Authors: Manil Shrestha, Edward Kim
Abstract: Multi-hop question answering over knowledge graphs remains computationally challenging due to the combinatorial explosion of possible reasoning paths. Recent approaches rely on expensive Large Language Model (LLM) inference for both entity linking and path ranking, limiting their practical deployment. Additionally, LLM-generated answers often lack verifiable grounding in structured knowledge. We present two complementary hybrid algorithms that address both efficiency and verifiability: (1) LLM-Guided Planning that uses a single LLM call to predict relation sequences executed via breadth-first search, achieving near-perfect accuracy (micro-F1 > 0.90) while ensuring all answers are grounded in the knowledge graph, and (2) Embedding-Guided Neural Search that eliminates LLM calls entirely by fusing text and graph embeddings through a lightweight 6.7M-parameter edge scorer, achieving over 100 times speedup with competitive accuracy. Through knowledge distillation, we compress planning capability into a 4B-parameter model that matches large-model performance at zero API cost. Evaluation on MetaQA demonstrates that grounded reasoning consistently outperforms ungrounded generation, with structured planning proving more transferable than direct answer generation. Our results show that verifiable multi-hop reasoning does not require massive models at inference time, but rather the right architectural inductive biases combining symbolic structure with learned representations.
Authors: Mobina Mehrazar, Mohammad Amin Yousefi, Parisa Abolfath Beygi, Behnam Bahrak
Abstract: Large language models (LLMs) are increasingly used to generate self-explanations alongside their predictions, a practice that raises concerns about the faithfulness of these explanations, especially in low-resource languages. This study evaluates the faithfulness of LLM-generated explanations in the context of emotion classification in Persian, a low-resource language, by comparing the influential words identified by the model against those identified by human annotators. We assess faithfulness using confidence scores derived from token-level log-probabilities. Two prompting strategies, differing in the order of explanation and prediction (Predict-then-Explain and Explain-then-Predict), are tested for their impact on explanation faithfulness. Our results reveal that while LLMs achieve strong classification performance, their generated explanations often diverge from faithful reasoning, showing greater agreement with each other than with human judgments. These results highlight the limitations of current explanation methods and metrics, emphasizing the need for more robust approaches to ensure LLM reliability in multilingual and low-resource contexts.
Authors: Richard J. Young, Alice M. Matthews
Abstract: Domain-specific text embeddings are critical for clinical natural language processing, yet systematic comparisons across model architectures remain limited. This study evaluates ten transformer-based embedding models adapted for cardiology through Low-Rank Adaptation (LoRA) fine-tuning on 106,535 cardiology text pairs derived from authoritative medical textbooks. Results demonstrate that encoder-only architectures, particularly BioLinkBERT, achieve superior domain-specific performance (separation score: 0.510) compared to larger decoder-based models, while requiring significantly fewer computational resources. The findings challenge the assumption that larger language models necessarily produce better domain-specific embeddings and provide practical guidance for clinical NLP system development. All models, training code, and evaluation datasets are publicly available to support reproducible research in medical informatics.
Authors: Colton Casto, Anna Ivanova, Evelina Fedorenko, Nancy Kanwisher
Abstract: Language understanding entails not just extracting the surface-level meaning of the linguistic input, but constructing rich mental models of the situation it describes. Here we propose that because processing within the brain's core language system is fundamentally limited, deeply understanding language requires exporting information from the language system to other brain regions that compute perceptual and motor representations, construct mental models, and store our world knowledge and autobiographical memories. We review the existing evidence for this hypothesis, and argue that recent progress in cognitive neuroscience provides both the conceptual foundation and the methods to directly test it, thus opening up a new strategy to reveal what it means, cognitively and neurally, to understand language.
Authors: Maureen Herbert, Katie Sun, Angelica Lim, Yasaman Etesam
Abstract: The rapid advancement of large language models (LLMs) and their growing integration into daily life underscore the importance of evaluating and ensuring their fairness. In this work, we examine fairness within the domain of emotional theory of mind, investigating whether LLMs exhibit gender biases when presented with a description of a person and their environment and asked, "How does this person feel?". Furthermore, we propose and evaluate several debiasing strategies, demonstrating that achieving meaningful reductions in bias requires training based interventions rather than relying solely on inference-time prompt-based approaches such as prompt engineering.
Authors: Saif M. Mohammad
Abstract: Factor analysis studies have shown that the primary dimensions of word meaning are Valence (V), Arousal (A), and Dominance (D). Existing lexicons such as the NRC VAD Lexicon, published in 2018, include VAD association ratings for words. Here, we present a complement to it, which has human ratings of valence, arousal, and dominance for 10k English Multiword Expressions (MWEs) and their constituent words. We also increase the coverage of unigrams, especially words that have become more common since 2018. In all, the new NRC VAD Lexicon v2 now has entries for 10k MWEs and 25k words, in addition to the entries in v1. We show that the associations are highly reliable. We use the lexicon to examine emotional characteristics of MWEs, including: 1. The degree to which MWEs (idioms, noun compounds, and verb particle constructions) exhibit strong emotionality; 2. The degree of emotional compositionality in MWEs. The lexicon enables a wide variety of research in NLP, Psychology, Public Health, Digital Humanities, and Social Sciences. The NRC VAD Lexicon v2 is freely available through the project webpage: http://saifmohammad.com/WebPages/nrc-vad.html
Authors: Koena Ronny Mabokela, Tim Schlippe, Mpho Raborife, Turgay Celik
Abstract: Sentiment analysis is a helpful task to automatically analyse opinions and emotions on various topics in areas such as AI for Social Good, AI in Education or marketing. While many of the sentiment analysis systems are developed for English, many African languages are classified as low-resource languages due to the lack of digital language resources like text labelled with corresponding sentiment classes. One reason for that is that manually labelling text data is time-consuming and expensive. Consequently, automatic and rapid processes are needed to reduce the manual effort as much as possible making the labelling process as efficient as possible. In this paper, we present and analyze an automatic language-independent sentiment labelling method that leverages information from sentiment-bearing emojis and words. Our experiments are conducted with tweets in the languages English, Sepedi and Setswana from SAfriSenti, a multilingual sentiment corpus for South African languages. We show that our sentiment labelling approach is able to label the English tweets with an accuracy of 66%, the Sepedi tweets with 69%, and the Setswana tweets with 63%, so that on average only 34% of the automatically generated labels remain to be corrected.
