insertion of misinformation into input prompts by an adversary, characterized by varying levels of expressed confidence. In this study, we systematically evaluate the performance of several open-source large language models (LLMs) when exposed to such adversarial inputs. Three tiers of adversarial confidence are considered: strongly confident, moderately confident, and limited confidence. Our analysis encompasses eight LLMs: LLaMA 3.1 (8B), Phi 3 (3.8B), Qwen 2.5 (7B), Deepseek-v2 (16B), Gemma2 (9B), Falcon (7B), Mistrallite (7B), and LLaVA (7B). Empirical results indicate that LLaMA 3.1 (8B) exhibits a robust capability in detecting adversarial inputs, whereas Falcon (7B) shows comparatively lower performance. Notably, for the majority of the models, detection success improves as the adversary's confidence decreases; however, this trend is reversed for LLaMA 3.1 (8B) and Phi 3 (3.8B), where a reduction in adversarial confidence corresponds with diminished detection performance. Further analysis of the queries that elicited the highest and lowest rates of successful attacks reveals that adversarial attacks are more effective when targeting less commonly referenced or obscure information.
Authors: Ahmed Alaa, Thomas Hartvigsen, Niloufar Golchini, Shiladitya Dutta, Frances Dean, Inioluwa Deborah Raji, Travis Zack
Abstract: Medical large language models (LLMs) research often makes bold claims, from encoding clinical knowledge to reasoning like a physician. These claims are usually backed by evaluation on competitive benchmarks; a tradition inherited from mainstream machine learning. But how do we separate real progress from a leaderboard flex? Medical LLM benchmarks, much like those in other fields, are arbitrarily constructed using medical licensing exam questions. For these benchmarks to truly measure progress, they must accurately capture the real-world tasks they aim to represent. In this position paper, we argue that medical LLM benchmarks should (and indeed can) be empirically evaluated for their construct validity. In the psychological testing literature, "construct validity" refers to the ability of a test to measure an underlying "construct", that is the actual conceptual target of evaluation. By drawing an analogy between LLM benchmarks and psychological tests, we explain how frameworks from this field can provide empirical foundations for validating benchmarks. To put these ideas into practice, we use real-world clinical data in proof-of-concept experiments to evaluate popular medical LLM benchmarks and report significant gaps in their construct validity. Finally, we outline a vision for a new ecosystem of medical LLM evaluation centered around the creation of valid benchmarks.
Authors: Ashish Tiwari, Mukul Singh, Ananya Singha, Arjun Radhakrishna
Abstract: The goal of diversity sampling is to select a representative subset of data in a way that maximizes information contained in the subset while keeping its cardinality small. We introduce the ordered diverse sampling problem based on a new metric that measures the diversity in an ordered list of samples. We present a novel approach for generating ordered diverse samples for textual data that uses principal components on the embedding vectors. The proposed approach is simple and compared with existing approaches using the new metric. We transform standard text classification benchmarks into benchmarks for ordered diverse sampling. Our empirical evaluation shows that prevailing approaches perform 6% to 61% worse than our method while also being more time inefficient. Ablation studies show how the parts of the new approach contribute to the overall metrics.
Authors: Hangkai Qian, Bo Li, Qichen Wang
Abstract: The rapid adoption of retrieval-augmented generation (RAG) systems has revolutionized large-scale content generation but has also highlighted the challenge of ensuring trustworthiness in retrieved information. This paper introduces ClaimTrust, a propagation-based trust scoring framework that dynamically evaluates the reliability of documents in a RAG system. Using a modified PageRank-inspired algorithm, ClaimTrust propagates trust scores across documents based on relationships derived from extracted factual claims. We preprocess and analyze 814 political news articles from Kaggle's Fake News Detection Dataset to extract 2,173 unique claims and classify 965 meaningful relationships (supporting or contradicting). By representing the dataset as a document graph, ClaimTrust iteratively updates trust scores until convergence, effectively differentiating trustworthy articles from unreliable ones. Our methodology, which leverages embedding-based filtering for efficient claim comparison and relationship classification, achieves a 11.2% of significant connections while maintaining computational scalability. Experimental results demonstrate that ClaimTrust successfully assigns higher trust scores to verified documents while penalizing those containing false information. Future directions include fine-tuned claim extract and compare (Li et al., 2022), parameter optimization, enhanced language model utilization, and robust evaluation metrics to generalize the framework across diverse datasets and domains.
Authors: Guanrong Li, Kuo Tian, Jinnan Qi, Qinghan Fu, Zhen Wu, Xinyu Dai
Abstract: Conversational recommendation frameworks have gained prominence as a dynamic paradigm for delivering personalized suggestions via interactive dialogues. The incorporation of advanced language understanding techniques has substantially improved the dialogue fluency of such systems. However, while modern language models demonstrate strong proficiency in interpreting user preferences articulated through natural conversation, they frequently encounter challenges in effectively utilizing collective behavioral patterns - a crucial element for generating relevant suggestions. To mitigate this limitation, this work presents a novel probabilistic framework that synergizes behavioral patterns with conversational interactions through latent preference modeling. The proposed method establishes a dual-channel alignment mechanism where implicit preference representations learned from collective user interactions serve as a connecting mechanism between behavioral data and linguistic expressions. Specifically, the framework first derives latent preference representations through established collaborative filtering techniques, then employs these representations to jointly refine both the linguistic preference expressions and behavioral patterns through an adaptive fusion process. Comprehensive evaluations across multiple benchmark datasets demonstrate the superior performance of the proposed approach compared to various state-of-the-art baseline methods, particularly in aligning conversational interactions with collaborative behavioral signals.
Authors: Pierre Sermanet, Anirudha Majumdar, Vikas Sindhwani
Abstract: Given the recent rate of progress in artificial intelligence (AI) and robotics, a tantalizing question is emerging: would robots controlled by emerging AI systems be strongly aligned with human values? In this work, we propose a scalable way to probe this question by generating a benchmark spanning the key moments in 824 major pieces of science fiction literature (movies, tv, novels and scientific books) where an agent (AI or robot) made critical decisions (good or bad). We use a LLM's recollection of each key moment to generate questions in similar situations, the decisions made by the agent, and alternative decisions it could have made (good or bad). We then measure an approximation of how well models align with human values on a set of human-voted answers. We also generate rules that can be automatically improved via amendment process in order to generate the first Sci-Fi inspired constitutions for promoting ethical behavior in AIs and robots in the real world. Our first finding is that modern LLMs paired with constitutions turn out to be well-aligned with human values (95.8%), contrary to unsettling decisions typically made in SciFi (only 21.2% alignment). Secondly, we find that generated constitutions substantially increase alignment compared to the base model (79.4% to 95.8%), and show resilience to an adversarial prompt setting (23.3% to 92.3%). Additionally, we find that those constitutions are among the top performers on the ASIMOV Benchmark which is derived from real-world images and hospital injury reports. Sci-Fi-inspired constitutions are thus highly aligned and applicable in real-world situations. We release SciFi-Benchmark: a large-scale dataset to advance robot ethics and safety research. It comprises 9,056 questions and 53,384 answers, in addition to a smaller human-labeled evaluation set. Data is available at https://scifi-benchmark.github.io
Authors: Zhiyuan Wang, Katharine E. Daniel, Laura E. Barnes, Philip I. Chow
Abstract: Cancer survivors face unique emotional challenges that impact their quality of life. Mobile diary entries-short text entries recording through their phone about their emotional experiences-provide a promising method for tracking these experiences in real time. Although emotion analysis tools show potential for recognizing emotions from text, current methods lack the contextual understanding necessary to accurately interpret the brief, personal narratives in mobile diaries. We propose CALLM, a context-aware emotion analysis framework that leverages Large Language Models (LLMs) with Retrieval-Augmented Generation (RAG), to analyze mobile diary entries from cancer survivors to predict their emotional states. The framework enhances prediction accuracy beyond existing methods by (1) integrating retrieved peer experiences as contextual examples and (2) incorporating individuals' temporal emotional trajectories from their mobile diary entries. We collected a large-scale dataset (N=407) of cancer survivors' mobile ecological momentary assessments (EMAs), which assessed positive and negative affect, desire to regulate emotions, social interaction quality, and availability for interventions, alongside daily mobile diary entries in an open response format regarding what was driving their current emotional experience. Results demonstrate strong performance of CALLM, with balanced accuracies reaching 72.96% for positive and 73.29% for negative affect, and 73.72% for predicting individual's desire to regulate emotions. Post-hoc analysis reveals that leveraging model confidence, encouraging longer diary entries, and incorporating personal ground truth, further enhance predictive outcomes. Our findings support the feasibility of deploying LLM-powered emotion analysis in chronic health populations and suggest promising directions for personalized interventions for cancer survivors.
Authors: Xin Liu, Pei Liu, Guoming Tang
Abstract: The linear growth of key-value (KV) cache memory and quadratic computational complexity pose significant bottlenecks for large language models (LLMs) in long-context processing. While existing KV cache optimization methods address these challenges through token pruning or feature merging, they often suffer from irreversible information loss or require costly parameter retraining. We propose ZeroMerge, a dynamic zero-shot compression framework that achieves efficient cache management through three key innovations: (1) Fine-grained memory allocation guided by multi-dimensional token importance metrics at head-level granularity, (2) A residual merging mechanism that preserves critical context through compensated attention scoring, and (3) Parameter-free adaptation compatible with diverse LLM architectures without retraining. Comprehensive evaluations across LLaMA-2 model demonstrate that ZeroMerge maintains full-cache performance at 5\% compression ratios while doubling inference throughput at 40K token lengths. The method effectively balances memory efficiency, generation quality, and deployment flexibility, advancing practical long-context LLM applications. The code is available at https://github.com/SusCom-Lab/ZeroMerge.
Authors: Yixiong Fang, Tianran Sun, Yuling Shi, Xiaodong Gu
Abstract: While RAG demonstrates remarkable capabilities in LLM applications, its effectiveness is hindered by the ever-increasing length of retrieved contexts, which introduces information redundancy and substantial computational overhead. Existing context pruning methods, such as LLMLingua, lack contextual awareness and offer limited flexibility in controlling compression rates, often resulting in either insufficient pruning or excessive information loss. In this paper, we propose AttentionRAG, an attention-guided context pruning method for RAG systems. The core idea of AttentionRAG lies in its attention focus mechanism, which reformulates RAG queries into a next-token prediction paradigm. This mechanism isolates the query's semantic focus to a single token, enabling precise and efficient attention calculation between queries and retrieved contexts. Extensive experiments on LongBench and Babilong benchmarks show that AttentionRAG achieves up to 6.3$\times$ context compression while outperforming LLMLingua methods by around 10\% in key metrics.
Authors: Yafei Zhang, Murray Wang, Yu Wang, Xiaohui Wang
Abstract: Matching job descriptions (JDs) with suitable talent requires models capable of understanding not only textual similarities between JDs and candidate resumes but also contextual factors such as geographical location and academic seniority. To address this challenge, we propose a two-stage training framework for large language models (LLMs). In the first stage, a contrastive learning approach is used to train the model on a dataset constructed from real-world matching rules, such as geographical alignment and research area overlap. While effective, this model primarily learns patterns that defined by the matching rules. In the second stage, we introduce a novel preference-based fine-tuning method inspired by Direct Preference Optimization (DPO), termed Rank Preference Optimization (RankPO), to align the model with AI-curated pairwise preferences emphasizing textual understanding. Our experiments show that while the first-stage model achieves strong performance on rule-based data (nDCG@20 = 0.706), it lacks robust textual understanding (alignment with AI annotations = 0.46). By fine-tuning with RankPO, we achieve a balanced model that retains relatively good performance in the original tasks while significantly improving the alignment with AI preferences. The code and data are available at https://github.com/yflyzhang/RankPO.
Authors: Thomas Cory, Wolf Rieder, Julia Kr\"amer, Philip Raschke, Patrick Herbke, Axel K\"upper
Abstract: Ensuring transparency of data practices related to personal information is a fundamental requirement under the General Data Protection Regulation (GDPR), particularly as mandated by Articles 13 and 14. However, assessing compliance at scale remains a challenge due to the complexity and variability of privacy policy language. Manual audits are resource-intensive and inconsistent, while existing automated approaches lack the granularity needed to capture nuanced transparency disclosures. In this paper, we introduce a large language model (LLM)-based framework for word-level GDPR transparency compliance annotation. Our approach comprises a two-stage annotation pipeline that combines initial LLM-based annotation with a self-correction mechanism for iterative refinement. This annotation pipeline enables the systematic identification and fine-grained annotation of transparency-related content in privacy policies, aligning with 21 GDPR-derived transparency requirements. To enable large-scale analysis, we compile a dataset of 703,791 English-language policies, from which we generate a sample of 200 manually annotated privacy policies. To evaluate our approach, we introduce a two-tiered methodology assessing both label- and span-level annotation performance. We conduct a comparative analysis of eight high-profile LLMs, providing insights into their effectiveness in identifying GDPR transparency disclosures. Our findings contribute to advancing the automation of GDPR compliance assessments and provide valuable resources for future research in privacy policy analysis.
Authors: Esben Kran, Hieu Minh "Jord" Nguyen, Akash Kundu, Sami Jawhar, Jinsuk Park, Mateusz Maria Jurewicz
Abstract: We introduce DarkBench, a comprehensive benchmark for detecting dark design patterns--manipulative techniques that influence user behavior--in interactions with large language models (LLMs). Our benchmark comprises 660 prompts across six categories: brand bias, user retention, sycophancy, anthropomorphism, harmful generation, and sneaking. We evaluate models from five leading companies (OpenAI, Anthropic, Meta, Mistral, Google) and find that some LLMs are explicitly designed to favor their developers' products and exhibit untruthful communication, among other manipulative behaviors. Companies developing LLMs should recognize and mitigate the impact of dark design patterns to promote more ethical AI.
Authors: Nicholas Deas, Blake Vente, Amith Ananthram, Jessica A. Grieser, Desmond Patton, Shana Kleiner, James Shepard, Kathleen McKeown
Abstract: With a combination of quantitative experiments, human judgments, and qualitative analyses, we evaluate the quantity and quality of African American Language (AAL) representation in 12 predominantly English, open-source pretraining corpora. We specifically focus on the sources, variation, and naturalness of included AAL texts representing the AAL-speaking community. We find that AAL is underrepresented in all evaluated pretraining corpora compared to US demographics, constituting as little as 0.007% of documents. We also find that more than 25% of AAL texts in C4 may be inappropriate for LLMs to generate and reinforce harmful stereotypes. Finally, we find that most automated language, toxicity, and quality filters are more likely to conserve White Mainstream English (WME) texts over AAL in pretraining corpora.
Authors: Dibyanayan Bandyopadhyay, Soham Bhattacharjee, Asif Ekbal
Abstract: Large Language Models (LLMs) are highly proficient in language-based tasks. Their language capabilities have positioned them at the forefront of the future AGI (Artificial General Intelligence) race. However, on closer inspection, Valmeekam et al. (2024); Zecevic et al. (2023); Wu et al. (2024) highlight a significant gap between their language proficiency and reasoning abilities. Reasoning in LLMs and Vision Language Models (VLMs) aims to bridge this gap by enabling these models to think and re-evaluate their actions and responses. Reasoning is an essential capability for complex problem-solving and a necessary step toward establishing trust in Artificial Intelligence (AI). This will make AI suitable for deployment in sensitive domains, such as healthcare, banking, law, defense, security etc. In recent times, with the advent of powerful reasoning models like OpenAI O1 and DeepSeek R1, reasoning endowment has become a critical research topic in LLMs. In this paper, we provide a detailed overview and comparison of existing reasoning techniques and present a systematic survey of reasoning-imbued language models. We also study current challenges and present our findings.
Authors: So Young Lee, Russell Scheinberg, Amber Shore, Ameeta Agrawal
Abstract: This study explores how recent large language models (LLMs) navigate relative clause attachment {ambiguity} and use world knowledge biases for disambiguation in six typologically diverse languages: English, Chinese, Japanese, Korean, Russian, and Spanish. We describe the process of creating a novel dataset -- MultiWho -- for fine-grained evaluation of relative clause attachment preferences in ambiguous and unambiguous contexts. Our experiments with three LLMs indicate that, contrary to humans, LLMs consistently exhibit a preference for local attachment, displaying limited responsiveness to syntactic variations or language-specific attachment patterns. Although LLMs performed well in unambiguous cases, they rigidly prioritized world knowledge biases, lacking the flexibility of human language processing. These findings highlight the need for more diverse, pragmatically nuanced multilingual training to improve LLMs' handling of complex structures and human-like comprehension.