Authors: Shi-Wei Dai, Yan-Wei Shie, Tsung-Huan Yang, Lun-Wei Ku, Yung-Hui Li
Abstract: Personalized Large Language Models (LLMs) have been shown to be an effective way to create more engaging and enjoyable user-AI interactions. While previous studies have explored using prompts to elicit specific personality traits in LLMs, they have not optimized these prompts to maximize personality expression. To address this limitation, we propose PersonaPulse: Dynamic Profile Optimization for Realistic Personality Expression in LLMs, a framework that leverages LLMs' inherent knowledge of personality traits to iteratively enhance role-play prompts while integrating a situational response benchmark as a scoring tool, ensuring a more realistic and contextually grounded evaluation to guide the optimization process. Quantitative evaluations demonstrate that the prompts generated by PersonaPulse outperform those of prior work, which were designed based on personality descriptions from psychological studies. Additionally, we explore the relationship between model size and personality modeling through extensive experiments. Finally, we find that, for certain personality traits, the extent of personality evocation can be partially controlled by pausing the optimization process. These findings underscore the importance of prompt optimization in shaping personality expression within LLMs, offering valuable insights for future research on adaptive AI interactions.
Authors: Farzad Ahmed, Joniel Augustine Jerome, Meliha Yetisgen, \"Ozlem Uzuner
Abstract: Objective: Clinical documentation contains factual, diagnostic, and management errors that can compromise patient safety. Large language models (LLMs) may help detect and correct such errors, but their behavior under different prompting strategies remains unclear. We evaluate zero-shot prompting, static prompting with random exemplars (SPR), and retrieval-augmented dynamic prompting (RDP) for three subtasks of medical error processing: error flag detection, error sentence detection, and error correction. Methods: Using the MEDEC dataset, we evaluated nine instruction-tuned LLMs (GPT, Claude, Gemini, and OpenAI o-series models). We measured performance using accuracy, recall, false-positive rate (FPR), and an aggregate score of ROUGE-1, BLEURT, and BERTScore for error correction. We also analyzed example outputs to identify failure modes and differences between LLM and clinician reasoning. Results: Zero-shot prompting showed low recall in both detection tasks, often missing abbreviation-heavy or atypical errors. SPR improved recall but increased FPR. Across all nine LLMs, RDP reduced FPR by about 15 percent, improved recall by 5 to 10 percent in error sentence detection, and generated more contextually accurate corrections. Conclusion: Across diverse LLMs, RDP outperforms zero-shot and SPR prompting. Using retrieved exemplars improves detection accuracy, reduces false positives, and enhances the reliability of medical error correction.
Authors: Tianyi Chen, Michael Solodko, Sen Wang, Jongwoo Ko, Junheng Hao, Colby Banbury, Sara Abdali, Saeed Amizadeh, Qing Xiao, Yinheng Li, Tianyu Ding, Kamran Ghasedi Dizaji, Suzhen Zheng, Hao Fan, Justin Wagle, Pashmina Cameron, Kazuhito Koishida
Abstract: Computer Using Agents (CUAs) are increasingly equipped with external tools, enabling them to perform complex and realistic tasks. For CUAs to operate effectively, application selection, which refers to deciding which application to use before invoking fine-grained tools such as APIs, is a fundamental capability. It determines whether the agent initializes the correct environment, avoids orchestration confusion, and efficiently focuses on relevant context. However, existing benchmarks primarily assess fine-grained API selection, offering limited insight into whether models can reason across and choose between different applications. To fill this gap, we introduce AppSelectBench, a comprehensive benchmark for evaluating application selection in CUAs. AppSelectBench contains a novel user task generation pipeline that produces realistic, diverse, and semantically grounded user intents at scale, together with unified evaluation protocols covering random, heuristic, zero-shot, few-shot, and retrieval-augmented-settings. AppSelectBench covers one hundred widely used desktop applications and includes more than one hundred thousand realistic, diverse, and semantically grounded user tasks. Extensive experiments across both closed-source and open-source large language models reveal systematic strengths and weaknesses in inter-application reasoning, showing that even the most capable models still struggle to make consistent application choices. Together, these results establish AppSelectBench as a foundation for studying and advancing application level reasoning, an essential yet underexplored capability of intelligent CUAs. The source is available at https://github.com/microsoft/appselectbench.
Authors: Xinyu Wang, Hanwei Wu, Qingchen Hu, Zhenghan Tai, Jingrui Tian, Lei Ding, Jijun Chi, Hailin He, Tung Sum Thomas Kwok, Yufei Cui, Sicheng Lyu, Muzhi Li, Mingze Li, Xinyue Yu, Ling Zhou, Peng Lu
Abstract: Decoder-only rerankers are central to Retrieval-Augmented Generation (RAG). However, generalist models miss domain-specific nuances in high-stakes fields like finance and law, and naive fine-tuning causes surface-form overfitting and catastrophic forgetting. To address this challenge, we introduce R2R, a domain-aware framework that combines dynamic expert routing with a two-stage training strategy, Entity Abstraction for Generalization (EAG). EAG introduces a counter-shortcut mechanism by masking the most predictive surface cues, forcing the reranker to learn domain-invariant relevance patterns rather than memorizing dataset-specific entities. To efficiently activate domain experts, R2R employs a lightweight Latent Semantic Router that probes internal representations from the frozen backbone decoder to select the optimal LoRA expert per query. Extensive experiments across different reranker backbones and diverse domains (legal, medical, and financial) demonstrate that R2R consistently surpasses generalist and single-domain fine-tuned baselines. Our results confirm that R2R is a model-agnostic and modular approach to domain specialization with strong cross-domain robustness.
Authors: Mihir Sahasrabudhe
Abstract: Transformers are theoretically reversal-invariant: their function class does not prefer left-to-right over right-to-left mappings. Yet empirical studies on natural language repeatedly report a "reversal curse," and recent work on temporal asymmetry in LLMs suggests that real-world corpora carry their own arrow of time. This leaves an unresolved question: do directional failures stem from linguistic statistics, or from the architecture itself? We cut through this ambiguity with a fully synthetic, entropy-controlled benchmark designed as a clean-room stress test for directional learning. Using random string mappings with tunable branching factor K, we construct forward tasks with zero conditional entropy and inverse tasks with analytically determined entropy floors. Excess loss above these floors reveals that even scratch-trained GPT-2 models exhibit a strong, reproducible directional optimization gap (e.g., 1.16 nats at K=5), far larger than that of an MLP trained on the same data. Pre-trained initializations shift optimization behavior but do not eliminate this gap, while LoRA encounters a sharp capacity wall on high-entropy inverse mappings. Together, these results isolate a minimal, semantics-free signature of directional friction intrinsic to causal Transformer training-one that persists even when linguistic priors, token frequencies, and corpus-level temporal asymmetries are removed. Our benchmark provides a controlled instrument for dissecting directional biases in modern sequence models and motivates deeper mechanistic study of why inversion remains fundamentally harder for Transformers.
Authors: Edward Ajayi, Martha Kachweka, Mawuli Deku, Emily Aiken
Abstract: Mental health challenges and cyberbullying are increasingly prevalent in digital spaces, necessitating scalable and interpretable detection systems. This paper introduces a unified multiclass classification framework for detecting ten distinct mental health and cyberbullying categories from social media data. We curate datasets from Twitter and Reddit, implementing a rigorous "split-then-balance" pipeline to train on balanced data while evaluating on a realistic, held-out imbalanced test set. We conducted a comprehensive evaluation comparing traditional lexical models, hybrid approaches, and several end-to-end fine-tuned transformers. Our results demonstrate that end-to-end fine-tuning is critical for performance, with the domain-adapted MentalBERT emerging as the top model, achieving an accuracy of 0.92 and a Macro F1 score of 0.76, surpassing both its generic counterpart and a zero-shot LLM baseline. Grounded in a comprehensive ethical analysis, we frame the system as a human-in-the-loop screening aid, not a diagnostic tool. To support this, we introduce a hybrid SHAPLLM explainability framework and present a prototype dashboard ("Social Media Screener") designed to integrate model predictions and their explanations into a practical workflow for moderators. Our work provides a robust baseline, highlighting future needs for multi-label, clinically-validated datasets at the critical intersection of online safety and computational mental health.
Authors: Huiyu Bai, Runze Wang, Zhuoyun Du, Yiyang Zhao, Fengji Zhang, Haoyu Chen, Xiaoyong Zhu, Bo Zheng, Xuejiao Zhao
Abstract: Personalized Visual Language Models (VLMs) are gaining increasing attention for their formidable ability in user-specific concepts aligned interactions (e.g., identifying a user's bike). Existing methods typically require the learning of separate embeddings for each new concept, which fails to support real-time adaptation during testing. This limitation becomes particularly pronounced in large-scale scenarios, where efficient retrieval of concept embeddings is not achievable. To alleviate this gap, we propose Online-PVLM, a framework for online concept learning by leveraging hyperbolic representations. Our approach makes a train-free paradigm for concept embeddings generation at test time, making the use of personalized VLMs both scalable and efficient. In addition, we develop OP-Eval, a comprehensive and large-scale benchmark comprising 1,292 concepts and over 30K high-quality instances with diverse question types, designed to rigorously assess online concept learning in realistic scenarios. Extensive experiments demonstrate the state-of-the-art performance of our proposed framework. Our source code and dataset will be made available.
Authors: Xiaopeng Li, Yuanjin Zheng, Wanyu Wang, wenlin zhang, Pengyue Jia, Yiqi Wang, Maolin Wang, Xuetao Wei, Xiangyu Zhao
Abstract: Personalized Large Language Models (PLLMs) aim to align model outputs with individual user preferences, a crucial capability for user-centric applications. However, the prevalent approach of fine-tuning a separate module for each user faces two major limitations: (1) storage costs scale linearly with the number of users, rendering the method unscalable; and (2) fine-tuning a static model from scratch often yields suboptimal performance for users with sparse data. To address these challenges, we propose MTA, a Merge-then-Adapt framework for PLLMs. MTA comprises three key stages. First, we construct a shared Meta-LoRA Bank by selecting anchor users and pre-training meta-personalization traits within meta-LoRA modules. Second, to ensure scalability and enable dynamic personalization combination beyond static models, we introduce an Adaptive LoRA Fusion stage. This stage retrieves and dynamically merges the most relevant anchor meta-LoRAs to synthesize a user-specific one, thereby eliminating the need for user-specific storage and supporting more flexible personalization. Third, we propose a LoRA Stacking for Few-Shot Personalization stage, which applies an additional ultra-low-rank, lightweight LoRA module on top of the merged LoRA. Fine-tuning this module enables effective personalization under few-shot settings. Extensive experiments on the LaMP benchmark demonstrate that our approach outperforms existing SOTA methods across multiple tasks.