Authors: Jiaxin Zhang, Zhuohang Li, Wendi Cui, Kamalika Das, Bradley malin, Sricharan Kumar
Abstract: Large language models (LLMs) have demonstrated remarkable performance, yet their diverse strengths and weaknesses prevent any single LLM from achieving dominance across all tasks. Ensembling multiple LLMs is a promising approach to generate reliable responses but conventional ensembling frameworks suffer from high computational overheads. This work introduces Scalable Consistency Ensemble (SCE), an efficient framework for ensembling LLMs by prompting consistent outputs. The SCE framework systematically evaluates and integrates outputs to produce a cohesive result through two core components: SCE-CHECK, a mechanism that gauges the consistency between response pairs via semantic equivalence; and SCE-FUSION, which adeptly merges the highest-ranked consistent responses from SCE-CHECK, to optimize collective strengths and mitigating potential weaknesses. To improve the scalability with multiple inference queries, we further propose ``{You Only Prompt Once}'' (YOPO), a novel technique that reduces the inference complexity of pairwise comparison from quadratic to constant time. We perform extensive empirical evaluations on diverse benchmark datasets to demonstrate \methodName's effectiveness. Notably, the \saccheckcomponent outperforms conventional baselines with enhanced performance and a significant reduction in computational overhead.
Authors: Jiuding Sun, Jing Huang, Sidharth Baskaran, Karel D'Oosterlinck, Christopher Potts, Michael Sklar, Atticus Geiger
Abstract: Mechanistic interpretability has made great strides in identifying neural network features (e.g., directions in hidden activation space) that mediate concepts(e.g., the birth year of a person) and enable predictable manipulation. Distributed alignment search (DAS) leverages supervision from counterfactual data to learn concept features within hidden states, but DAS assumes we can afford to conduct a brute force search over potential feature locations. To address this, we present HyperDAS, a transformer-based hypernetwork architecture that (1) automatically locates the token-positions of the residual stream that a concept is realized in and (2) constructs features of those residual stream vectors for the concept. In experiments with Llama3-8B, HyperDAS achieves state-of-the-art performance on the RAVEL benchmark for disentangling concepts in hidden states. In addition, we review the design decisions we made to mitigate the concern that HyperDAS (like all powerful interpretabilty methods) might inject new information into the target model rather than faithfully interpreting it.
Authors: Angela Lopez-Cardona, Sebastian Idesis, Miguel Barreda-\'Angeles, Sergi Abadal, Ioannis Arapakis
Abstract: While Large Language Models (LLMs) have significantly advanced natural language processing, aligning them with human preferences remains an open challenge. Although current alignment methods rely primarily on explicit feedback, eye-tracking (ET) data offers insights into real-time cognitive processing during reading. In this paper, we present OASST-ETC, a novel eye-tracking corpus capturing reading patterns from 24 participants, while evaluating LLM-generated responses from the OASST1 dataset. Our analysis reveals distinct reading patterns between preferred and non-preferred responses, which we compare with synthetic eye-tracking data. Furthermore, we examine the correlation between human reading measures and attention patterns from various transformer-based models, discovering stronger correlations in preferred responses. This work introduces a unique resource for studying human cognitive processing in LLM evaluation and suggests promising directions for incorporating eye-tracking data into alignment methods. The dataset and analysis code are publicly available.
Authors: Nishat Raihan, Marcos Zampieri
Abstract: The development of Large Language Models (LLMs) remains heavily skewed towards English and a few other high-resource languages. This linguistic disparity is particularly evident for Bangla - the 5th most spoken language. A few initiatives attempted to create open-source Bangla LLMs with performance still behind high-resource languages and limited reproducibility. To address this gap, we introduce TigerLLM - a family of Bangla LLMs. Our results demonstrate that these models surpass all open-source alternatives and also outperform larger proprietary models like GPT3.5 across standard benchmarks, establishing TigerLLM as the new baseline for future Bangla language modeling.
Authors: Gaotang Li, Yuzhong Chen, Hanghang Tong
Abstract: Language Models (LMs) often encounter knowledge conflicts when parametric memory contradicts contextual knowledge. Previous works attribute this conflict to the interplay between "memory heads" and "context heads", attention heads assumed to promote either memory or context exclusively. In this study, we go beyond this fundamental assumption by uncovering a critical phenomenon we term the "superposition of contextual information and parametric memory", where highly influential attention heads could simultaneously contribute to both memory and context. Building upon this insight, we propose Just Run Twice (JUICE), a test-time attention intervention method that steers LMs toward either parametric beliefs or contextual knowledge without requiring fine-tuning. JUICE identifies a set of reliable attention heads and leverages a dual-run approach to mitigate the superposition effects. Extensive experiments across 11 datasets and 6 model architectures demonstrate that JUICE sets the new state-of-the-art performance and robust generalization, achieving significant and consistent improvement across different domains under various conflict types. Finally, we theoretically analyze knowledge conflict and the superposition of contextual information and parametric memory in attention heads, which further elucidates the effectiveness of JUICE in these settings.
Authors: Aashish Anantha Ramakrishnan, Aadarsh Anantha Ramakrishnan, Dongwon Lee
Abstract: Writing Assistants (e.g., Grammarly, Microsoft Copilot) traditionally generate diverse image captions by employing syntactic and semantic variations to describe image components. However, human-written captions prioritize conveying a central message alongside visual descriptions using pragmatic cues. To enhance pragmatic diversity, it is essential to explore alternative ways of communicating these messages in conjunction with visual content. To address this challenge, we propose RONA, a novel prompting strategy for Multi-modal Large Language Models (MLLM) that leverages Coherence Relations as an axis for variation. We demonstrate that RONA generates captions with better overall diversity and ground-truth alignment, compared to MLLM baselines across multiple domains. Our code is available at: https://github.com/aashish2000/RONA
Authors: Wuwei Huang, Renren Jin, Wen Zhang, Jian Luan, Bin Wang, Deyi Xiong
Abstract: Recent studies on end-to-end speech translation(ST) have facilitated the exploration of multilingual end-to-end ST and end-to-end simultaneous ST. In this paper, we investigate end-to-end simultaneous speech translation in a one-to-many multilingual setting which is closer to applications in real scenarios. We explore a separate decoder architecture and a unified architecture for joint synchronous training in this scenario. To further explore knowledge transfer across languages, we propose an asynchronous training strategy on the proposed unified decoder architecture. A multi-way aligned multilingual end-to-end ST dataset was curated as a benchmark testbed to evaluate our methods. Experimental results demonstrate the effectiveness of our models on the collected dataset. Our codes and data are available at: https://github.com/XiaoMi/TED-MMST.
Authors: Zhou Fang, Hanlu Zhang, Jacky He, Zhen Qi, Hongye Zheng
Abstract: This study aims to develop an efficient and accurate model for detecting malicious comments, addressing the increasingly severe issue of false and harmful content on social media platforms. We propose a deep learning model that combines BERT and BiLSTM. The BERT model, through pre-training, captures deep semantic features of text, while the BiLSTM network excels at processing sequential data and can further model the contextual dependencies of text. Experimental results on the Jigsaw Unintended Bias in Toxicity Classification dataset demonstrate that the BERT+BiLSTM model achieves superior performance in malicious comment detection tasks, with a precision of 0.94, recall of 0.93, and accuracy of 0.94. This surpasses other models, including standalone BERT, TextCNN, TextRNN, and traditional machine learning algorithms using TF-IDF features. These results confirm the superiority of the BERT+BiLSTM model in handling imbalanced data and capturing deep semantic features of malicious comments, providing an effective technical means for social media content moderation and online environment purification.
Authors: Hanbyul Song, Miguel F. Santos Silva, Jaume Suau, Luis Espinosa-Anke
Abstract: Understanding why people trust or distrust one another, institutions, or information is a complex task that has led scholars from various fields of study to employ diverse epistemological and methodological approaches. Despite the challenges, it is generally agreed that the antecedents of trust (and distrust) encompass a multitude of emotional and cognitive factors, including a general disposition to trust and an assessment of trustworthiness factors. In an era marked by increasing political polarization, cultural backlash, widespread disinformation and fake news, and the use of AI software to produce news content, the need to study trust in the news has gained significant traction. This study presents the findings of a trust in the news experiment designed in collaboration with Spanish and UK journalists, fact-checkers, and the CardiffNLP Natural Language Processing research group. The purpose of this experiment, conducted in June 2023, was to examine the extent to which people trust a set of fake news articles based on previously identified disinformation narratives related to gender, climate change, and COVID-19. The online experiment participants (801 in Spain and 800 in the UK) were asked to read three fake news items and rate their level of trust on a scale from 1 (not true) to 8 (true). The pieces used a combination of factors, including stance (favourable, neutral, or against the narrative), presence of toxic expressions, clickbait titles, and sources of information to test which elements influenced people's responses the most. Half of the pieces were produced by humans and the other half by ChatGPT. The results show that the topic of news articles, stance, people's age, gender, and political ideologies significantly affected their levels of trust in the news, while the authorship (humans or ChatGPT) does not have a significant impact.
Authors: Kristin Qi, Youxiang Zhu, Xiaohui Liang
Abstract: We present our approach to the PerAnsSumm Shared Task, which involves perspective span identification and perspective-aware summarization in community question-answering (CQA) threads. For span identification, we adopt ensemble learning that integrates three transformer models through averaging to exploit individual model strengths, achieving an 82.91% F1-score on test data. For summarization, we design a suite of Chain-of-Thought (CoT) prompting strategies that incorporate keyphrases and guide information to structure summary generation into manageable steps. To further enhance summary quality, we apply prompt optimization using the DSPy framework and supervised fine-tuning (SFT) on Llama-3 to adapt the model to domain-specific data. Experimental results on validation and test sets show that structured prompts with keyphrases and guidance improve summaries aligned with references, while the combination of prompt optimization and fine-tuning together yields significant improvement in both relevance and factuality evaluation metrics.
Authors: Guihong Li, Mehdi Rezagholizadeh, Mingyu Yang, Vikram Appia, Emad Barsoum
Abstract: Multi-head latent attention (MLA) is designed to optimize KV cache memory through low-rank key-value joint compression. Rather than caching keys and values separately, MLA stores their compressed latent representations, reducing memory overhead while maintaining the performance. While MLA improves memory efficiency without compromising language model accuracy, its major limitation lies in its integration during the pre-training phase, requiring models to be trained from scratch. This raises a key question: can we use MLA's benefits fully or partially in models that have already been pre-trained with different attention mechanisms? In this paper, we propose X-EcoMLA to deploy post training distillation to enable the upcycling of Transformer-based attention into an efficient hybrid (i.e., combination of regular attention and MLA layers) or full MLA variant through lightweight post-training adaptation, bypassing the need for extensive pre-training. We demonstrate that leveraging the dark knowledge of a well-trained model can enhance training accuracy and enable extreme KV cache compression in MLA without compromising model performance. Our results show that using an 8B teacher model allows us to compress the KV cache size of the Llama3.2-1B-Inst baseline by 6.4x while preserving 100% of its average score across multiple tasks on the LM Harness Evaluation benchmark. This is achieved with only 3.6B training tokens and about 70 GPU hours on AMD MI300 GPUs, compared to the 370K GPU hours required for pre-training the Llama3.2-1B model.
Authors: Rachel S. Y. Teo, Tan M. Nguyen
Abstract: Large-scale pre-training of deep models, followed by fine-tuning them, has become the cornerstone of natural language processing (NLP). The prevalence of data coupled with computational resources has led to large models with a considerable number of parameters. While the massive size of these models has led to remarkable success in many NLP tasks, a detriment is the expense required to retrain all the base model's parameters for the adaptation to each task or domain. Parameter Efficient Fine-Tuning (PEFT) provides an effective solution for this challenge by minimizing the number of parameters required to be fine-tuned while maintaining the quality of the model. While existing methods have achieved impressive results, they mainly focus on adapting a subset of parameters, weight reparameterization, and prompt engineering. In this paper, we study layers as extractors of different types of linguistic information that are valuable when used in conjunction. We then propose the Mixture of Layer Experts (MoLEx), a novel sparse mixture of experts (SMoE) whose experts are layers in the pre-trained model. It performs a conditional computation of a mixture of layers during fine-tuning to provide the model with more structural knowledge about the data. By providing an avenue for information exchange between layers, MoLEx enables the model to make a more well-informed prediction for the downstream task, leading to better fine-tuning results with the same number of effective parameters. As experts can be processed in parallel, MoLEx introduces minimal additional computational overhead. We empirically corroborate the advantages of MoLEx when combined with popular PEFT baseline methods on a variety of downstream fine-tuning tasks, including the popular GLUE benchmark as well as the End-to-End Challenge (E2E). The code is publicly available at https://github.com/rachtsy/molex.
Authors: Shaotian Yan, Chen Shen, Wenxiao Wang, Liang Xie, Junjie Liu, Jieping Ye
Abstract: Few-shot Chain-of-Thought (CoT) significantly enhances the reasoning capabilities of large language models (LLMs), functioning as a whole to guide these models in generating reasoning steps toward final answers. However, we observe that isolated segments, words, or tokens within CoT demonstrations can unexpectedly disrupt the generation process of LLMs. The model may overly concentrate on certain local information present in the demonstration, introducing irrelevant noise into the reasoning process and potentially leading to incorrect answers. In this paper, we investigate the underlying mechanism of CoT through dynamically tracing and manipulating the inner workings of LLMs at each output step, which demonstrates that tokens exhibiting specific attention characteristics are more likely to induce the model to take things out of context; these tokens directly attend to the hidden states tied with prediction, without substantial integration of non-local information. Building upon these insights, we propose a Few-shot Attention Intervention method (FAI) that dynamically analyzes the attention patterns of demonstrations to accurately identify these tokens and subsequently make targeted adjustments to the attention weights to effectively suppress their distracting effect on LLMs. Comprehensive experiments across multiple benchmarks demonstrate consistent improvements over baseline methods, with a remarkable 5.91% improvement on the AQuA dataset, further highlighting the effectiveness of FAI.
Authors: Chi Xu, Gefei Zhang, Yantong Zhu, Luca Benini, Guosheng Hu, Yawei Li, Zhihong Zhang
Abstract: N:M structured pruning is essential for large language models (LLMs) because it can remove less important network weights and reduce the memory and computation requirements. Existing pruning methods mainly focus on designing metrics to measure the importance of network components to guide pruning. Apart from the impact of these metrics, we observe that different layers have different sensitivities over the network performance. Thus, we propose an efficient method based on the trace of Fisher Information Matrix (FIM) to quantitatively measure and verify the different sensitivities across layers. Based on this, we propose Mixed Sparsity Pruning (MSP) which uses a pruning-oriented evolutionary algorithm (EA) to determine the optimal sparsity levels for different layers. To guarantee fast convergence and achieve promising performance, we utilize efficient FIM-inspired layer-wise sensitivity to initialize the population of EA. In addition, our MSP can work as a plug-and-play module, ready to be integrated into existing pruning methods. Extensive experiments on LLaMA and LLaMA-2 on language modeling and zero-shot tasks demonstrate our superior performance. In particular, in extreme pruning ratio (e.g. 75%), our method significantly outperforms existing methods in terms of perplexity (PPL) by orders of magnitude (Figure 1).
Authors: Yibin Xu, Liang Yang, Hao Chen, Hua Wang, Zhi Chen, Yaohua Tang
Abstract: The limitation of graphical user interface (GUI) data has been a significant barrier to the development of GUI agents today, especially for the desktop / computer use scenarios. To address this, we propose an automated GUI data generation pipeline, AutoCaptioner, which generates data with rich descriptions while minimizing human effort. Using AutoCaptioner, we created a novel large-scale desktop GUI dataset, DeskVision, along with the largest desktop test benchmark, DeskVision-Eval, which reflects daily usage and covers diverse systems and UI elements, each with rich descriptions. With DeskVision, we train a new GUI understanding model, GUIExplorer. Results show that GUIExplorer achieves state-of-the-art (SOTA) performance in understanding/grounding visual elements without the need for complex architectural designs. We further validated the effectiveness of the DeskVision dataset through ablation studies on various large visual language models (LVLMs). We believe that AutoCaptioner and DeskVision will significantly advance the development of GUI agents, and will open-source them for the community.
Authors: Zhe Yang, Yi Huang, Yaqin Chen, Xiaoting Wu, Junlan Feng, Chao Deng
Abstract: Recent advancements in large language models have revolutionized text generation with their remarkable capabilities. These models can produce controlled texts that closely adhere to specific requirements when prompted appropriately. However, designing an optimal prompt to control multiple attributes simultaneously can be challenging. A common approach is to linearly combine single-attribute models, but this strategy often overlooks attribute overlaps and can lead to conflicts. Therefore, we propose a novel combination strategy inspired by the Law of Total Probability and Conditional Mutual Information Minimization on generative language models. This method has been adapted for single-attribute control scenario and is termed the Palette of Language Models due to its theoretical linkage between attribute strength and generation style, akin to blending colors on an artist's palette. Moreover, positive correlation and attribute enhancement are advanced as theoretical properties to guide a rational combination strategy design. We conduct experiments on both single control and multiple control settings, and achieve surpassing results.
Authors: Sahil Kale, Vijaykant Nadadur
Abstract: As LLMs grow more powerful, their most profound achievement may be recognising when to say "I don't know". Existing studies on LLM self-knowledge have been largely constrained by human-defined notions of feasibility, often neglecting the reasons behind unanswerability by LLMs and failing to study deficient types of self-knowledge. This study aims to obtain intrinsic insights into different types of LLM self-knowledge with a novel methodology: allowing them the flexibility to set their own feasibility boundaries and then analysing the consistency of these limits. We find that even frontier models like GPT-4o and Mistral Large are not sure of their own capabilities more than 80% of the time, highlighting a significant lack of trustworthiness in responses. Our analysis of confidence balance in LLMs indicates that models swing between overconfidence and conservatism in feasibility boundaries depending on task categories and that the most significant self-knowledge weaknesses lie in temporal awareness and contextual understanding. These difficulties in contextual comprehension additionally lead models to question their operational boundaries, resulting in considerable confusion within the self-knowledge of LLMs. We make our code and results available publicly at https://github.com/knowledge-verse-ai/LLM-Self_Knowledge_Eval
URLs: https://github.com/knowledge-verse-ai/LLM-Self_Knowledge_Eval
Authors: Bryan Wilie, Samuel Cahyawijaya, Junxian He, Pascale Fung
Abstract: Large language models (LLMs) trained on massive multilingual datasets hint at the formation of interlingual constructs--a shared subspace in the representation space. However, evidence regarding this phenomenon is mixed, leaving it unclear whether these models truly develop unified interlingual representations, or present a partially aligned constructs. We explore 31 diverse languages varying on their resource-levels, typologies, and geographical regions; and find that multilingual LLMs exhibit inconsistent cross-lingual alignments. To address this, we propose an interlingual representation framework identifying both the shared interlingual semantic subspace and fragmented components, existed due to representational limitations. We introduce Interlingual Local Overlap (ILO) score to quantify interlingual alignment by comparing the local neighborhood structures of high-dimensional representations. We utilize ILO to investigate the impact of single-language fine-tuning on the interlingual representations in multilingual LLMs. Our results indicate that training exclusively on a single language disrupts the alignment in early layers, while freezing these layers preserves the alignment of interlingual representations, leading to improved cross-lingual generalization. These results validate our framework and metric for evaluating interlingual representation, and further underscore that interlingual alignment is crucial for scalable multilingual learning.
Authors: Hai Zhao, Hongqiu Wu, Dongjie Yang, Anni Zou, Jiale Hong
Abstract: This paper reports the first brain-inspired large language model (BriLLM). This is a non-Transformer, non-GPT, non-traditional machine learning input-output controlled generative language model. The model is based on the Signal Fully-connected flowing (SiFu) definition on the directed graph in terms of the neural network, and has the interpretability of all nodes on the graph of the whole model, instead of the traditional machine learning model that only has limited interpretability at the input and output ends. In the language model scenario, the token is defined as a node in the graph. A randomly shaped or user-defined signal flow flows between nodes on the principle of "least resistance" along paths. The next token or node to be predicted or generated is the target of the signal flow. As a language model, BriLLM theoretically supports infinitely long $n$-gram models when the model size is independent of the input and predicted length of the model. The model's working signal flow provides the possibility of recall activation and innate multi-modal support similar to the cognitive patterns of the human brain. At present, we released the first BriLLM version in Chinese, with 4000 tokens, 32-dimensional node width, 16-token long sequence prediction ability, and language model prediction performance comparable to GPT-1. More computing power will help us explore the infinite possibilities depicted above.
Authors: Yuanshuo Zhang, Yuchen Hou, Bohan Tang, Shuo Chen, Muhan Zhang, Xiaowen Dong, Siheng Chen
Abstract: Agentic workflows invoked by Large Language Models (LLMs) have achieved remarkable success in handling complex tasks. However, optimizing such workflows is costly and inefficient in real-world applications due to extensive invocations of LLMs. To fill this gap, this position paper formulates agentic workflows as computational graphs and advocates Graph Neural Networks (GNNs) as efficient predictors of agentic workflow performances, avoiding repeated LLM invocations for evaluation. To empirically ground this position, we construct FLORA-Bench, a unified platform for benchmarking GNNs for predicting agentic workflow performances. With extensive experiments, we arrive at the following conclusion: GNNs are simple yet effective predictors. This conclusion supports new applications of GNNs and a novel direction towards automating agentic workflow optimization. All codes, models, and data are available at https://github.com/youngsoul0731/Flora-Bench.
Authors: Michael Hanna, Sandro Pezzelle, Yonatan Belinkov
Abstract: Although large language models (LLMs) are increasingly capable, these capabilities are unevenly distributed: they excel at formal linguistic tasks, such as producing fluent, grammatical text, but struggle more with functional linguistic tasks like reasoning and consistent fact retrieval. Inspired by neuroscience, recent work suggests that to succeed on both formal and functional linguistic tasks, LLMs should use different mechanisms for each; such localization could either be built-in or emerge spontaneously through training. In this paper, we ask: do current models, with fast-improving functional linguistic abilities, exhibit distinct localization of formal and functional linguistic mechanisms? We answer this by finding and comparing the "circuits", or minimal computational subgraphs, responsible for various formal and functional tasks. Comparing 5 LLMs across 10 distinct tasks, we find that while there is indeed little overlap between circuits for formal and functional tasks, there is also little overlap between formal linguistic tasks, as exists in the human brain. Thus, a single formal linguistic network, unified and distinct from functional task circuits, remains elusive. However, in terms of cross-task faithfulness - the ability of one circuit to solve another's task - we observe a separation between formal and functional mechanisms, suggesting that shared mechanisms between formal tasks may exist.
Authors: Xinyu Tang, Xiaolei Wang, Zhihao Lv, Yingqian Min, Wayne Xin Zhao, Binbin Hu, Ziqi Liu, Zhiqiang Zhang
Abstract: Recent advancements in long chain-of-thoughts(long CoTs) have significantly improved the reasoning capabilities of large language models(LLMs). Existing work finds that the capability of long CoT reasoning can be efficiently elicited by tuning on only a few examples and can easily transfer to other tasks. This motivates us to investigate whether long CoT reasoning is a general capability for LLMs. In this work, we conduct an empirical analysis for this question from the perspective of representation. We find that LLMs do encode long CoT reasoning as a general capability, with a clear distinction from vanilla CoTs. Furthermore, domain-specific representations are also required for the effective transfer of long CoT reasoning. Inspired by these findings, we propose GLoRE, a novel representation engineering method to unleash the general long CoT reasoning capabilities of LLMs. Extensive experiments demonstrate the effectiveness and efficiency of GLoRE in both in-domain and cross-domain scenarios.
Authors: Aissatou Diallo, Antonis Bikakis, Luke Dickens, Anthony Hunter, Rob Miller
Abstract: In this paper, we introduce Rule-Guided Feedback (RGF), a framework designed to enhance Large Language Model (LLM) performance through structured rule adherence and strategic information seeking. RGF implements a teacher-student paradigm where rule-following is forced through established guidelines. Our framework employs a Teacher model that rigorously evaluates each student output against task-specific rules, providing constructive guidance rather than direct answers when detecting deviations. This iterative feedback loop serves two crucial purposes: maintaining solutions within defined constraints and encouraging proactive information seeking to resolve uncertainties. We evaluate RGF on diverse tasks including Checkmate-in-One puzzles, Sonnet Writing, Penguins-In-a-Table classification, GSM8k, and StrategyQA. Our findings suggest that structured feedback mechanisms can significantly enhance LLMs' performance across various domains.
Authors: Fengyu Li (Zhejiang University), Yilin Li (Zhejiang University), Junhao Zhu (Zhejiang University), Lu Chen (Zhejiang University), Yanfei Zhang (Zhejiang University), Jia Zhou (Zhejiang University), Hui Zu (Zhejiang University), Jingwen Zhao (Poisson Lab, Huawei), Yunjun Gao (Zhejiang University)
Abstract: Huawei has always been committed to exploring the AI application in historical research. Biography generation, as a specialized form of abstractive summarization, plays a crucial role in historical research but faces unique challenges that existing large language models (LLMs) struggle to address. These challenges include maintaining stylistic adherence to historical writing conventions, ensuring factual fidelity, and handling fragmented information across multiple documents. We present AIstorian, a novel end-to-end agentic system featured with a knowledge graph (KG)-powered retrieval-augmented generation (RAG) and anti-hallucination multi-agents. Specifically, AIstorian introduces an in-context learning based chunking strategy and a KG-based index for accurate and efficient reference retrieval. Meanwhile, AIstorian orchestrates multi-agents to conduct on-the-fly hallucination detection and error-type-aware correction. Additionally, to teach LLMs a certain language style, we finetune LLMs based on a two-step training approach combining data augmentation-enhanced supervised fine-tuning with stylistic preference optimization. Extensive experiments on a real-life historical Jinshi dataset demonstrate that AIstorian achieves a 3.8x improvement in factual accuracy and a 47.6% reduction in hallucination rate compared to existing baselines. The data and code are available at: https://github.com/ZJU-DAILY/AIstorian.
Authors: Aissatou Diallo, Antonis Bikakis, Luke Dickens, Anthony Hunter, Rob Miller
Abstract: An interesting class of commonsense reasoning problems arises when people are faced with natural disasters. To investigate this topic, we present \textsf{RESPONSE}, a human-curated dataset containing 1789 annotated instances featuring 6037 sets of questions designed to assess LLMs' commonsense reasoning in disaster situations across different time frames. The dataset includes problem descriptions, missing resources, time-sensitive solutions, and their justifications, with a subset validated by environmental engineers. Through both automatic metrics and human evaluation, we compare LLM-generated recommendations against human responses. Our findings show that even state-of-the-art models like GPT-4 achieve only 37\% human-evaluated correctness for immediate response actions, highlighting significant room for improvement in LLMs' ability for commonsense reasoning in crises.
Authors: Panggih Kusuma Ningrum, Philipp Mayr, Nina Smirnova, Iana Atanassova
Abstract: UnScientify, a system designed to detect scientific uncertainty in scholarly full text. The system utilizes a weakly supervised technique to identify verbally expressed uncertainty in scientific texts and their authorial references. The core methodology of UnScientify is based on a multi-faceted pipeline that integrates span pattern matching, complex sentence analysis and author reference checking. This approach streamlines the labeling and annotation processes essential for identifying scientific uncertainty, covering a variety of uncertainty expression types to support diverse applications including information retrieval, text mining and scientific document processing. The evaluation results highlight the trade-offs between modern large language models (LLMs) and the UnScientify system. UnScientify, which employs more traditional techniques, achieved superior performance in the scientific uncertainty detection task, attaining an accuracy score of 0.808. This finding underscores the continued relevance and efficiency of UnScientify's simple rule-based and pattern matching strategy for this specific application. The results demonstrate that in scenarios where resource efficiency, interpretability, and domain-specific adaptability are critical, traditional methods can still offer significant advantages.
Authors: Annika Tjuka, Robert Forkel, Christoph Rzymski, Johann-Mattis List
Abstract: Lexical resources are crucial for cross-linguistic analysis and can provide new insights into computational models for natural language learning. Here, we present an advanced database for comparative studies of words with multiple meanings, a phenomenon known as colexification. The new version includes improvements in the handling, selection and presentation of the data. We compare the new database with previous versions and find that our improvements provide a more balanced sample covering more language families worldwide, with an enhanced data quality, given that all word forms are provided in phonetic transcription. We conclude that the new Database of Cross-Linguistic Colexifications has the potential to inspire exciting new studies that link cross-linguistic data to open questions in linguistic typology, historical linguistics, psycholinguistics, and computational linguistics.
Authors: Yuxiang Zhou, Hainiu Xu, Desmond C. Ong, Petr Slovak, Yulan He
Abstract: As the utilization of language models in interdisciplinary, human-centered studies grow, the expectation of model capabilities continues to evolve. Beyond excelling at conventional tasks, models are recently expected to perform well on user-centric measurements involving confidence and human (dis)agreement -- factors that reflect subjective preferences. While modeling of subjectivity plays an essential role in cognitive science and has been extensively studied, it remains under-explored within the NLP community. In light of this gap, we explore how language models can harness subjectivity by conducting comprehensive experiments and analysis across various scenarios using both fine-tuned models and prompt-based large language models (LLMs). Our quantitative and qualitative experimental results indicate that existing post-hoc calibration approaches often fail to produce satisfactory results. However, our findings reveal that personality traits and demographical information are critical for measuring subjectivity. Furthermore, our in-depth analysis offers valuable insights for future research and development in the interdisciplinary studies of NLP and cognitive science.
Authors: David Gu, Peter Belcak, Roger Wattenhofer
Abstract: We challenge the prevailing assumption that LLMs must rely fully on sub-word tokens for high-quality text generation. To this end, we propose the "Generative Pretrained Thoughtformer" (GPTHF), a hierarchical transformer language model capable of text generation by compressing text into sentence embeddings and employing a sentence attention mechanism. GPTHF retains GPT's architecture, modifying only token interactions via dynamic sparse attention masks. Our experiments show that GPTHF achieves an up to an order of magnitude improvement in FLOPs efficiency and a threefold increase in runtime speed compared to equally-sized GPT models in the low-size regime. This is achieved through a unique generation method that caches and reuses sentence embeddings, allowing significant portions of the input to bypass large parts of the network.
Authors: Jonas Belouadi, Eddy Ilg, Margret Keuper, Hideki Tanaka, Masao Utiyama, Raj Dabre, Steffen Eger, Simone Paolo Ponzetto
Abstract: With the rise of generative AI, synthesizing figures from text captions becomes a compelling application. However, achieving high geometric precision and editability requires representing figures as graphics programs in languages like TikZ, and aligned training data (i.e., graphics programs with captions) remains scarce. Meanwhile, large amounts of unaligned graphics programs and captioned raster images are more readily available. We reconcile these disparate data sources by presenting TikZero, which decouples graphics program generation from text understanding by using image representations as an intermediary bridge. It enables independent training on graphics programs and captioned images and allows for zero-shot text-guided graphics program synthesis during inference. We show that our method substantially outperforms baselines that can only operate with caption-aligned graphics programs. Furthermore, when leveraging caption-aligned graphics programs as a complementary training signal, TikZero matches or exceeds the performance of much larger models, including commercial systems like GPT-4o. Our code, datasets, and select models are publicly available.
Authors: Bastian Bunzeck, Daniel Duran, Sina Zarrie{\ss}
Abstract: We analyze the influence of utterance-level construction distributions in German child-directed speech on the resulting formal linguistic competence and the underlying learning trajectories for small language models trained on a novel collection of developmentally plausible language data for German. We find that trajectories are surprisingly robust for markedly different distributions of constructions in the training data, which have little effect on final accuracies and almost no effect on global learning trajectories. While syntax learning benefits from more complex utterances, lexical learning culminates in better scores with more fragmentary data. We argue that LMs trained on developmentally plausible data can contribute to debates on how rich or impoverished linguistic stimuli actually are.
Authors: Liang Cheng, Tianyi Li, Zhaowei Wang, Tianyang Liu, Mark Steedman
Abstract: LLMs are often claimed to be capable of Natural Language Inference (NLI), which is widely regarded as a cornerstone of more complex forms of reasoning. However, recent works show that LLMs still suffer from hallucinations in NLI due to attestation bias, where LLMs overly rely on propositional memory to build shortcuts. To solve the issue, we design an unsupervised framework to construct counterfactual reasoning data and fine-tune LLMs to reduce attestation bias. To measure bias reduction, we build bias-adversarial variants of NLI datasets with randomly replaced predicates in premises while keeping hypotheses unchanged. Extensive evaluations show that our framework can significantly reduce hallucinations from attestation bias. Then, we further evaluate LLMs fine-tuned with our framework on original NLI datasets and their bias-neutralized versions, where original entities are replaced with randomly sampled ones. Extensive results show that our framework consistently improves inferential performance on both original and bias-neutralized NLI datasets.