Authors: Duc Anh Vu, Thong Nguyen, Cong-Duy Nguyen, Viet Anh Nguyen, Anh Tuan Luu
Abstract: With the advancement of large language models (LLMs), their performance on multiple-choice question (MCQ) tasks has improved significantly. However, existing approaches face key limitations: answer choices are typically presented to LLMs without contextual grounding or explanation. This absence of context can lead to incomplete exploration of all possible answers, ultimately degrading the models' reasoning capabilities. To address these challenges, we introduce BiasPrompting, a novel inference framework that guides LLMs to generate and critically evaluate reasoning across all plausible answer options before reaching a final prediction. It consists of two components: first, a reasoning generation stage, where the model is prompted to produce supportive reasonings for each answer option, and then, a reasoning-guided agreement stage, where the generated reasonings are synthesized to select the most plausible answer. Through comprehensive evaluations, BiasPrompting demonstrates significant improvements in five widely used multiple-choice question answering benchmarks. Our experiments showcase that BiasPrompting enhances the reasoning capabilities of LLMs and provides a strong foundation for tackling complex and challenging questions, particularly in settings where existing methods underperform.
Authors: Zhenyi Shen, Junru Lu, Lin Gui, Jiazheng Li, Yulan He, Di Yin, Xing Sun
Abstract: The quadratic complexity of full attention limits efficient long-context processing in large language models (LLMs). Sparse attention mitigates this cost by restricting each query to attend to a subset of previous tokens; however, training-free approaches often lead to severe performance degradation. Native sparse-attention methods (e.g., NSA, MoBA) alleviate this issue, yet exhibit a critical paradox: they produce lower attention sparsity than full-attention models, despite aiming to approximate full attention, which may constrain their effectiveness. We attribute this paradox to gradient update deficiency: low-ranked key-value pairs excluded during sparse training receive neither forward contribution nor backward gradients, and thus never learn proper suppression. To overcome this limitation, we propose SSA (Sparse Sparse Attention), a unified training framework that considers both sparse and full attention and enforces bidirectional alignment at every layer. This design preserves gradient flow to all tokens while explicitly encouraging sparse-attention outputs to align with their full-attention counterparts, thereby promoting stronger sparsity. As a result, SSA achieves state-of-the-art performance under both sparse and full attention inference across multiple commonsense benchmarks. Furthermore, SSA enables models to adapt smoothly to varying sparsity budgets; performance improves consistently as more tokens are allowed to attend, supporting flexible compute-performance trade-offs at inference time. Finally, we show that native sparse-attention training surprisingly improves long-context extrapolation by mitigating the over-allocation of attention values in sink areas, with SSA demonstrating the strongest extrapolation capability.
Authors: Xingfeng Li, Xiaohan Shi, Junjie Li, Yongwei Li, Masashi Unoki, Tomoki Toda, Masato Akagi
Abstract: This study introduces EM2LDL, a novel multilingual speech corpus designed to advance mixed emotion recognition through label distribution learning. Addressing the limitations of predominantly monolingual and single-label emotion corpora \textcolor{black}{that restrict linguistic diversity, are unable to model mixed emotions, and lack ecological validity}, EM2LDL comprises expressive utterances in English, Mandarin, and Cantonese, capturing the intra-utterance code-switching prevalent in multilingual regions like Hong Kong and Macao. The corpus integrates spontaneous emotional expressions from online platforms, annotated with fine-grained emotion distributions across 32 categories. Experimental baselines using self-supervised learning models demonstrate robust performance in speaker-independent gender-, age-, and personality-based evaluations, with HuBERT-large-EN achieving optimal results. By incorporating linguistic diversity and ecological validity, EM2LDL enables the exploration of complex emotional dynamics in multilingual settings. This work provides a versatile testbed for developing adaptive, empathetic systems for applications in affective computing, including mental health monitoring and cross-cultural communication. The dataset, annotations, and baseline codes are publicly available at https://github.com/xingfengli/EM2LDL.
Authors: Huu Tuong Tu, Ha Viet Khanh, Tran Tien Dat, Vu Huan, Thien Van Luong, Nguyen Tien Cuong, Nguyen Thi Thu Trang
Abstract: Mispronunciation Detection and Diagnosis (MDD) is crucial for language learning and speech therapy. Unlike conventional methods that require scoring models or training phoneme-level models, we propose a novel training-free framework that leverages retrieval techniques with a pretrained Automatic Speech Recognition model. Our method avoids phoneme-specific modeling or additional task-specific training, while still achieving accurate detection and diagnosis of pronunciation errors. Experiments on the L2-ARCTIC dataset show that our method achieves a superior F1 score of 69.60% while avoiding the complexity of model training.
Authors: Somsubhra De, Harsh Kumar, Arun Prakash A
Abstract: Grammatical error correction (GEC) is an important task in Natural Language Processing that aims to automatically detect and correct grammatical mistakes in text. While recent advances in transformer-based models and large annotated datasets have greatly improved GEC performance for high-resource languages such as English, the progress has not extended equally. For most Indic languages, GEC remains a challenging task due to limited resources, linguistic diversity and complex morphology. In this work, we explore prompting-based approaches using state-of-the-art large language models (LLMs), such as GPT-4.1, Gemini-2.5 and LLaMA-4, combined with few-shot strategy to adapt them to low-resource settings. We observe that even basic prompting strategies, such as zero-shot and few-shot approaches, enable these LLMs to substantially outperform fine-tuned Indic-language models like Sarvam-22B, thereby illustrating the exceptional multilingual generalization capabilities of contemporary LLMs for GEC. Our experiments show that carefully designed prompts and lightweight adaptation significantly enhance correction quality across multiple Indic languages. We achieved leading results in the shared task--ranking 1st in Tamil (GLEU: 91.57) and Hindi (GLEU: 85.69), 2nd in Telugu (GLEU: 85.22), 4th in Bangla (GLEU: 92.86), and 5th in Malayalam (GLEU: 92.97). These findings highlight the effectiveness of prompt-driven NLP techniques and underscore the potential of large-scale LLMs to bridge resource gaps in multilingual GEC.