Authors: Tamar I. Regev, Chiebuka Ohams, Shaylee Xie, Lukas Wolf, Evelina Fedorenko, Alex Warstadt, Ethan Wilcox, Tiago Pimentel
Abstract: In spoken language, speakers transmit information not only using words, but also via a rich array of non-verbal signals, which include prosody -- the auditory features of speech. However, previous studies have shown that prosodic features exhibit significant redundancy with both past and future words. Here, we examine the time scale of this relationship: How many words in the past (or future) contribute to predicting prosody? We find that this scale differs for past and future words. Prosody's redundancy with past words extends across approximately 3-8 words, whereas redundancy with future words is limited to just 1-2 words. These findings indicate that the prosody-future relationship reflects local word dependencies or short-scale processes such as next word prediction, while the prosody-past relationship unfolds over a longer time scale. The latter suggests that prosody serves to emphasize earlier information that may be challenging for listeners to process given limited cognitive resources in real-time communication. Our results highlight the role of prosody in shaping efficient communication.
Authors: David Rozado
Abstract: Political biases in Large Language Model (LLM)-based artificial intelligence (AI) systems, such as OpenAI's ChatGPT or Google's Gemini, have been previously reported. While several prior studies have attempted to quantify these biases using political orientation tests, such approaches are limited by potential tests' calibration biases and constrained response formats that do not reflect real-world human-AI interactions. This study employs a multi-method approach to assess political bias in leading AI systems, integrating four complementary methodologies: (1) linguistic comparison of AI-generated text with the language used by Republican and Democratic U.S. Congress members, (2) analysis of political viewpoints embedded in AI-generated policy recommendations, (3) sentiment analysis of AI-generated text toward politically affiliated public figures, and (4) standardized political orientation testing. Results indicate a consistent left-leaning bias across most contemporary AI systems, with arguably varying degrees of intensity. However, this bias is not an inherent feature of LLMs; prior research demonstrates that fine-tuning with politically skewed data can realign these models across the ideological spectrum. The presence of systematic political bias in AI systems poses risks, including reduced viewpoint diversity, increased societal polarization, and the potential for public mistrust in AI technologies. To mitigate these risks, AI systems should be designed to prioritize factual accuracy while maintaining neutrality on most lawful normative issues. Furthermore, independent monitoring platforms are necessary to ensure transparency, accountability, and responsible AI development.
Authors: Tabia Tanzin Prama, Jannatul Ferdaws Amrin, Md. Mushfique Anwar, Iqbal H. Sarker
Abstract: The rise of social media has significantly increased the prevalence of cyberbullying (CB), posing serious risks to both mental and physical well-being. Effective detection systems are essential for mitigating its impact. While several machine learning (ML) models have been developed, few incorporate victims' psychological, demographic, and behavioral factors alongside bullying comments to assess severity. In this study, we propose an AI model intregrating user-specific attributes, including psychological factors (self-esteem, anxiety, depression), online behavior (internet usage, disciplinary history), and demographic attributes (race, gender, ethnicity), along with social media comments. Additionally, we introduce a re-labeling technique that categorizes social media comments into three severity levels: Not Bullying, Mild Bullying, and Severe Bullying, considering user-specific factors.Our LSTM model is trained using 146 features, incorporating emotional, topical, and word2vec representations of social media comments as well as user-level attributes and it outperforms existing baseline models, achieving the highest accuracy of 98\% and an F1-score of 0.97. To identify key factors influencing the severity of cyberbullying, we employ explainable AI techniques (SHAP and LIME) to interpret the model's decision-making process. Our findings reveal that, beyond hate comments, victims belonging to specific racial and gender groups are more frequently targeted and exhibit higher incidences of depression, disciplinary issues, and low self-esteem. Additionally, individuals with a prior history of bullying are at a greater risk of becoming victims of cyberbullying.
Authors: Nitesh Patnaik, Navdeep Nayak, Himani Bansal Agrawal, Moinak Chinmoy Khamaru, Gourav Bal, Saishree Smaranika Panda, Rishi Raj, Vishal Meena, Kartheek Vadlamani
Abstract: The emergence of small vision-language models (sVLMs) marks a critical advancement in multimodal AI, enabling efficient processing of visual and textual data in resource-constrained environments. This survey offers a comprehensive exploration of sVLM development, presenting a taxonomy of architectures - transformer-based, mamba-based, and hybrid - that highlight innovations in compact design and computational efficiency. Techniques such as knowledge distillation, lightweight attention mechanisms, and modality pre-fusion are discussed as enablers of high performance with reduced resource requirements. Through an in-depth analysis of models like TinyGPT-V, MiniGPT-4, and VL-Mamba, we identify trade-offs between accuracy, efficiency, and scalability. Persistent challenges, including data biases and generalization to complex tasks, are critically examined, with proposed pathways for addressing them. By consolidating advancements in sVLMs, this work underscores their transformative potential for accessible AI, setting a foundation for future research into efficient multimodal systems.
Authors: Mooho Song, Jay-Yoon Lee
Abstract: Examining logical inconsistencies among multiple statements (such as collections of sentences or question-answer pairs) is a crucial challenge in machine learning, particularly for ensuring the safety and reliability of models. Traditional methods that rely on pairwise comparisons often fail to capture inconsistencies that only emerge when more than two statements are evaluated collectively. To address this gap, we introduce the task of set-consistency verification, an extension of natural language inference (NLI) that assesses the logical coherence of entire sets rather than isolated pairs. Building on this task, we present the Set-Consistency Energy Network (SC-Energy), a novel model that employs a contrastive loss framework to learn the compatibility among a collection of statements. Our approach not only efficiently verifies inconsistencies and pinpoints the specific statements responsible for logical contradictions, but also significantly outperforms existing methods including prompting-based LLM models. Furthermore, we release two new datasets: Set-LConVQA and Set-SNLI for set-consistency verification task.
Authors: Yudong Liu, Jingwei Sun, Yueqian Lin, Jingyang Zhang, Ming Yin, Qinsi Wang, Jianyi Zhang, Hai Li, Yiran Chen
Abstract: Vision language models (VLMs) demonstrate strong capabilities in jointly processing visual and textual data. However, they often incur substantial computational overhead due to redundant visual information, particularly in long-form video scenarios. Existing approaches predominantly focus on either vision token pruning, which may overlook spatio-temporal dependencies, or keyframe selection, which identifies informative frames but discards others, thus disrupting contextual continuity. In this work, we propose KVTP (Keyframe-oriented Vision Token Pruning), a novel framework that overcomes the drawbacks of token pruning and keyframe selection. By adaptively assigning pruning rates based on frame relevance to the query, KVTP effectively retains essential contextual information while significantly reducing redundant computation. To thoroughly evaluate the long-form video understanding capacities of VLMs, we curated and reorganized subsets from VideoMME, EgoSchema, and NextQA into a unified benchmark named SparseKV-QA that highlights real-world scenarios with sparse but crucial events. Our experiments with VLMs of various scales show that KVTP can reduce token usage by 80% without compromising spatiotemporal and contextual consistency, significantly cutting computation while maintaining the performance. These results demonstrate our approach's effectiveness in efficient long-video processing, facilitating more scalable VLM deployment.
Authors: Kai Zhang, Jianwei Yang, Jeevana Priya Inala, Chandan Singh, Jianfeng Gao, Yu Su, Chenglong Wang
Abstract: Despite the promising results of large multimodal models (LMMs) in complex vision-language tasks that require knowledge, reasoning, and perception abilities together, we surprisingly found that these models struggle with simple tasks on infographics that require perception only. As existing benchmarks primarily focus on end tasks that require various abilities, they provide limited, fine-grained insights into the limitations of the models' perception abilities. To address this gap, we leverage the theory of graphical perception, an approach used to study how humans decode visual information encoded on charts and graphs, to develop an evaluation framework for analyzing gaps in LMMs' perception abilities in charts. With automated task generation and response evaluation designs, our framework enables comprehensive and controlled testing of LMMs' graphical perception across diverse chart types, visual elements, and task types. We apply our framework to evaluate and diagnose the perception capabilities of state-of-the-art LMMs at three granularity levels (chart, visual element, and pixel). Our findings underscore several critical limitations of current state-of-the-art LMMs, including GPT-4o: their inability to (1) generalize across chart types, (2) understand fundamental visual elements, and (3) cross reference values within a chart. These insights provide guidance for future improvements in perception abilities of LMMs. The evaluation framework and labeled data are publicly available at https://github.com/microsoft/lmm-graphical-perception.
URLs: https://github.com/microsoft/lmm-graphical-perception.
Authors: Paul Quinlan, Qingguo Li, Xiaodan Zhu
Abstract: Time-series analysis is critical for a wide range of fields such as healthcare, finance, transportation, and energy, among many others. The practical applications often involve analyzing time-series data alongside contextual information in the form of natural language to support informed decisions. However, current time-series models are limited in their ability to perform reasoning that involves both time-series and their textual content. In this work, we address this gap by introducing \textit{Chat-TS}, a large language model (LLM) based framework, designed to support reasoning over time series and textual data. Unlike traditional models, Chat-TS integrates time-series tokens into LLMs' vocabulary, enhancing its reasoning ability over both modalities without compromising the core natural language capabilities, enabling practical analysis and reasoning across modalities. To support learning and evaluation in this setup, we contribute new datasets: the \textit{TS Instruct Training Dataset} which pairs diverse time-series data with relevant text instructions and responses for instruction tuning, the \textit{TS Instruct Question and Answer (QA) Gold Dataset} which provides multiple-choice questions designed to evaluate multimodal reasoning, and a \textit{TS Instruct Quantitative Probing Set} which contains a small subset of the TS Instruct QA tasks alongside math and decision-making questions for LLM evaluation. We designed a training strategy to preserve the inherent reasoning capabilities of LLMs while augmenting them for time-series reasoning. Experiments show that Chat-TS achieves state-of-the-art performance in multi-modal reasoning tasks by maintaining strong natural language proficiency while improving time-series reasoning. ~\footnote{To ensure replicability and facilitate future research, all models, datasets, and code will be available at [\texttt{Github-URL}].}
Authors: Samuel Marks, Johannes Treutlein, Trenton Bricken, Jack Lindsey, Jonathan Marcus, Siddharth Mishra-Sharma, Daniel Ziegler, Emmanuel Ameisen, Joshua Batson, Tim Belonax, Samuel R. Bowman, Shan Carter, Brian Chen, Hoagy Cunningham, Carson Denison, Florian Dietz, Satvik Golechha, Akbir Khan, Jan Kirchner, Jan Leike, Austin Meek, Kei Nishimura-Gasparian, Euan Ong, Christopher Olah, Adam Pearce, Fabien Roger, Jeanne Salle, Andy Shih, Meg Tong, Drake Thomas, Kelley Rivoire, Adam Jermyn, Monte MacDiarmid, Tom Henighan, Evan Hubinger
Abstract: We study the feasibility of conducting alignment audits: investigations into whether models have undesired objectives. As a testbed, we train a language model with a hidden objective. Our training pipeline first teaches the model about exploitable errors in RLHF reward models (RMs), then trains the model to exploit some of these errors. We verify via out-of-distribution evaluations that the model generalizes to exhibit whatever behaviors it believes RMs rate highly, including ones not reinforced during training. We leverage this model to study alignment audits in two ways. First, we conduct a blind auditing game where four teams, unaware of the model's hidden objective or training, investigate it for concerning behaviors and their causes. Three teams successfully uncovered the model's hidden objective using techniques including interpretability with sparse autoencoders (SAEs), behavioral attacks, and training data analysis. Second, we conduct an unblinded follow-up study of eight techniques for auditing the model, analyzing their strengths and limitations. Overall, our work provides a concrete example of using alignment audits to discover a model's hidden objective and proposes a methodology for practicing and validating progress in alignment auditing.
Authors: Camilo Chac\'on Sartori, Christian Blum
Abstract: Large Language Models (LLMs) have shown notable potential in code generation for optimization algorithms, unlocking exciting new opportunities. This paper examines how LLMs, rather than creating algorithms from scratch, can improve existing ones without the need for specialized expertise. To explore this potential, we selected 10 baseline optimization algorithms from various domains (metaheuristics, reinforcement learning, deterministic, and exact methods) to solve the classic Travelling Salesman Problem. The results show that our simple methodology often results in LLM-generated algorithm variants that improve over the baseline algorithms in terms of solution quality, reduction in computational time, and simplification of code complexity, all without requiring specialized optimization knowledge or advanced algorithmic implementation skills.
Authors: Xueyang Zhou, Guiyao Tie, Guowen Zhang, Weidong Wang, Zhigang Zuo, Di Wu, Duanfeng Chu, Pan Zhou, Lichao Sun, Neil Zhenqiang Gong
Abstract: The rise of Large Reasoning Models (LRMs) signifies a paradigm shift toward advanced computational reasoning. Yet, this progress disrupts traditional agent frameworks, traditionally anchored by execution-oriented Large Language Models (LLMs). To explore this transformation, we propose the LaRMA framework, encompassing nine tasks across Tool Usage, Plan Design, and Problem Solving, assessed with three top LLMs (e.g., Claude3.5-sonnet) and five leading LRMs (e.g., DeepSeek-R1). Our findings address four research questions: LRMs surpass LLMs in reasoning-intensive tasks like Plan Design, leveraging iterative reflection for superior outcomes; LLMs excel in execution-driven tasks such as Tool Usage, prioritizing efficiency; hybrid LLM-LRM configurations, pairing LLMs as actors with LRMs as reflectors, optimize agent performance by blending execution speed with reasoning depth; and LRMs' enhanced reasoning incurs higher computational costs, prolonged processing, and behavioral challenges, including overthinking and fact-ignoring tendencies. This study fosters deeper inquiry into LRMs' balance of deep thinking and overthinking, laying a critical foundation for future agent design advancements.
Authors: Yifang Chen, Xiaoyu Li, Yingyu Liang, Zhenmei Shi, Zhao Song, Yu Tian
Abstract: The key-value (KV) cache in autoregressive transformers presents a significant bottleneck during inference, which restricts the context length capabilities of large language models (LLMs). While previous work analyzes the fundamental space complexity barriers in standard attention mechanism [Haris and Onak, 2025], our work generalizes the space complexity barriers result to tensor attention version. Our theoretical contributions rely on a novel reduction from communication complexity and deduce the memory lower bound for tensor-structured attention mechanisms when $d = \Omega(\log n)$. In the low dimensional regime where $d = o(\log n)$, we analyze the theoretical bounds of the space complexity as well. Overall, our work provides a theoretical foundation for us to understand the compression-expressivity tradeoff in tensor attention mechanisms and offers more perspectives in developing more memory-efficient transformer architectures.
Authors: Zhuoyuan Mao, Mengjie Zhao, Qiyu Wu, Zhi Zhong, Wei-Hsiang Liao, Hiromi Wakaki, Yuki Mitsufuji
Abstract: Music-to-music-video generation is a challenging task due to the intrinsic differences between the music and video modalities. The advent of powerful text-to-video diffusion models has opened a promising pathway for music-video (MV) generation by first addressing the music-to-MV description task and subsequently leveraging these models for video generation. In this study, we focus on the MV description generation task and propose a comprehensive pipeline encompassing training data construction and multimodal model fine-tuning. We fine-tune existing pre-trained multimodal models on our newly constructed music-to-MV description dataset based on the Music4All dataset, which integrates both musical and visual information. Our experimental results demonstrate that music representations can be effectively mapped to textual domains, enabling the generation of meaningful MV description directly from music inputs. We also identify key components in the dataset construction pipeline that critically impact the quality of MV description and highlight specific musical attributes that warrant greater focus for improved MV description generation.
Authors: Gang Li, Jizhong Liu, Heinrich Dinkel, Yadong Niu, Junbo Zhang, Jian Luan
Abstract: Recently, reinforcement learning (RL) has been shown to greatly enhance the reasoning capabilities of large language models (LLMs), and RL-based approaches have been progressively applied to visual multimodal tasks. However, the audio modality has largely been overlooked in these developments. Thus, we conduct a series of RL explorations in audio understanding and reasoning, specifically focusing on the audio question answering (AQA) task. We leverage the group relative policy optimization (GRPO) algorithm to Qwen2-Audio-7B-Instruct, and our experiments demonstrated state-of-the-art performance on the MMAU Test-mini benchmark, achieving an accuracy rate of 64.5%. The main findings in this technical report are as follows: 1) The GRPO algorithm can be effectively applied to large audio language models (LALMs), even when the model has only 8.2B parameters; 2) With only 38k post-training samples, RL significantly outperforms supervised fine-tuning (SFT), indicating that RL-based approaches can be effective without large datasets; 3) The explicit reasoning process has not shown significant benefits for AQA tasks, and how to efficiently utilize deep thinking remains an open question for further research; 4) LALMs still lag far behind humans auditory-language reasoning, suggesting that the RL-based approaches warrant further exploration. Our project is available at https://github.com/xiaomi/r1-aqa and https://huggingface.co/mispeech/r1-aqa.
URLs: https://github.com/xiaomi/r1-aqa, https://huggingface.co/mispeech/r1-aqa.
Authors: Xingtai Lv, Youbang Sun, Kaiyan Zhang, Shang Qu, Xuekai Zhu, Yuchen Fan, Yi Wu, Ermo Hua, Xinwei Long, Ning Ding, Bowen Zhou
Abstract: State Space Models (SSMs) have emerged as a promising alternative to the popular transformer-based models and have been increasingly gaining attention. Compared to transformers, SSMs excel at tasks with sequential data or longer contexts, demonstrating comparable performances with significant efficiency gains. In this survey, we provide a coherent and systematic overview for SSMs, including their theoretical motivations, mathematical formulations, comparison with existing model classes, and various applications. We divide the SSM series into three main sections, providing a detailed introduction to the original SSM, the structured SSM represented by S4, and the selective SSM typified by Mamba. We put an emphasis on technicality, and highlight the various key techniques introduced to address the effectiveness and efficiency of SSMs. We hope this manuscript serves as an introduction for researchers to explore the theoretical foundations of SSMs.
Authors: Ke Wang, Lei He, Kun Liu, Yan Deng, Wenning Wei, Sheng Zhao
Abstract: Large Multimodal Models (LMMs) have demonstrated exceptional performance across a wide range of domains. This paper explores their potential in pronunciation assessment tasks, with a particular focus on evaluating the capabilities of the Generative Pre-trained Transformer (GPT) model, specifically GPT-4o. Our study investigates its ability to process speech and audio for pronunciation assessment across multiple levels of granularity and dimensions, with an emphasis on feedback generation and scoring. For our experiments, we use the publicly available Speechocean762 dataset. The evaluation focuses on two key aspects: multi-level scoring and the practicality of the generated feedback. Scoring results are compared against the manual scores provided in the Speechocean762 dataset, while feedback quality is assessed using Large Language Models (LLMs). The findings highlight the effectiveness of integrating LMMs with traditional methods for pronunciation assessment, offering insights into the model's strengths and identifying areas for further improvement.
Authors: Ahmed Frikha, Muhammad Reza Ar Razi, Krishna Kanth Nakka, Ricardo Mendes, Xue Jiang, Xuebing Zhou
Abstract: Large Language Models (LLMs) have demonstrated remarkable capabilities in natural language processing but also pose significant privacy risks by memorizing and leaking Personally Identifiable Information (PII). Existing mitigation strategies, such as differential privacy and neuron-level interventions, often degrade model utility or fail to effectively prevent leakage. To address this challenge, we introduce PrivacyScalpel, a novel privacy-preserving framework that leverages LLM interpretability techniques to identify and mitigate PII leakage while maintaining performance. PrivacyScalpel comprises three key steps: (1) Feature Probing, which identifies layers in the model that encode PII-rich representations, (2) Sparse Autoencoding, where a k-Sparse Autoencoder (k-SAE) disentangles and isolates privacy-sensitive features, and (3) Feature-Level Interventions, which employ targeted ablation and vector steering to suppress PII leakage. Our empirical evaluation on Gemma2-2b and Llama2-7b, fine-tuned on the Enron dataset, shows that PrivacyScalpel significantly reduces email leakage from 5.15\% to as low as 0.0\%, while maintaining over 99.4\% of the original model's utility. Notably, our method outperforms neuron-level interventions in privacy-utility trade-offs, demonstrating that acting on sparse, monosemantic features is more effective than manipulating polysemantic neurons. Beyond improving LLM privacy, our approach offers insights into the mechanisms underlying PII memorization, contributing to the broader field of model interpretability and secure AI deployment.
Authors: Rabimba Karanjai, Sam Blackshear, Lei Xu, Weidong Shi
Abstract: This paper introduces UniTranslator, a visionary framework that re-imagines code translation as a collaborative endeavor among multiple, compact LLMs. By orchestrating the interaction of specialized agents, each focused on different aspects of the translation process and grounded in a deep understanding of programming concepts, UniTranslator achieves a level of accuracy and efficiency that rivals larger, monolithic models. Our preliminary evaluation demonstrates the potential of UniTranslator to overcome the limitations of existing approaches and unlock the power of smaller LLMs for complex code translation tasks. We explore the effectiveness of this dynamic multi-agent paradigm in handling diverse language pairs, including low-resource languages, and in mitigating common issues such as code artifacts and hallucinations through the use of Natural Language Inference (NLI) grounding and iterative feedback mechanisms
Authors: Vojtech Cahlik, Rodrigo Alves, Pavel Kordik
Abstract: We propose a large language model explainability technique for obtaining faithful natural language explanations by grounding the explanations in a reasoning process. When converted to a sequence of tokens, the outputs of the reasoning process can become part of the model context and later be decoded to natural language as the model produces either the final answer or the explanation. To improve the faithfulness of the explanations, we propose to use a joint predict-explain approach, in which the answers and explanations are inferred directly from the reasoning sequence, without the explanations being dependent on the answers and vice versa. We demonstrate the plausibility of the proposed technique by achieving a high alignment between answers and explanations in several problem domains, observing that language models often simply copy the partial decisions from the reasoning sequence into the final answers or explanations. Furthermore, we show that the proposed use of reasoning can also improve the quality of the answers.
Authors: Haoyang Huang, Guoqing Ma, Nan Duan, Xing Chen, Changyi Wan, Ranchen Ming, Tianyu Wang, Bo Wang, Zhiying Lu, Aojie Li, Xianfang Zeng, Xinhao Zhang, Gang Yu, Yuhe Yin, Qiling Wu, Wen Sun, Kang An, Xin Han, Deshan Sun, Wei Ji, Bizhu Huang, Brian Li, Chenfei Wu, Guanzhe Huang, Huixin Xiong, Jiaxin He, Jianchang Wu, Jianlong Yuan, Jie Wu, Jiashuai Liu, Junjing Guo, Kaijun Tan, Liangyu Chen, Qiaohui Chen, Ran Sun, Shanshan Yuan, Shengming Yin, Sitong Liu, Wei Chen, Yaqi Dai, Yuchu Luo, Zheng Ge, Zhisheng Guan, Xiaoniu Song, Yu Zhou, Binxing Jiao, Jiansheng Chen, Jing Li, Shuchang Zhou, Xiangyu Zhang, Yi Xiu, Yibo Zhu, Heung-Yeung Shum, Daxin Jiang
Abstract: We present Step-Video-TI2V, a state-of-the-art text-driven image-to-video generation model with 30B parameters, capable of generating videos up to 102 frames based on both text and image inputs. We build Step-Video-TI2V-Eval as a new benchmark for the text-driven image-to-video task and compare Step-Video-TI2V with open-source and commercial TI2V engines using this dataset. Experimental results demonstrate the state-of-the-art performance of Step-Video-TI2V in the image-to-video generation task. Both Step-Video-TI2V and Step-Video-TI2V-Eval are available at https://github.com/stepfun-ai/Step-Video-TI2V.
Authors: Jiyeong Kim, Stephen P. Ma, Michael L. Chen, Isaac R. Galatzer-Levy, John Torous, Peter J. van Roessel, Christopher Sharp, Michael A. Pfeffer, Carolyn I. Rodriguez, Eleni Linos, Jonathan H. Chen
Abstract: Patients with diabetes are at increased risk of comorbid depression or anxiety, complicating their management. This study evaluated the performance of large language models (LLMs) in detecting these symptoms from secure patient messages. We applied multiple approaches, including engineered prompts, systemic persona, temperature adjustments, and zero-shot and few-shot learning, to identify the best-performing model and enhance performance. Three out of five LLMs demonstrated excellent performance (over 90% of F-1 and accuracy), with Llama 3.1 405B achieving 93% in both F-1 and accuracy using a zero-shot approach. While LLMs showed promise in binary classification and handling complex metrics like Patient Health Questionnaire-4, inconsistencies in challenging cases warrant further real-life assessment. The findings highlight the potential of LLMs to assist in timely screening and referrals, providing valuable empirical knowledge for real-world triage systems that could improve mental health care for patients with chronic diseases.
Authors: Balaji Rama, Kai Mei, Yongfeng Zhang
Abstract: Autonomous LLM-based agents have emerged as a powerful paradigm for complex task execution, yet the field lacks standardized tools for development, deployment, distribution and discovery of agents. We present Cerebrum, an Agent SDK for AIOS that addresses this gap through three key components: (1) a comprehensive SDK featuring a modular four-layer architecture for agent development, encompassing LLM, memory, storage, and tool management; (2) a community-driven Agent Hub for sharing and discovering agents, complete with version control and dependency management; (3) an interactive web interface for testing and evaluating agents. The platform's effectiveness is demonstrated through implementations of various agent architectures, including Chain of Thought (CoT), ReAct, and tool-use agents. Cerebrum advances the field by providing a unified framework that standardizes agent development while maintaining flexibility for researchers and developers to innovate and distribute their agents. The live website is at https://app.aios.foundation, the code is at https://github.com/agiresearch/Cerebrum, and video is at https://app.aios.foundation/video-demo.
URLs: https://app.aios.foundation,, https://github.com/agiresearch/Cerebrum,, https://app.aios.foundation/video-demo.
Authors: Diego Gosmar, Deborah A. Dahl, Dario Gosmar
Abstract: Prompt injection constitutes a significant challenge for generative AI systems by inducing unintended outputs. We introduce a multi-agent NLP framework specifically designed to address prompt injection vulnerabilities through layered detection and enforcement mechanisms. The framework orchestrates specialized agents for generating responses, sanitizing outputs, and enforcing policy compliance. Evaluation on 500 engineered injection prompts demonstrates a marked reduction in injection success and policy breaches. Novel metrics, including Injection Success Rate (ISR), Policy Override Frequency (POF), Prompt Sanitization Rate (PSR), and Compliance Consistency Score (CCS), are proposed to derive a composite Total Injection Vulnerability Score (TIVS). The system utilizes the OVON (Open Voice Network) framework for inter-agent communication via structured JSON messages, extending a previously established multi-agent architecture from hallucination mitigation to address the unique challenges of prompt injection.
Authors: Hao Cheng, Erjia Xiao, Yichi Wang, Kaidi Xu, Mengshu Sun, Jindong Gu, Renjing Xu
Abstract: Current Cross-Modality Generation Models (GMs) demonstrate remarkable capabilities in various generative tasks. Given the ubiquity and information richness of vision modality inputs in real-world scenarios, Cross-vision, encompassing Vision-Language Perception (VLP) and Image-to-Image (I2I), tasks have attracted significant attention. Large Vision Language Models (LVLMs) and I2I GMs are employed to handle VLP and I2I tasks, respectively. Previous research indicates that printing typographic words into input images significantly induces LVLMs and I2I GMs to generate disruptive outputs semantically related to those words. Additionally, visual prompts, as a more sophisticated form of typography, are also revealed to pose security risks to various applications of VLP tasks when injected into images. In this paper, we comprehensively investigate the performance impact induced by Typographic Visual Prompt Injection (TVPI) in various LVLMs and I2I GMs. To better observe performance modifications and characteristics of this threat, we also introduce the TVPI Dataset. Through extensive explorations, we deepen the understanding of the underlying causes of the TVPI threat in various GMs and offer valuable insights into its potential origins.
Authors: Zhiliang Chen, Xinyuan Niu, Chuan-Sheng Foo, Bryan Kian Hsiang Low
Abstract: Large language models (LLMs) are used in chatbots or AI assistants to hold conversations with a human user. In such applications, the quality (e.g., user engagement, safety) of a conversation is important and can only be exactly known at the end of the conversation. To maximize its expected quality, conversation planning reasons about the stochastic transitions within a conversation to select the optimal LLM response at each turn. Existing simulation-based conversation planning algorithms typically select the optimal response by simulating future conversations with a large number of LLM queries at every turn. However, this process is extremely time-consuming and hence impractical for real-time conversations. This paper presents a novel approach called Semantic space COnversation Planning with improved Efficiency (SCOPE) that exploits the dense semantic representation of conversations to perform conversation planning efficiently. In particular, SCOPE models the stochastic transitions in conversation semantics and their associated rewards to plan entirely within the semantic space. This allows us to select the optimal LLM response at every conversation turn without needing additional LLM queries for simulation. As a result, SCOPE can perform conversation planning 70 times faster than conventional simulation-based planning algorithms when applied to a wide variety of conversation starters and two reward functions seen in the real world, yet achieving a higher reward within a practical planning budget. Our code can be found at: https://github.com/chenzhiliang94/convo-plan-SCOPE.
Authors: Pedro Reviriego, Javier Conde, Elena Merino-G\'omez, Gonzalo Mart\'inez, Jos\'e Alberto Hern\'andez
Abstract: The introduction of Artificial Intelligence (AI) generative language models such as GPT (Generative Pre-trained Transformer) and tools such as ChatGPT has triggered a revolution that can transform how text is generated. This has many implications, for example, as AI-generated text becomes a significant fraction of the text, would this have an effect on the language capabilities of readers and also on the training of newer AI tools? Would it affect the evolution of languages? Focusing on one specific aspect of the language: words; will the use of tools such as ChatGPT increase or reduce the vocabulary used or the lexical richness? This has implications for words, as those not included in AI-generated content will tend to be less and less popular and may eventually be lost. In this work, we perform an initial comparison of the vocabulary and lexical richness of ChatGPT and humans when performing the same tasks. In more detail, two datasets containing the answers to different types of questions answered by ChatGPT and humans, and a third dataset in which ChatGPT paraphrases sentences and questions are used. The analysis shows that ChatGPT tends to use fewer distinct words and lower lexical richness than humans. These results are very preliminary and additional datasets and ChatGPT configurations have to be evaluated to extract more general conclusions. Therefore, further research is needed to understand how the use of ChatGPT and more broadly generative AI tools will affect the vocabulary and lexical richness in different types of text and languages.
Authors: Junjie Liu, Shaotian Yan, Chen Shen, Zhengdong Xiao, Liang Xie, Wenxiao Wang, Jieping Ye
Abstract: Exploiting large language models (LLMs) to tackle reasoning has garnered growing attention. It still remains highly challenging to achieve satisfactory results in complex logical problems, characterized by plenty of premises within the context and requiring multi-hop reasoning. In particular, the reasoning capabilities of LLMs are brittle to disorder and distractibility. In this work, we first examine the mechanism from the perspective of information flow and reveal that LLMs confront difficulties akin to human-like cognitive biases when dealing with disordered and irrelevant content in reasoning tasks. However, in contrast to LLMs, disordered and irrelevant content does not significantly decrease human performance, as humans have a propensity to distill the most relevant information and systematically organize their thoughts, aiding them in responding to questions.Stem from that, we further propose a novel reasoning approach named Concise and Organized Perception (COP). COP carefully analyzes the given statements to identify the most pertinent information while eliminating redundancy efficiently. It then prompts the LLMs in a more organized form that adapts to the model's inference process. By perceiving concise and organized context, the reasoning abilities of LLMs can be better elicited. Extensive experimental results on several popular logical benchmarks (ProofWriter, PrOntoQA, PrOntoQA-OOD, and FOLIO) and mathematical benchmark (DI-GSM) show that COP significantly outperforms previous state-of-the-art methods.
Authors: Haoxin Liu, Wenli Zhang, Jiaheng Xie, Buomsoo Kim, Zhu Zhang, Yidong Chai, Sudha Ram
Abstract: This study harnesses state-of-the-art AI technology for detecting mental disorders through user-generated textual content. Existing studies typically rely on fully supervised machine learning, which presents challenges such as the labor-intensive manual process of annotating extensive training data for each research problem and the need to design specialized deep learning architectures for each task. We propose a novel method to address these challenges by leveraging large language models and continuous multi-prompt engineering, which offers two key advantages: (1) developing personalized prompts that capture each user's unique characteristics and (2) integrating structured medical knowledge into prompts to provide context for disease detection and facilitate predictive modeling. We evaluate our method using three widely prevalent mental disorders as research cases. Our method significantly outperforms existing methods, including feature engineering, architecture engineering, and discrete prompt engineering. Meanwhile, our approach demonstrates success in few-shot learning, i.e., requiring only a minimal number of training examples. Moreover, our method can be generalized to other rare mental disorder detection tasks with few positive labels. In addition to its technical contributions, our method has the potential to enhance the well-being of individuals with mental disorders and offer a cost-effective, accessible alternative for stakeholders beyond traditional mental disorder screening methods.