Authors: Wen-Fang Su, Hsiao-Wei Chou, Wen-Yang Lin
Abstract: Named Entity Recognition (NER) is a critical task in natural language processing, yet it remains particularly challenging for discontinuous entities. The primary difficulty lies in text segmentation, as traditional methods often missegment or entirely miss cross-sentence discontinuous entities, significantly affecting recognition accuracy. Therefore, we aim to address the segmentation and omission issues associated with such entities. Recent studies have shown that grid-tagging methods are effective for information extraction due to their flexible tagging schemes and robust architectures. Building on this, we integrate image data augmentation techniques, such as cropping, scaling, and padding, into grid-based models to enhance their ability to recognize discontinuous entities and handle segmentation challenges. Experimental results demonstrate that traditional segmentation methods often fail to capture cross-sentence discontinuous entities, leading to decreased performance. In contrast, our augmented grid models achieve notable improvements. Evaluations on the CADEC, ShARe13, and ShARe14 datasets show F1 score gains of 1-2.5% overall and 3.7-8.4% for discontinuous entities, confirming the effectiveness of our approach.
Authors: Adilet Metinov, Gulida M. Kudakeeva, Gulnara D. Kabaeva
Abstract: Kyrgyz remains a low-resource language with limited foundational NLP tools. To address this gap, we introduce KyrgyzBERT, the first publicly available monolingual BERT-based language model for Kyrgyz. The model has 35.9M parameters and uses a custom tokenizer designed for the language's morphological structure. To evaluate performance, we create kyrgyz-sst2, a sentiment analysis benchmark built by translating the Stanford Sentiment Treebank and manually annotating the full test set. KyrgyzBERT fine-tuned on this dataset achieves an F1-score of 0.8280, competitive with a fine-tuned mBERT model five times larger. All models, data, and code are released to support future research in Kyrgyz NLP.
Authors: Chuyi Kong, Gao Wei, Jing Ma, Hongzhan Lin, Zhiyuan Fan
Abstract: The prevalence of misinformation on social media threatens public trust, demanding automated fact-checking systems that provide accurate verdicts with interpretable explanations. However, existing large language model-based (LLM-based) approaches often rely heavily on external knowledge sources, introducing substantial latency and even hallucinations that undermine reliability, interpretability, and responsiveness, which is crucial for real-time use. To address these challenges, we propose REason-guided Fact-checking with Latent EXplanations REFLEX paradigm, a plug-and-play, self-refining paradigm that leverages the internal knowledge in backbone model to improve both verdict accuracy and explanation quality. REFLEX reformulates fact-checking as a role-play dialogue and jointly trains verdict prediction and explanation generation. It adaptively extracts contrastive activation pairs between the backbone model and its fine-tuned variant to construct steering vectors that disentangle truth into style and substance naturally. These activation-level signals guide inference and suppress noisy explanations, enabling more faithful and efficient reasoning. Experiments on real-world datasets show that REFLEX outperforms previous methods that steer toward a single truth direction and underscores the challenge traditional approaches face when handling the subtle, human-unknown truth in fact-checking tasks. Remarkably, with only 465 self-refined training samples, RELFEX achieves state-of-the-art performance. Furthermore, models trained with explanatory objectives can effectively guide those without them, yielding up to a 7.57% improvement, highlighting that internal explanation signals play a dual role in both interpreting and enhancing factual reasoning.
Authors: Luohe Shi, Zuchao Li, Lefei Zhang, Baoyuan Qi, Guoming Liu, Hai Zhao
Abstract: Speculative decoding accelerates LLM inference by utilizing otherwise idle computational resources during memory-to-chip data transfer. Current speculative decoding methods typically assume a considerable amount of available computing power, then generate a complex and massive draft tree using a small autoregressive language model to improve overall prediction accuracy. However, methods like batching have been widely applied in mainstream model inference systems as a superior alternative to speculative decoding, as they compress the available idle computing power. Therefore, performing speculative decoding with low verification resources and low scheduling costs has become an important research problem. We believe that more capable models that allow for parallel generation on draft sequences are what we truly need. Recognizing the fundamental nature of draft models to only generate sequences of limited length, we propose SpecFormer, a novel architecture that integrates unidirectional and bidirectional attention mechanisms. SpecFormer combines the autoregressive model's ability to extract information from the entire input sequence with the parallel generation benefits of non-autoregressive models. This design eliminates the reliance on large prefix trees and achieves consistent acceleration, even in large-batch scenarios. Through lossless speculative decoding experiments across models of various scales, we demonstrate that SpecFormer sets a new standard for scaling LLM inference with lower training demands and reduced computational costs.
Authors: Taewhoo Lee, Minju Song, Chanwoong Yoon, Jungwoo Park, Jaewoo Kang
Abstract: Analogical reasoning is at the core of human cognition, serving as an important foundation for a variety of intellectual activities. While prior work has shown that LLMs can represent task patterns and surface-level concepts, it remains unclear whether these models can encode high-level relational concepts and apply them to novel situations through structured comparisons. In this work, we explore this fundamental aspect using proportional and story analogies, and identify three key findings. First, LLMs effectively encode the underlying relationships between analogous entities; both attributive and relational information propagate through mid-upper layers in correct cases, whereas reasoning failures reflect missing relational information within these layers. Second, unlike humans, LLMs often struggle not only when relational information is missing, but also when attempting to apply it to new entities. In such cases, strategically patching hidden representations at critical token positions can facilitate information transfer to a certain extent. Lastly, successful analogical reasoning in LLMs is marked by strong structural alignment between analogous situations, whereas failures often reflect degraded or misplaced alignment. Overall, our findings reveal that LLMs exhibit emerging but limited capabilities in encoding and applying high-level relational concepts, highlighting both parallels and gaps with human cognition.
Authors: Abdullah Al Sefat
Abstract: Large language models excel on broad multilingual benchmarks but remain to be evaluated extensively in figurative and culturally grounded reasoning, especially in low-resource contexts. We present BengaliFig, a compact yet richly annotated challenge set that targets this gap in Bengali, a widely spoken low-resourced language. The dataset contains 435 unique riddles drawn from Bengali oral and literary traditions. Each item is annotated along five orthogonal dimensions capturing reasoning type, trap type, cultural depth, answer category, and difficulty, and is automatically converted to multiple-choice format through a constraint-aware, AI-assisted pipeline. We evaluate eight frontier LLMs from major providers under zero-shot and few-shot chain-of-thought prompting, revealing consistent weaknesses in metaphorical and culturally specific reasoning. BengaliFig thus contributes both a diagnostic probe for evaluating LLM robustness in low-resource cultural contexts and a step toward inclusive and heritage-aware NLP evaluation.
Authors: Md Abdullah Al Kafi, Raka Moni, Sumit Kumar Banshal
Abstract: Text normalization is an essential preprocessing step in many natural language processing (NLP) tasks, and stemming is one such normalization technique that reduces words to their base or root form. However, evaluating stemming methods is challenging because current evaluation approaches are limited and do not capture the potential harm caused by excessive stemming; therefore, it is essential to develop new approaches to evaluate stemming methods. To address this issue, this study propose a novel, task-oriented approach to evaluate stemming methods, which considers three aspects: (1) the utility of stemming using Stemming Effectiveness Score (SES), (2) the impact of stemming on downstream tasks using Model Performance Delta (MPD), and (3) the semantic similarity between stemmed and original words using Average Normalized Levenshtein Distance (ANLD), thus providing a comprehensive evaluation framework. We apply our evaluation framework to compare two stemmers for Bangla (BNLTK) and English (Snowball), and our results reveal a significant issue, prompting us to analyze their performance in detail. While the Bangla stemmer achieves the highest SES (1.67) due to effective word reduction (CR = 1.90), SES alone is insufficient because our proposed safety measure, ANLD, reveals that this high SES is due to harmful over-stemming (ANLD = 0.26), which correlates with the observed decrease in downstream performance.In contrast, the English stemmer achieves a moderate SES (1.31) with a safe meaning distance (ANLD = 0.14), allowing its word reduction to contribute positively to downstream performance; therefore, it is a more reliable stemmer. Our study provides a valuable tool for distinguishing between potential efficiency gains (high SES) and meaning preservation (low ANLD).
Authors: Mosab Rezaei, Mina Rajaei Moghadam, Abdul Rahman Shaikh, Hamed Alhoori, Reva Freedman
Abstract: Recent advances in large language models have created new opportunities for stylometry, the study of writing styles and authorship. Two challenges, however, remain central: training generative models when no paired data exist, and evaluating stylistic text without relying only on human judgment. In this work, we present a framework for both generating and evaluating sentences in the style of 19th-century novelists. Large language models are fine-tuned with minimal, single-token prompts to produce text in the voices of authors such as Dickens, Austen, Twain, Alcott, and Melville. To assess these generative models, we employ a transformer-based detector trained on authentic sentences, using it both as a classifier and as a tool for stylistic explanation. We complement this with syntactic comparisons and explainable AI methods, including attention-based and gradient-based analyses, to identify the linguistic cues that drive stylistic imitation. Our findings show that the generated text reflects the authors' distinctive patterns and that AI-based evaluation offers a reliable alternative to human assessment. All artifacts of this work are published online.
Authors: Jakub Hoscilowicz, Artur Janicki
Abstract: We introduce the Adversarial Confusion Attack, a new class of threats against multimodal large language models (MLLMs). Unlike jailbreaks or targeted misclassification, the goal is to induce systematic disruption that makes the model generate incoherent or confidently incorrect outputs. Applications include embedding adversarial images into websites to prevent MLLM-powered agents from operating reliably. The proposed attack maximizes next-token entropy using a small ensemble of open-source MLLMs. In the white-box setting, we show that a single adversarial image can disrupt all models in the ensemble, both in the full-image and adversarial CAPTCHA settings. Despite relying on a basic adversarial technique (PGD), the attack generates perturbations that transfer to both unseen open-source (e.g., Qwen3-VL) and proprietary (e.g., GPT-5.1) models.