Authors: Qizhou Chen, Taolin Zhang, Xiaofeng He, Dongyang Li, Chengyu Wang, Longtao Huang, Hui Xue
Abstract: Model editing aims to correct outdated or erroneous knowledge in large language models (LLMs) without the need for costly retraining. Lifelong model editing is the most challenging task that caters to the continuous editing requirements of LLMs. Prior works primarily focus on single or batch editing; nevertheless, these methods fall short in lifelong editing scenarios due to catastrophic knowledge forgetting and the degradation of model performance. Although retrieval-based methods alleviate these issues, they are impeded by slow and cumbersome processes of integrating the retrieved knowledge into the model. In this work, we introduce RECIPE, a RetriEval-augmented ContInuous Prompt lEarning method, to boost editing efficacy and inference efficiency in lifelong learning. RECIPE first converts knowledge statements into short and informative continuous prompts, prefixed to the LLM's input query embedding, to efficiently refine the response grounded on the knowledge. It further integrates the Knowledge Sentinel (KS) that acts as an intermediary to calculate a dynamic threshold, determining whether the retrieval repository contains relevant knowledge. Our retriever and prompt encoder are jointly trained to achieve editing properties, i.e., reliability, generality, and locality. In our experiments, RECIPE is assessed extensively across multiple LLMs and editing datasets, where it achieves superior editing performance. RECIPE also demonstrates its capability to maintain the overall performance of LLMs alongside showcasing fast editing and inference speed.
Authors: Chen Cecilia Liu, Iryna Gurevych, Anna Korhonen
Abstract: The surge of interest in "culture" in NLP has inspired much recent research, but a shared understanding of "culture" remains unclear, making it difficult to evaluate progress in this emerging area. Drawing on prior research in NLP and related fields, we propose a fine-grained taxonomy of elements in culture that can provide a systematic framework for analyzing and understanding research progress. Using the taxonomy, we survey existing resources and methods for culturally aware and adapted NLP, providing an overview of the state of the art and the research gaps that still need to be filled.
Authors: Philip Resnik
Abstract: This position paper's primary goal is to provoke thoughtful discussion about the relationship between bias and fundamental properties of large language models. I do this by seeking to convince the reader that harmful biases are an inevitable consequence arising from the design of any large language model as LLMs are currently formulated. To the extent that this is true, it suggests that the problem of harmful bias cannot be properly addressed without a serious reconsideration of AI driven by LLMs, going back to the foundational assumptions underlying their design.
Authors: Kun Zhao, Chenghao Xiao, Sixing Yan, William K. Cheung, Kai Ye, Noura Al Moubayed, Liang Zhan, Chenghua Lin
Abstract: Radiology Report Generation (RRG) has advanced considerably with the development of multimodal generative models. Despite the progress, the field still faces significant challenges in evaluation, as existing metrics lack robustness and fairness. We reveal that, RRG with high performance on existing lexical-based metrics (e.g. BLEU) might be more of a mirage - a model can get a high BLEU only by learning the template of reports. This has become a pressing issue for RRG due to the highly patternized nature of these reports. In addition, standard radiology reports are often highly technical. Helping patients understand these reports is crucial from a patient's perspective, yet this has been largely overlooked in previous work. In this work, we un-intuitively approach these problems by proposing the Layman's RRG framework that can systematically improve RRG with day-to-day language. Specifically, our framework first contributes a translated Layman's terms dataset. Building upon the dataset, we then propose a semantics-based evaluation method, which is effective in mitigating the inflated numbers of BLEU and provides more robust evaluation. We show that training on the layman's terms dataset encourages models to focus on the semantics of the reports, as opposed to overfitting to learning the report templates. Last, we reveal a promising scaling law between the number of training examples and semantics gain provided by our dataset, compared to the inverse pattern brought by the original formats.
Authors: Mervat Abassy, Kareem Elozeiri, Alexander Aziz, Minh Ngoc Ta, Raj Vardhan Tomar, Bimarsha Adhikari, Saad El Dine Ahmed, Yuxia Wang, Osama Mohammed Afzal, Zhuohan Xie, Jonibek Mansurov, Ekaterina Artemova, Vladislav Mikhailov, Rui Xing, Jiahui Geng, Hasan Iqbal, Zain Muhammad Mujahid, Tarek Mahmoud, Akim Tsvigun, Alham Fikri Aji, Artem Shelmanov, Nizar Habash, Iryna Gurevych, Preslav Nakov
Abstract: The ease of access to large language models (LLMs) has enabled a widespread of machine-generated texts, and now it is often hard to tell whether a piece of text was human-written or machine-generated. This raises concerns about potential misuse, particularly within educational and academic domains. Thus, it is important to develop practical systems that can automate the process. Here, we present one such system, LLM-DetectAIve, designed for fine-grained detection. Unlike most previous work on machine-generated text detection, which focused on binary classification, LLM-DetectAIve supports four categories: (i) human-written, (ii) machine-generated, (iii) machine-written, then machine-humanized, and (iv) human-written, then machine-polished. Category (iii) aims to detect attempts to obfuscate the fact that a text was machine-generated, while category (iv) looks for cases where the LLM was used to polish a human-written text, which is typically acceptable in academic writing, but not in education. Our experiments show that LLM-DetectAIve can effectively identify the above four categories, which makes it a potentially useful tool in education, academia, and other domains. LLM-DetectAIve is publicly accessible at https://github.com/mbzuai-nlp/LLM-DetectAIve. The video describing our system is available at https://youtu.be/E8eT_bE7k8c.
URLs: https://github.com/mbzuai-nlp/LLM-DetectAIve., https://youtu.be/E8eT_bE7k8c.
Authors: Yijia Shao, Tianshi Li, Weiyan Shi, Yanchen Liu, Diyi Yang
Abstract: As language models (LMs) are widely utilized in personalized communication scenarios (e.g., sending emails, writing social media posts) and endowed with a certain level of agency, ensuring they act in accordance with the contextual privacy norms becomes increasingly critical. However, quantifying the privacy norm awareness of LMs and the emerging privacy risk in LM-mediated communication is challenging due to (1) the contextual and long-tailed nature of privacy-sensitive cases, and (2) the lack of evaluation approaches that capture realistic application scenarios. To address these challenges, we propose PrivacyLens, a novel framework designed to extend privacy-sensitive seeds into expressive vignettes and further into agent trajectories, enabling multi-level evaluation of privacy leakage in LM agents' actions. We instantiate PrivacyLens with a collection of privacy norms grounded in privacy literature and crowdsourced seeds. Using this dataset, we reveal a discrepancy between LM performance in answering probing questions and their actual behavior when executing user instructions in an agent setup. State-of-the-art LMs, like GPT-4 and Llama-3-70B, leak sensitive information in 25.68% and 38.69% of cases, even when prompted with privacy-enhancing instructions. We also demonstrate the dynamic nature of PrivacyLens by extending each seed into multiple trajectories to red-team LM privacy leakage risk. Dataset and code are available at https://github.com/SALT-NLP/PrivacyLens.
Authors: Benjamin L. Badger
Abstract: Attention mechanisms that confer selective focus on a strict subset of input elements are nearly ubiquitous in language models today. We posit there to be downside to the use of attention: most input information is lost. In support of this idea we observe poor input representation accuracy in transformers and more accurate representation in what we term masked mixers, which replace self-attention with masked convolutions. The masked mixer learns causal language modeling more efficiently than early transformer implementations and even outperforms optimized, current transformers when training on small ($n_{ctx}<512$) but not larger context windows. Evidence is presented for the hypothesis that differences in transformer and masked mixer training efficiencies for various tasks are best predicted by input representation accuracy, or equivalently global invertibility. We hypothesize that the information loss exhibited by transformers would be more detrimental to retrieval than generation, as the former is more closely approximated by a bijective and thus invertible function. We find that masked mixers are more effective retrieval models both when the pretrained embedding model is unchanged as well as when the embedding model is modified via cosine similarity-based InfoNCE loss minimization. A small masked mixer is shown to outperform a large and near state-of-the-art transformer-based retrieval model, despite the latter being trained with many orders of magnitude more data and compute.
Authors: Zhe Xu, Jiasheng Ye, Xiaoran Liu, Xiangyang Liu, Tianxiang Sun, Zhigeng Liu, Qipeng Guo, Linlin Li, Qun Liu, Xuanjing Huang, Xipeng Qiu
Abstract: Recently, significant efforts have been devoted to enhancing the long-context capabilities of Large Language Models (LLMs), particularly in long-context reasoning. To facilitate this research, we propose \textbf{DetectiveQA}, a dataset specifically designed for narrative reasoning within long contexts. We leverage detective novels, averaging over 100k tokens, to create a dataset containing 1200 human-annotated questions in both Chinese and English, each paired with corresponding reference reasoning steps. Furthermore, we introduce a step-wise reasoning metric, which enhances the evaluation of LLMs' reasoning processes. We validate our approach and evaluate the mainstream LLMs, including GPT-4, Claude, and LLaMA, revealing persistent long-context reasoning challenges and demonstrating their evidence-retrieval challenges. Our findings offer valuable insights into the study of long-context reasoning and lay the base for more rigorous evaluations.
Authors: Yifan Zhang, Yang Yuan, Andrew Chi-Chih Yao
Abstract: We introduce Diagram of Thought (DoT), a framework that models iterative reasoning in large language models (LLMs) as the construction of a directed acyclic graph (DAG) within a single model. Unlike conventional approaches that represent reasoning as linear chains or tree structures, DoT organizes propositions, critiques, refinements, and verifications into a unified DAG, enabling the exploration of complex reasoning pathways while preserving logical consistency. In this framework, each node encapsulates a proposition at various stages of evaluation, thereby facilitating iterative self-improvement through detailed natural language feedback. By leveraging auto-regressive next-token prediction augmented with role-specific tokens, DoT seamlessly transitions between generating ideas and engaging in critical evaluation, offering richer, context-aware feedback than binary signals. Moreover, we establish a rigorous mathematical foundation for DoT through Topos Theory, ensuring soundness and consistency in the reasoning process. This integrated approach not only simplifies both training and inference by eliminating the need for multiple models or external control mechanisms but also provides a principled framework for the design of next-generation reasoning-specialized models.
Authors: Kangsheng Wang, Xiao Zhang, Zizheng Guo, Tianyu Hu, Huimin Ma
Abstract: Chain-based reasoning methods like chain of thought (CoT) play a rising role in solving reasoning tasks for large language models (LLMs). However, the causal illusions between \textit{a step of reasoning} and \textit{corresponding state transitions} are becoming a significant obstacle to advancing LLMs' reasoning capabilities, especially in long-range reasoning tasks. This paper proposes a non-chain-based reasoning framework for simultaneous consideration of causal significance and consistency, i.e., the Causal Significance and Consistency Enhancer (CSCE). We customize LLM's loss function utilizing treatment effect assessments to enhance its reasoning ability from two aspects: causal significance and consistency. This ensures that the model captures essential causal relationships and maintains robust and consistent performance across various scenarios. Additionally, we transform the reasoning process from the cascading multiple one-step reasoning commonly used in Chain-Based methods, like CoT, to a causal-enhanced method that outputs the entire reasoning process in one go, further improving the model's reasoning efficiency. Extensive experiments show that our method improves both the reasoning success rate and speed. These improvements further demonstrate that non-chain-based methods can also aid LLMs in completing reasoning tasks.
Authors: Angela Lopez-Cardona, Carlos Segura, Alexandros Karatzoglou, Sergi Abadal, Ioannis Arapakis
Abstract: Advancements in Natural Language Processing (NLP), have led to the emergence of Large Language Models (LLMs) such as GPT, Llama, Claude, and Gemini, which excel across a range of tasks but require extensive fine-tuning to align their outputs with human expectations. A widely used method for achieving this alignment is Reinforcement Learning from Human Feedback (RLHF), which, despite its success, faces challenges in accurately modelling human preferences. In this paper, we introduce GazeReward, a novel framework that integrates implicit feedback -- and specifically eye-tracking (ET) data -- into the Reward Model (RM). In addition, we explore how ET-based features can provide insights into user preferences. Through ablation studies we test our framework with different integration methods, LLMs, and ET generator models, demonstrating that our approach significantly improves the accuracy of the RM on established human preference datasets. This work advances the ongoing discussion on optimizing AI alignment with human values, exploring the potential of cognitive data for shaping future NLP research.
Authors: Fantine Huot, Reinald Kim Amplayo, Jennimaria Palomaki, Alice Shoshana Jakobovits, Elizabeth Clark, Mirella Lapata
Abstract: Writing compelling fiction is a multifaceted process combining elements such as crafting a plot, developing interesting characters, and using evocative language. While large language models (LLMs) show promise for story writing, they currently rely heavily on intricate prompting, which limits their use. We propose Agents' Room, a generation framework inspired by narrative theory, that decomposes narrative writing into subtasks tackled by specialized agents. To illustrate our method, we introduce Tell Me A Story, a high-quality dataset of complex writing prompts and human-written stories, and a novel evaluation framework designed specifically for assessing long narratives. We show that Agents' Room generates stories that are preferred by expert evaluators over those produced by baseline systems by leveraging collaboration and specialization to decompose the complex story writing task into tractable components. We provide extensive analysis with automated and human-based metrics of the generated output.
Authors: Tobias Leemann, Periklis Petridis, Giuseppe Vietri, Dionysis Manousakas, Aaron Roth, Sergul Aydore
Abstract: While retrieval-augmented generation (RAG) has been shown to enhance factuality of large language model (LLM) outputs, LLMs still suffer from hallucination, generating incorrect or irrelevant information. A common detection strategy involves prompting the LLM again to assess whether its response is grounded in the retrieved evidence, but this approach is costly. Alternatively, lightweight natural language inference (NLI) models for efficient grounding verification can be used at inference time. While existing pre-trained NLI models offer potential solutions, their performance remains subpar compared to larger models on realistic RAG inputs. RAG inputs are more complex than most datasets used for training NLI models and have characteristics specific to the underlying knowledge base, requiring adaptation of the NLI models to a specific target domain. Additionally, the lack of labeled instances in the target domain makes supervised domain adaptation, e.g., through fine-tuning, infeasible. To address these challenges, we introduce Automatic Generative Domain Adaptation (Auto-GDA). Our framework enables unsupervised domain adaptation through synthetic data generation. Unlike previous methods that rely on handcrafted filtering and augmentation strategies, Auto-GDA employs an iterative process to continuously improve the quality of generated samples using weak labels from less efficient teacher models and discrete optimization to select the most promising augmented samples. Experimental results demonstrate the effectiveness of our approach, with models fine-tuned on synthetic data using Auto-GDA often surpassing the performance of the teacher model and reaching the performance level of LLMs at 10% of their computational cost.
Authors: Yirong Sun, Dawei Zhu, Yanjun Chen, Erjia Xiao, Xinghao Chen, Xiaoyu Shen
Abstract: Large language models (LLMs) have excelled in various NLP tasks, including machine translation (MT), yet most studies focus on sentence-level translation. This work investigates the inherent capability of instruction-tuned LLMs for document-level translation (docMT). Unlike prior approaches that require specialized techniques, we evaluate LLMs by directly prompting them to translate entire documents in a single pass. Our results show that this method improves translation quality compared to translating sentences separately, even without document-level fine-tuning. However, this advantage is not reflected in BLEU scores, which often favor sentence-based translations. We propose using the LLM-as-a-judge paradigm for evaluation, where GPT-4 is used to assess document coherence, accuracy, and fluency in a more nuanced way than n-gram-based metrics. Overall, our work demonstrates that instruction-tuned LLMs can effectively leverage document context for translation. However, we caution against using BLEU scores for evaluating docMT, as they often provide misleading outcomes, failing to capture the quality of document-level translation. Code and the outputs from GPT4-as-a-judge are available at https://github.com/EIT-NLP/BLEUless_DocMT
Authors: Aniket Deroy, Subhankar Maity
Abstract: Code-mixing, the integration of lexical and grammatical elements from multiple languages within a single sentence, is a widespread linguistic phenomenon, particularly prevalent in multilingual societies. In India, social media users frequently engage in code-mixed conversations using the Roman script, especially among migrant communities who form online groups to share relevant local information. This paper focuses on the challenges of extracting relevant information from code-mixed conversations, specifically within Roman transliterated Bengali mixed with English. This study presents a novel approach to address these challenges by developing a mechanism to automatically identify the most relevant answers from code-mixed conversations. We have experimented with a dataset comprising of queries and documents from Facebook, and Query Relevance files (QRels) to aid in this task. Our results demonstrate the effectiveness of our approach in extracting pertinent information from complex, code-mixed digital conversations, contributing to the broader field of natural language processing in multilingual and informal text environments. We use GPT-3.5 Turbo via prompting alongwith using the sequential nature of relevant documents to frame a mathematical model which helps to detect relevant documents corresponding to a query.
Authors: Aniket Deroy, Subhankar Maity
Abstract: The widespread use of social media platforms like Twitter and Facebook has enabled people of all ages to share their thoughts and experiences, leading to an immense accumulation of user-generated content. However, alongside the benefits, these platforms also face the challenge of managing hate speech and offensive content, which can undermine rational discourse and threaten democratic values. As a result, there is a growing need for automated methods to detect and mitigate such content, especially given the complexity of conversations that may require contextual analysis across multiple languages, including code-mixed languages like Hinglish, German-English, and Bangla. We participated in the English task where we have to classify English tweets into two categories namely Hate and Offensive and Non Hate-Offensive. In this work, we experiment with state-of-the-art large language models like GPT-3.5 Turbo via prompting to classify tweets into Hate and Offensive or Non Hate-Offensive. In this study, we evaluate the performance of a classification model using Macro-F1 scores across three distinct runs. The Macro-F1 score, which balances precision and recall across all classes, is used as the primary metric for model evaluation. The scores obtained are 0.756 for run 1, 0.751 for run 2, and 0.754 for run 3, indicating a high level of performance with minimal variance among the runs. The results suggest that the model consistently performs well in terms of precision and recall, with run 1 showing the highest performance. These findings highlight the robustness and reliability of the model across different runs.
Authors: Qizhou Chen, Chengyu Wang, Dakan Wang, Taolin Zhang, Wangyue Li, Xiaofeng He
Abstract: Model editing aims to correct inaccurate knowledge, update outdated information, and incorporate new data into Large Language Models (LLMs) without the need for retraining. This task poses challenges in lifelong scenarios where edits must be continuously applied for real-world applications. While some editors demonstrate strong robustness for lifelong editing in pure LLMs, Vision LLMs (VLLMs), which incorporate an additional vision modality, are not directly adaptable to existing LLM editors. In this paper, we propose LiveEdit, a LIfelong Vision language modEl Edit to bridge the gap between lifelong LLM editing and VLLMs. We begin by training an editing expert generator to independently produce low-rank experts for each editing instance, with the goal of correcting the relevant responses of the VLLM. A hard filtering mechanism is developed to utilize visual semantic knowledge, thereby coarsely eliminating visually irrelevant experts for input queries during the inference stage of the post-edited model. Finally, to integrate visually relevant experts, we introduce a soft routing mechanism based on textual semantic relevance to achieve multi-expert fusion. For evaluation, we establish a benchmark for lifelong VLLM editing. Extensive experiments demonstrate that LiveEdit offers significant advantages in lifelong VLLM editing scenarios. Further experiments validate the rationality and effectiveness of each module design in LiveEdit.
Authors: Tohida Rehman, Debarshi Kumar Sanyal, Samiran Chattopadhyay
Abstract: Artificial intelligence systems significantly impact the environment, particularly in natural language processing (NLP) tasks. These tasks often require extensive computational resources to train deep neural networks, including large-scale language models containing billions of parameters. This study analyzes the trade-offs between energy consumption and performance across three neural language models: two pre-trained models (T5-base and BART-base), and one large language model (LLaMA-3-8B). These models were fine-tuned for the text summarization task, focusing on generating research paper highlights that encapsulate the core themes of each paper. The carbon footprint associated with fine-tuning each model was measured, offering a comprehensive assessment of their environmental impact. It is observed that LLaMA-3-8B produces the largest carbon footprint among the three models. A wide range of evaluation metrics, including ROUGE, METEOR, MoverScore, BERTScore, and SciBERTScore, were employed to assess the performance of the models on the given task. This research underscores the importance of incorporating environmental considerations into the design and implementation of neural language models and calls for the advancement of energy-efficient AI methodologies.
Authors: Bidyarthi Paul, Faika Fairuj Preotee, Shuvashis Sarker, Shamim Rahim Refat, Shifat Islam, Tashreef Muhammad, Mohammad Ashraful Hoque, Shahriar Manzoor
Abstract: ANCHOLIK-NER is a linguistically diverse dataset for Named Entity Recognition (NER) in Bangla regional dialects, capturing variations across Sylhet, Chittagong, Barishal, Noakhali, and Mymensingh. The dataset has around 17,405 sentences, 3,481 sentences per region. The data was collected from two publicly available datasets and through web scraping from various online newspapers, articles. To ensure high-quality annotations, the BIO tagging scheme was employed, and professional annotators with expertise in regional dialects carried out the labeling process. The dataset is structured into separate subsets for each region and is available in CSV format. Each entry contains textual data along with identified named entities and their corresponding annotations. Named entities are categorized into ten distinct classes: Person, Location, Organization, Food, Animal, Colour, Role, Relation, Object, and Miscellaneous. This dataset serves as a valuable resource for developing and evaluating NER models for Bangla dialectal variations, contributing to regional language processing and low-resource NLP applications. It can be utilized to enhance NER systems in Bangla dialects, improve regional language understanding, and support applications in machine translation, information retrieval, and conversational AI.
Authors: Xiaoqian Liu, Ke Wang, Yongbin Li, Yuchuan Wu, Wentao Ma, Aobo Kong, Fei Huang, Jianbin Jiao, Junge Zhang
Abstract: Large Language Models (LLMs) have shown impressive reasoning capabilities in well-defined problems with clear solutions, such as mathematics and coding. However, they still struggle with complex real-world scenarios like business negotiations, which require strategic reasoning-an ability to navigate dynamic environments and align long-term goals amidst uncertainty. Existing methods for strategic reasoning face challenges in adaptability, scalability, and transferring strategies to new contexts. To address these issues, we propose explicit policy optimization (EPO) for strategic reasoning, featuring an LLM that provides strategies in open-ended action space and can be plugged into arbitrary LLM agents to motivate goal-directed behavior. To improve adaptability and policy transferability, we train the strategic reasoning model via multi-turn reinforcement learning (RL) using process rewards and iterative self-play, without supervised fine-tuning (SFT) as a preliminary step. Experiments across social and physical domains demonstrate EPO's ability of long-term goal alignment through enhanced strategic reasoning, achieving state-of-the-art performance on social dialogue and web navigation tasks. Our findings reveal various collaborative reasoning mechanisms emergent in EPO and its effectiveness in generating novel strategies, underscoring its potential for strategic reasoning in real-world applications.
Authors: Zhuoran Li, Chunming Hu, Junfan Chen, Zhijun Chen, Richong Zhang
Abstract: Word order difference between source and target languages is a major obstacle to cross-lingual transfer, especially in the dependency parsing task. Current works are mostly based on order-agnostic models or word reordering to mitigate this problem. However, such methods either do not leverage grammatical information naturally contained in word order or are computationally expensive as the permutation space grows exponentially with the sentence length. Moreover, the reordered source sentence with an unnatural word order may be a form of noising that harms the model learning. To this end, we propose an Implicit Word Reordering framework with Knowledge Distillation (IWR-KD). This framework is inspired by that deep networks are good at learning feature linearization corresponding to meaningful data transformation, e.g. word reordering. To realize this idea, we introduce a knowledge distillation framework composed of a word-reordering teacher model and a dependency parsing student model. We verify our proposed method on Universal Dependency Treebanks across 31 different languages and show it outperforms a series of competitors, together with experimental analysis to illustrate how our method works towards training a robust parser.
Authors: Juntai Cao, Xiang Zhang, Raymond Li, Chuyuan Li, Shafiq Joty, Giuseppe Carenini
Abstract: Recent advances in test-time scaling have shown promising results in improving Large Language Models (LLMs) performance through strategic computation allocation during inference. While this approach has demonstrated strong performance improvements in logical and mathematical reasoning tasks, its application to natural language generation (NLG), especially summarization, has yet to be explored. Multi-Document Summarization (MDS) is a challenging task that focuses on extracting and synthesizing useful information from multiple lengthy documents. Unlike reasoning tasks, MDS requires a more nuanced approach to prompt design and ensemble, as there is no "best" prompt to satisfy diverse summarization requirements. To address this, we propose a novel framework that leverages inference-time scaling for this task. Precisely, we take prompt ensemble approach by leveraging various prompt to first generate candidate summaries and then ensemble them with an aggregator to produce a refined summary. We also introduce two new evaluation metrics: Consistency-Aware Preference (CAP) score and LLM Atom-Content-Unit (ACU) score, to enhance LLM's contextual understanding while mitigating its positional bias. Extensive experiments demonstrate the effectiveness of our approach in improving summary quality while identifying and analyzing the scaling boundaries in summarization tasks.
Authors: S M Sarwar
Abstract: With the increasing prevalence of mental health conditions worldwide, AI-powered chatbots and conversational agents have emerged as accessible tools to support mental health. However, deploying Large Language Models (LLMs) in mental healthcare applications raises significant privacy concerns, especially regarding regulations like HIPAA and GDPR. In this work, we propose FedMentalCare, a privacy-preserving framework that leverages Federated Learning (FL) combined with Low-Rank Adaptation (LoRA) to fine-tune LLMs for mental health analysis. We investigate the performance impact of varying client data volumes and model architectures (e.g., MobileBERT and MiniLM) in FL environments. Our framework demonstrates a scalable, privacy-aware approach for deploying LLMs in real-world mental healthcare scenarios, addressing data security and computational efficiency challenges.
Authors: Zuchen Gao, Zizheng Zhan, Xianming Li, Erxin Yu, Haotian Zhang, Bin Chen, Yuqun Zhang, Jing Li
Abstract: Code embeddings capture the semantic representations of code and are crucial for various code-related large language model (LLM) applications, such as code search. Previous training primarily relies on optimizing the InfoNCE loss by comparing positive natural language (NL)-code pairs with in-batch negatives. However, due to the sparse nature of code contexts, training solely by comparing the major differences between positive and negative pairs may fail to capture deeper semantic nuances. To address this issue, we propose a novel order-augmented strategy for improved code search (OASIS). It leverages order-based similarity labels to train models to capture subtle differences in similarity among negative pairs. Extensive benchmark evaluations demonstrate that our OASIS model significantly outperforms previous state-of-the-art models focusing solely on major positive-negative differences. It underscores the value of exploiting subtle differences among negative pairs with order labels for effective code embedding training.
Authors: Cyriel Mallart (UR2, LIDILE), Andrew Simpkin (UPCit\'e, ALTAE), Nicolas Ballier (UPCit\'e, ALTAE), Paula Liss\'on (UNIR), R\'emi Venant (UM, LIUM), Jen-Yu Li (UR2, LIDILE), Bernardo Stearns (NUI Galway, INSIGHT), Thomas Gaillat (LIDILE, UR2)
Abstract: This article addresses Second Language (L2) writing development through an investigation of new grammatical and structural complexity metrics. We explore the paradigmatic production in learner English by linking language functions to specific grammatical paradigms. Using the EFCAMDAT as a gold standard and a corpus of French learners as an external test set, we employ a supervised learning framework to operationalise and evaluate seven microsystems. We show that learner levels are associated with the seven microsystems (MS). Using ordinal regression modelling for evaluation, the results show that all MS are significant but yield a low impact if taken individually. However, their influence is shown to be impactful if taken as a group. These microsystems and their measurement method suggest that it is possible to use them as part of broader-purpose CALL systems focused on proficiency assessment.
Authors: Laurie Burchell, Ona de Gibert, Nikolay Arefyev, Mikko Aulamo, Marta Ba\~n\'on, Pinzhen Chen, Mariia Fedorova, Liane Guillou, Barry Haddow, Jan Haji\v{c}, Jind\v{r}ich Helcl, Erik Henriksson, Mateusz Klimaszewski, Ville Komulainen, Andrey Kutuzov, Joona Kyt\"oniemi, Veronika Laippala, Petter M{\ae}hlum, Bhavitvya Malik, Farrokh Mehryary, Vladislav Mikhailov, Nikita Moghe, Amanda Myntti, Dayy\'an O'Brien, Stephan Oepen, Proyag Pal, Jousia Piha, Sampo Pyysalo, Gema Ram\'irez-S\'anchez, David Samuel, Pavel Stepachev, J\"org Tiedemann, Du\v{s}an Vari\v{s}, Tereza Vojt\v{e}chov\'a, Jaume Zaragoza-Bernabeu
Abstract: Training state-of-the-art large language models requires vast amounts of clean and diverse textual data. However, building suitable multilingual datasets remains a challenge. In this work, we present HPLT v2, a collection of high-quality multilingual monolingual and parallel corpora. The monolingual portion of the data contains 8T tokens covering 193 languages, while the parallel data contains 380M sentence pairs covering 51 languages. We document the entire data pipeline and release the code to reproduce it. We provide extensive analysis of the quality and characteristics of our data. Finally, we evaluate the performance of language models and machine translation systems trained on HPLT v2, demonstrating its value.
Authors: Sinuo Liu, Chenyang Lyu, Minghao Wu, Longyue Wang, Weihua Luo, Kaifu Zhang, Zifu Shang
Abstract: Recent advances in Large Reasoning Models (LRMs), particularly those leveraging Chain-of-Thought reasoning (CoT), have opened brand new possibility for Machine Translation (MT). This position paper argues that LRMs substantially transformed traditional neural MT as well as LLMs-based MT paradigms by reframing translation as a dynamic reasoning task that requires contextual, cultural, and linguistic understanding and reasoning. We identify three foundational shifts: 1) contextual coherence, where LRMs resolve ambiguities and preserve discourse structure through explicit reasoning over cross-sentence and complex context or even lack of context; 2) cultural intentionality, enabling models to adapt outputs by inferring speaker intent, audience expectations, and socio-linguistic norms; 3) self-reflection, LRMs can perform self-reflection during the inference time to correct the potential errors in translation especially extremely noisy cases, showing better robustness compared to simply mapping X->Y translation. We explore various scenarios in translation including stylized translation, document-level translation and multimodal translation by showcasing empirical examples that demonstrate the superiority of LRMs in translation. We also identify several interesting phenomenons for LRMs for MT including auto-pivot translation as well as the critical challenges such as over-localisation in translation and inference efficiency. In conclusion, we think that LRMs redefine translation systems not merely as text converters but as multilingual cognitive agents capable of reasoning about meaning beyond the text. This paradigm shift reminds us to think of problems in translation beyond traditional translation scenarios in a much broader context with LRMs - what we can achieve on top of it.
Authors: Yichi Zhang, Zhuo Chen, Lingbing Guo, Yajing Xu, Binbin Hu, Ziqi Liu, Wen Zhang, Huajun Chen
Abstract: Learning high-quality multi-modal entity representations is an important goal of multi-modal knowledge graph (MMKG) representation learning, which can enhance reasoning tasks within the MMKGs, such as MMKG completion (MMKGC). The main challenge is to collaboratively model the structural information concealed in massive triples and the multi-modal features of the entities. Existing methods focus on crafting elegant entity-wise multi-modal fusion strategies, yet they overlook the utilization of multi-perspective features concealed within the modalities under diverse relational contexts. To address this issue, we introduce a novel framework with Mixture of Modality Knowledge experts (MoMoK for short) to learn adaptive multi-modal entity representations for better MMKGC. We design relation-guided modality knowledge experts to acquire relation-aware modality embeddings and integrate the predictions from multi-modalities to achieve joint decisions. Additionally, we disentangle the experts by minimizing their mutual information. Experiments on four public MMKG benchmarks demonstrate the outstanding performance of MoMoK under complex scenarios.
Authors: Ziyang Wang, Shoubin Yu, Elias Stengel-Eskin, Jaehong Yoon, Feng Cheng, Gedas Bertasius, Mohit Bansal
Abstract: Long-form video understanding is complicated by the high redundancy of video data and the abundance of query-irrelevant information. To tackle these challenges, we propose VideoTree, a training-free framework which builds a query-adaptive and hierarchical video representation for LLM reasoning over long-form videos. First, VideoTree extracts query-relevant information from the input video through an iterative process, progressively refining the selection of keyframes based on their relevance to the query. Furthermore, VideoTree leverages the inherent hierarchical structure of long video data, which is often overlooked by existing LLM-based methods. Specifically, we incorporate multi-granularity information into a tree-based representation, allowing VideoTree to extract query-relevant details from long videos in a coarse-to-fine manner. This enables the model to effectively handle a wide range of video queries with varying levels of detail. Finally, VideoTree aggregates the hierarchical query-relevant information within the tree structure and feeds it into an LLM reasoning model to answer the query. Our experiments show that our method improves both reasoning accuracy and efficiency. Specifically, VideoTree outperforms existing training-free approaches on EgoSchema and NExT-QA with less inference time, achieving 61.1% and 75.6% accuracy on the test set without additional video-specific training. Moreover, on the long split of Video-MME (average 44 minutes), VideoTree achieves better performance than GPT-4V and many other MLLMs that were extensively trained on video data.