Authors: Nathan Roll, Jill Kries, Flora Jin, Catherine Wang, Ann Marie Finley, Meghan Sumner, Cory Shain, Laura Gwilliams
Abstract: Large language models (LLMs) have emerged as a candidate "model organism" for human language, offering an unprecedented opportunity to study the computational basis of linguistic disorders like aphasia. However, traditional clinical assessments are ill-suited for LLMs, as they presuppose human-like pragmatic pressures and probe cognitive processes not inherent to artificial architectures. We introduce the Text Aphasia Battery (TAB), a text-only benchmark adapted from the Quick Aphasia Battery (QAB) to assess aphasic-like deficits in LLMs. The TAB comprises four subtests: Connected Text, Word Comprehension, Sentence Comprehension, and Repetition. This paper details the TAB's design, subtests, and scoring criteria. To facilitate large-scale use, we validate an automated evaluation protocol using Gemini 2.5 Flash, which achieves reliability comparable to expert human raters (prevalence-weighted Cohen's kappa = 0.255 for model--consensus agreement vs. 0.286 for human--human agreement). We release TAB as a clinically-grounded, scalable framework for analyzing language deficits in artificial systems.
Authors: Wesley Bian, Xiaofeng Lin, Guang Cheng
Abstract: Modern machine learning models for audio tasks often exhibit superior performance on English and other well-resourced languages, primarily due to the abundance of available training data. This disparity leads to an unfair performance gap for low-resource languages, where data collection is both challenging and costly. In this work, we introduce a novel data augmentation technique for speech corpora designed to mitigate this gap. Through comprehensive experiments, we demonstrate that our method significantly improves the performance of automatic speech recognition systems on low-resource languages. Furthermore, we show that our approach outperforms existing augmentation strategies, offering a practical solution for enhancing speech technology in underrepresented linguistic communities.
Authors: Farjana Sultana Mim, Shuchin Aeron, Eric Miller, Kristen Wendell
Abstract: Identifying discourse features in student conversations is quite important for educational researchers to recognize the curricular and pedagogical variables that cause students to engage in constructing knowledge rather than merely completing tasks. The manual analysis of student conversations to identify these discourse features is time-consuming and labor-intensive, which limits the scale and scope of studies. Leveraging natural language processing (NLP) techniques can facilitate the automatic detection of these discourse features, offering educational researchers scalable and data-driven insights. However, existing studies in NLP that focus on discourse in dialogue rarely address educational data. In this work, we address this gap by introducing an annotated educational dialogue dataset of student conversations featuring knowledge construction and task production discourse. We also establish baseline models for automatically predicting these discourse properties for each turn of talk within conversations, using pre-trained large language models GPT-3.5 and Llama-3.1. Experimental results indicate that these state-of-the-art models perform suboptimally on this task, indicating the potential for future research.
Authors: Yixin Liu, Pengfei Liu, Arman Cohan
Abstract: Alignment with human preferences is an important evaluation aspect of LLMs, requiring them to be helpful, honest, safe, and to precisely follow human instructions. Evaluating large language models' (LLMs) alignment typically involves directly assessing their open-ended responses, requiring human annotators or strong LLM judges. Conversely, LLMs themselves have also been extensively evaluated as judges for assessing alignment. In this work, we examine the relationship between LLMs' generation and evaluation capabilities in aligning with human preferences. To this end, we first conduct a comprehensive analysis of the generation-evaluation consistency (GE-consistency) among various LLMs, revealing a strong correlation between their generation and evaluation capabilities when evaluated by a strong LLM preference oracle. Utilizing this finding, we propose a benchmarking paradigm that measures LLM alignment with human preferences without directly evaluating their generated outputs, instead assessing LLMs in their role as evaluators. Our evaluation shows that our proposed benchmark, AlignEval, matches or surpasses widely used automatic LLM evaluation benchmarks, such as AlpacaEval and Arena-Hard, in capturing human preferences when ranking LLMs. Our study offers valuable insights into the connection between LLMs' generation and evaluation capabilities, and introduces a benchmark that assesses alignment without directly evaluating model outputs.
Authors: Jiaru Zou, Xiyuan Yang, Ruizhong Qiu, Gaotang Li, Katherine Tieu, Pan Lu, Ke Shen, Hanghang Tong, Yejin Choi, Jingrui He, James Zou, Mengdi Wang, Ling Yang
Abstract: Multi-agent systems (MAS) extend large language models (LLMs) from independent single-model reasoning to coordinative system-level intelligence. While existing LLM agents depend on text-based mediation for reasoning and communication, we take a step forward by enabling models to collaborate directly within the continuous latent space. We introduce LatentMAS, an end-to-end training-free framework that enables pure latent collaboration among LLM agents. In LatentMAS, each agent first performs auto-regressive latent thoughts generation through last-layer hidden embeddings. A shared latent working memory then preserves and transfers each agent's internal representations, ensuring lossless information exchange. We provide theoretical analyses establishing that LatentMAS attains higher expressiveness and lossless information preservation with substantially lower complexity than vanilla text-based MAS. In addition, empirical evaluations across 9 comprehensive benchmarks spanning math and science reasoning, commonsense understanding, and code generation show that LatentMAS consistently outperforms strong single-model and text-based MAS baselines, achieving up to 14.6% higher accuracy, reducing output token usage by 70.8%-83.7%, and providing 4x-4.3x faster end-to-end inference. These results demonstrate that our new latent collaboration framework enhances system-level reasoning quality while offering substantial efficiency gains without any additional training. Code and data are fully open-sourced at https://github.com/Gen-Verse/LatentMAS.