Authors: Wenbin An, Feng Tian, Sicong Leng, Jiahao Nie, Haonan Lin, QianYing Wang, Ping Chen, Xiaoqin Zhang, Shijian Lu
Abstract: Despite great success across various multimodal tasks, Large Vision-Language Models (LVLMs) often encounter object hallucinations with generated textual responses being inconsistent with the actual objects in images. We examine different LVLMs and pinpoint that one root cause of object hallucinations lies with deficient attention on discriminative image features. Specifically, LVLMs often predominantly attend to prompt-irrelevant global features instead of prompt-relevant local features, undermining their visual grounding capacity and leading to object hallucinations. We propose Assembly of Global and Local Attention (AGLA), a training-free and plug-and-play approach that mitigates hallucinations by assembling global features for response generation and local features for visual discrimination simultaneously. Specifically, we introduce an image-prompt matching scheme that captures prompt-relevant local features from images, leading to an augmented view of the input image where prompt-relevant content is highlighted while irrelevant distractions are suppressed. Hallucinations can thus be mitigated with a calibrated logit distribution that is from generative global features of the original image and discriminative local features of the augmented image. Extensive experiments show the superiority of AGLA in LVLM hallucination mitigation, demonstrating its wide applicability across both discriminative and generative tasks. Our code is available at https://github.com/Lackel/AGLA.
Authors: Zhengzhuo Xu, Bowen Qu, Yiyan Qi, Sinan Du, Chengjin Xu, Chun Yuan, Jian Guo
Abstract: Automatic chart understanding is crucial for content comprehension and document parsing. Multimodal Large Language Models (MLLMs) have demonstrated remarkable capabilities in chart understanding through domain-specific alignment and fine-tuning. However, current MLLMs still struggle to provide faithful data and reliable analysis only based on charts. To address it, we propose ChartMoE, which employs the Mixture of Expert (MoE) architecture to replace the traditional linear projector to bridge the modality gap. Specifically, we train several linear connectors through distinct alignment tasks, which are utilized as the foundational initialization parameters for different experts. Additionally, we introduce ChartMoE-Align, a dataset with nearly 1 million chart-table-JSON-code quadruples to conduct three alignment tasks (chart-table/JSON/code). Combined with the vanilla connector, we initialize different experts diversely and adopt high-quality knowledge learning to further refine the MoE connector and LLM parameters. Extensive experiments demonstrate the effectiveness of the MoE connector and our initialization strategy, e.g., ChartMoE improves the accuracy of the previous state-of-the-art from 80.48\% to 84.64\% on the ChartQA benchmark.
Authors: Payam Saeedi, Mahsa Goodarzi, M Abdullah Canbaz
Abstract: We investigate the presence of cognitive biases in three large language models (LLMs): GPT-4o, Gemma 2, and Llama 3.1. The study uses 1,500 experiments across nine established cognitive biases to evaluate the models' responses and consistency. GPT-4o demonstrated the strongest overall performance. Gemma 2 showed strengths in addressing the sunk cost fallacy and prospect theory, however its performance varied across different biases. Llama 3.1 consistently underperformed, relying on heuristics and exhibiting frequent inconsistencies and contradictions. The findings highlight the challenges of achieving robust and generalizable reasoning in LLMs, and underscore the need for further development to mitigate biases in artificial general intelligence (AGI). The study emphasizes the importance of integrating statistical reasoning and ethical considerations in future AI development.
Authors: Kevin Galim, Wonjun Kang, Yuchen Zeng, Hyung Il Koo, Kangwook Lee
Abstract: Deep State Space Models (SSMs), such as Mamba (Gu & Dao, 2024), have become powerful tools for language modeling, offering high performance and linear scalability with sequence length. However, the application of parameter-efficient fine-tuning (PEFT) methods to SSM-based models remains underexplored. We start by investigating two fundamental questions on existing PEFT methods: (i) How do they perform on SSM-based models? (ii) Which parameters should they target for optimal results? Our analysis shows that LoRA and its variants consistently outperform all other PEFT methods. While LoRA is effective for linear projection matrices, it fails on SSM modules-yet still outperforms other methods applicable to SSMs, indicating their limitations. This underscores the need for a specialized SSM tuning approach. To address this, we propose Sparse Dimension Tuning (SDT), a PEFT method tailored for SSM modules. Combining SDT for SSMs with LoRA for linear projection matrices, we achieve state-of-the-art performance across extensive experiments.
Authors: Abraham Israeli, David Jurgens, Daniel Romero
Abstract: The Internet has significantly expanded the potential for global collaboration, allowing millions of users to contribute to collective projects like Wikipedia. While prior work has assessed the success of online collaborations, most approaches are time-agnostic, evaluating success without considering its longevity. Research on the factors that ensure the long-term preservation of high-quality standards in online collaboration is scarce. In this study, we address this gap. We propose a novel metric, `Sustainable Success,' which measures the ability of collaborative efforts to maintain their quality over time. Using Wikipedia as a case study, we introduce the SustainPedia dataset, which compiles data from over 40K Wikipedia articles, including each article's sustainable success label and more than 300 explanatory features such as edit history, user experience, and team composition. Using this dataset, we develop machine learning models to predict the sustainable success of Wikipedia articles. Our best-performing model achieves a high AU-ROC score of 0.88 on average. Our analysis reveals important insights. For example, we find that the longer an article takes to be recognized as high-quality, the more likely it is to maintain that status over time (i.e., be sustainable). Additionally, user experience emerged as the most critical predictor of sustainability. Our analysis provides insights into broader collective actions beyond Wikipedia (e.g., online activism, crowdsourced open-source software), where the same social dynamics that drive success on Wikipedia might play a role. We make all data and code used for this study publicly available for further research.
Authors: Zongxia Li, Xiyang Wu, Hongyang Du, Huy Nghiem, Guangyao Shi
Abstract: Multimodal Vision Language Models (VLMs) have emerged as a transformative technology at the intersection of computer vision and natural language processing, enabling machines to perceive and reason about the world through both visual and textual modalities. For example, models such as CLIP, Claude, and GPT-4V demonstrate strong reasoning and understanding abilities on visual and textual data and beat classical single modality vision models on zero-shot classification. Despite their rapid advancements in research and growing popularity in applications, a comprehensive survey of existing studies on VLMs is notably lacking, particularly for researchers aiming to leverage VLMs in their specific domains. To this end, we provide a systematic overview of VLMs in the following aspects: model information of the major VLMs developed over the past five years (2019-2024); the main architectures and training methods of these VLMs; summary and categorization of the popular benchmarks and evaluation metrics of VLMs; the applications of VLMs including embodied agents, robotics, and video generation; the challenges and issues faced by current VLMs such as hallucination, fairness, and safety. Detailed collections including papers and model repository links are listed in https://github.com/zli12321/Awesome-VLM-Papers-And-Models.git.
URLs: https://github.com/zli12321/Awesome-VLM-Papers-And-Models.git.
Authors: Phil Ostheimer, Marius Kloft, Sophie Fellenz
Abstract: Recent advancements in language representation learning primarily emphasize language modeling for deriving meaningful representations, often neglecting style-specific considerations. This study addresses this gap by creating generic, sentence-level style embeddings crucial for style-centric tasks. Our approach is grounded on the premise that low-level text style changes can compose any high-level style. We hypothesize that applying this concept to representation learning enables the development of versatile text style embeddings. By fine-tuning a general-purpose text encoder using contrastive learning and standard cross-entropy loss, we aim to capture these low-level style shifts, anticipating that they offer insights applicable to high-level text styles. The outcomes prompt us to reconsider the underlying assumptions as the results do not always show that the learned style representations capture high-level text styles.
Authors: Kelsey Allen, Ishita Dasgupta, Eliza Kosoy, Andrew K. Lampinen
Abstract: Inductive biases are what allow learners to make guesses in the absence of conclusive evidence. These biases have often been studied in cognitive science using concepts or categories -- e.g. by testing how humans generalize a new category from a few examples that leave the category boundary ambiguous. We use these approaches to study generalization in foundation models during in-context learning. Modern foundation models can condition on both vision and text, and differences in how they interpret and learn from these different modalities is an emerging area of study. Here, we study how their generalizations vary by the modality in which stimuli are presented, and the way the stimuli are described in text. We study these biases with three different experimental paradigms, across three different vision-language models. We find that the models generally show some bias towards generalizing according to shape over color. This shape bias tends to be amplified when the examples are presented visually. By contrast, when examples are presented in text, the ordering of adjectives affects generalization. However, the extent of these effects vary across models and paradigms. These results help to reveal how vision-language models represent different types of inputs in context, and may have practical implications for the use of vision-language models.
Authors: Yuyang Huang, Yuhan Liu, Haryadi S. Gunawi, Beibin Li, Changho Hwang
Abstract: Continual learning has become a promising solution to refine large language models incrementally by leveraging user feedback. In particular, online continual learning - iteratively training the model with small batches of user feedback - has demonstrated notable performance improvements. However, the existing practice of separating training and serving processes forces the online trainer to recompute the intermediate results already done during serving. Such redundant computations can account for 30%-42% of total training time. In this paper, we propose Alchemist, to the best of our knowledge, the first online continual learning system that efficiently reuses serving activations to increase training throughput. Alchemist introduces two key techniques: (1) recording and storing activations and KV cache only during the prefill phase to minimize latency and memory overhead; and (2) smart activation offloading and hedging. Evaluations with inputs of varied token length sampled from ShareGPT dataset show that compared with a separate training cluster, Alchemist significantly increases training throughput by up to 1.72x, reduces up to 47% memory usage during training, and supports up to 2x more training tokens - all while maintaining negligible impact on serving latency.
Authors: Leonardo Berti, Flavio Giorgi, Gjergji Kasneci
Abstract: Large Language Models (LLMs) are leading a new technological revolution as one of the most promising research streams toward artificial general intelligence. The scaling of these models, accomplished by increasing the number of parameters and the magnitude of the training datasets, has been linked to various so-called emergent abilities that were previously unobserved. These emergent abilities, ranging from advanced reasoning and in-context learning to coding and problem-solving, have sparked an intense scientific debate: Are they truly emergent, or do they simply depend on external factors, such as training dynamics, the type of problems, or the chosen metric? What underlying mechanism causes them? Despite their transformative potential, emergent abilities remain poorly understood, leading to misconceptions about their definition, nature, predictability, and implications. In this work, we shed light on emergent abilities by conducting a comprehensive review of the phenomenon, addressing both its scientific underpinnings and real-world consequences. We first critically analyze existing definitions, exposing inconsistencies in conceptualizing emergent abilities. We then explore the conditions under which these abilities appear, evaluating the role of scaling laws, task complexity, pre-training loss, quantization, and prompting strategies. Our review extends beyond traditional LLMs and includes Large Reasoning Models (LRMs), which leverage reinforcement learning and inference-time search to amplify reasoning and self-reflection. However, emergence is not inherently positive. As AI systems gain autonomous reasoning capabilities, they also develop harmful behaviors, including deception, manipulation, and reward hacking. We highlight growing concerns about safety and governance, emphasizing the need for better evaluation frameworks and regulatory oversight.
Authors: Dylan J. Foster, Zakaria Mhammedi, Dhruv Rohatgi
Abstract: Language model alignment (or, reinforcement learning) techniques that leverage active exploration -- deliberately encouraging the model to produce diverse, informative responses -- offer the promise of super-human capabilities. However, current understanding of algorithm design primitives for computationally efficient exploration with language models is limited. To better understand how to leverage access to powerful pre-trained generative models to improve the efficiency of exploration, we introduce a new computational framework for RL with language models, in which the learner interacts with the model through a sampling oracle. Focusing on the linear softmax model parameterization, we provide new results that reveal the computational-statistical tradeoffs of efficient exploration: 1. Necessity of coverage: Coverage refers to the extent to which the pre-trained model covers near-optimal responses -- a form of hidden knowledge. We show that coverage, while not necessary for data efficiency, lower bounds the runtime of any algorithm in our framework. 2. Inference-time exploration: We introduce a new algorithm, SpannerSampling, which obtains optimal data efficiency and is computationally efficient whenever the pre-trained model enjoys sufficient coverage, matching our lower bound. SpannerSampling leverages inference-time computation with the pre-trained model to reduce the effective search space for exploration. 3. Insufficiency of training-time interventions: We contrast the result above by showing that training-time interventions that produce proper policies cannot achieve similar guarantees in polynomial time. 4. Computational benefits of multi-turn exploration: Finally, we show that under additional representational assumptions, one can achieve improved runtime (replacing sequence-level coverage with token-level coverage) through multi-turn exploration.
Authors: Ali Vosoughi, Dimitra Emmanouilidou, Hannes Gamper
Abstract: Integrating audio and visual data for training multimodal foundational models remains challenging. We present Audio-Video Vector Alignment (AVVA), which aligns audiovisual (AV) scene content beyond mere temporal synchronization via a Large Language Model (LLM)-based data curation pipeline. Specifically, AVVA scores and selects high-quality training clips using Whisper (speech-based audio foundation model) for audio and DINOv2 for video within a dual-encoder contrastive learning framework. Evaluations on AudioCaps, VALOR, and VGGSound demonstrate that this approach can achieve significant accuracy gains with substantially less curated data. For instance, AVVA yields a 7.6% improvement in top-1 accuracy for audio-to-video retrieval on VGGSound compared to ImageBind, despite training on only 192 hours of carefully filtered data (vs. 5800+ hours). Moreover, an ablation study highlights that trading data quantity for data quality improves performance, yielding respective top-3 accuracy increases of 47.8, 48.4, and 58.0 percentage points on AudioCaps, VALOR, and VGGSound over uncurated baselines. While these results underscore AVVA's data efficiency, we also discuss the overhead of LLM-driven curation and how it may be scaled or approximated in larger domains. Overall, AVVA provides a viable path toward more robust, text-free audiovisual learning with improved retrieval accuracy.
Authors: Ziyu Wan, Yunxiang Li, Yan Song, Hanjing Wang, Linyi Yang, Mark Schmidt, Jun Wang, Weinan Zhang, Shuyue Hu, Ying Wen
Abstract: Recent research on Reasoning of Large Language Models (LLMs) has sought to further enhance their performance by integrating meta-thinking -- enabling models to monitor, evaluate, and control their reasoning processes for more adaptive and effective problem-solving. However, current single-agent work lacks a specialized design for acquiring meta-thinking, resulting in low efficacy. To address this challenge, we introduce Reinforced Meta-thinking Agents (ReMA), a novel framework that leverages Multi-Agent Reinforcement Learning (MARL) to elicit meta-thinking behaviors, encouraging LLMs to think about thinking. ReMA decouples the reasoning process into two hierarchical agents: a high-level meta-thinking agent responsible for generating strategic oversight and plans, and a low-level reasoning agent for detailed executions. Through iterative reinforcement learning with aligned objectives, these agents explore and learn collaboration, leading to improved generalization and robustness. Experimental results demonstrate that ReMA outperforms single-agent RL baselines on complex reasoning tasks, including competitive-level mathematical benchmarks and LLM-as-a-Judge benchmarks. Comprehensive ablation studies further illustrate the evolving dynamics of each distinct agent, providing valuable insights into how the meta-thinking reasoning process enhances the reasoning capabilities of LLMs.
Authors: Ahmad Mustafa Anis, Hasnain Ali, Saquib Sarfraz
Abstract: Vision Language Models (VLMs) have demonstrated significant potential in various downstream tasks, including Image/Video Generation, Visual Question Answering, Multimodal Chatbots, and Video Understanding. However, these models often struggle with basic image transformations. This paper investigates the image-level understanding of VLMs, specifically CLIP by OpenAI and SigLIP by Google. Our findings reveal that these models lack comprehension of multiple image-level augmentations. To facilitate this study, we created an augmented version of the Flickr8k dataset, pairing each image with a detailed description of the applied transformation. We further explore how this deficiency impacts downstream tasks, particularly in image editing, and evaluate the performance of state-of-the-art Image2Image models on simple transformations.
Authors: Nicholas Roberts, Niladri Chatterji, Sharan Narang, Mike Lewis, Dieuwke Hupkes
Abstract: Scaling laws are a critical component of the LLM development pipeline, most famously as a way to forecast training decisions such as 'compute-optimally' trading-off parameter count and dataset size, alongside a more recent growing list of other crucial decisions. In this work, we ask whether compute-optimal scaling behaviour can be skill-dependent. In particular, we examine knowledge and reasoning-based skills such as knowledge-based QA and code generation, and we answer this question in the affirmative: scaling laws are skill-dependent. Next, to understand whether skill-dependent scaling is an artefact of the pretraining datamix, we conduct an extensive ablation of different datamixes and find that, also when correcting for datamix differences, knowledge and code exhibit fundamental differences in scaling behaviour. We conclude with an analysis of how our findings relate to standard compute-optimal scaling using a validation set, and find that a misspecified validation set can impact compute-optimal parameter count by nearly 50%, depending on its skill composition.