Authors: Sandro Andric
Abstract: Mechanistic interpretability aspires to reverse-engineer neural networks into explicit algorithms, while model editing seeks to modify specific behaviours without retraining. Both areas are typically evaluated with informal evidence and ad-hoc experiments, with few explicit guarantees about how far an extracted or edited model can drift from the original on relevant inputs. We introduce BlockCert, a framework for certified blockwise extraction of transformer mechanisms, and outline how a lightweight extension can support certified local edits. Given a pre-trained transformer and a prompt distribution, BlockCert extracts structured surrogate implementations for residual blocks together with machine-checkable certificates that bound approximation error, record coverage metrics, and hash the underlying artifacts. We formalize a simple Lipschitz-based composition theorem in Lean 4 that lifts these local guarantees to a global deviation bound. Empirically, we apply the framework to GPT-2 small, TinyLlama-1.1B-Chat, and Llama-3.2-3B. Across these models we obtain high per-block coverage and small residual errors on the evaluated prompts, and in the TinyLlama setting we show that a fully stitched model matches the baseline perplexity within approximately 6e-5 on stress prompts. Our results suggest that blockwise extraction with explicit certificates is feasible for real transformer language models and offers a practical bridge between mechanistic interpretability and formal reasoning about model behaviour.
Authors: Padegal Amit, Omkar Mahesh Kashyap, Namitha Rayasam, Nidhi Shekhar, Surabhi Narayan
Abstract: Quantifying modality contributions in multimodal models remains a challenge, as existing approaches conflate the notion of contribution itself. Prior work relies on accuracy-based approaches, interpreting performance drops after removing a modality as indicative of its influence. However, such outcome-driven metrics fail to distinguish whether a modality is inherently informative or whether its value arises only through interaction with other modalities. This distinction is particularly important in cross-attention architectures, where modalities influence each other's representations. In this work, we propose a framework based on Partial Information Decomposition (PID) that quantifies modality contributions by decomposing predictive information in internal embeddings into unique, redundant, and synergistic components. To enable scalable, inference-only analysis, we develop an algorithm based on the Iterative Proportional Fitting Procedure (IPFP) that computes layer and dataset-level contributions without retraining. This provides a principled, representation-level view of multimodal behavior, offering clearer and more interpretable insights than outcome-based metrics.
Authors: Remi Petitpierre
Abstract: This thesis presents methods and datasets to investigate cartographic heritage on a large scale and from a cultural perspective. Heritage institutions worldwide have digitized more than one million maps, and automated techniques now enable large-scale recognition and extraction of map content. Yet these methods have engaged little with the history of cartography, or the view that maps are semantic-symbolic systems, and cultural objects reflecting political and epistemic expectations. This work leverages a diverse corpus of 771,561 map records and 99,715 digitized images aggregated from 38 digital catalogs. After normalization, the dataset includes 236,925 contributors and spans six centuries, from 1492 to 1948. These data make it possible to chart geographic structures and the global chronology of map publication. The spatial focus of cartography is analyzed in relation to political dynamics, evidencing links between Atlantic maritime charting, the triangular trade, and colonial expansion. Further results document the progression of national, domestic focus and the impact of military conflicts on publication volumes. The research introduces semantic segmentation techniques and object detection models for the generic recognition of land classes and cartographic signs, trained on annotated data and synthetic images. The analysis of land classes shows that maps are designed images whose framing and composition emphasize features through centering and semantic symmetries. The study of cartographic figuration encodes 63 M signs and 25 M fragments into a latent visual space, revealing figurative shifts such as the replacement of relief hachures by terrain contours and showing that signs tend to form locally consistent systems. Analyses of collaboration and diffusion highlight the role of legitimacy, larger actors, and major cities in the spread of figurative norms and semiotic cultures.
Authors: Ahmed Awadallah, Yash Lara, Raghav Magazine, Hussein Mozannar, Akshay Nambi, Yash Pandya, Aravind Rajeswaran, Corby Rosset, Alexey Taymanov, Vibhav Vineet, Spencer Whitehead, Andrew Zhao
Abstract: Progress in computer use agents (CUAs) has been constrained by the absence of large and high-quality datasets that capture how humans interact with a computer. While LLMs have thrived on abundant textual data, no comparable corpus exists for CUA trajectories. To address these gaps, we introduce FaraGen, a novel synthetic data generation system for multi-step web tasks. FaraGen can propose diverse tasks from frequently used websites, generate multiple solution attempts, and filter successful trajectories using multiple verifiers. It achieves high throughput, yield, and diversity for multi-step web tasks, producing verified trajectories at approximately $1 each. We use this data to train Fara-7B, a native CUA model that perceives the computer using only screenshots, executes actions via predicted coordinates, and is small enough to run on-device. We find that Fara-7B outperforms other CUA models of comparable size on benchmarks like WebVoyager, Online-Mind2Web, and WebTailBench -- our novel benchmark that better captures under-represented web tasks in pre-existing benchmarks. Furthermore, Fara-7B is competitive with much larger frontier models, illustrating key benefits of scalable data generation systems in advancing small efficient agentic models. We are making Fara-7B open-weight on Microsoft Foundry and HuggingFace, and we are releasing WebTailBench.
Authors: Meng Lu, Ran Xu, Yi Fang, Wenxuan Zhang, Yue Yu, Gaurav Srivastava, Yuchen Zhuang, Mohamed Elhoseiny, Charles Fleming, Carl Yang, Zhengzhong Tu, Yang Xie, Guanghua Xiao, Hanrui Wang, Di Jin, Wenqi Shi, Xuan Wang
Abstract: While recent vision-language models (VLMs) demonstrate strong image understanding, their ability to "think with images", i.e., to reason through multi-step visual interactions, remains limited. We introduce VISTA-Gym, a scalable training environment for incentivizing tool-integrated visual reasoning capabilities in VLMs. VISTA-Gym unifies diverse real-world multimodal reasoning tasks (7 tasks from 13 datasets in total) with a standardized interface for visual tools (e.g., grounding, parsing), executable interaction loops, verifiable feedback signals, and efficient trajectory logging, enabling visual agentic reinforcement learning at scale. While recent VLMs exhibit strong text-only reasoning, both proprietary and open-source models still struggle with tool selection, invocation, and coordination. With VISTA-Gym, we train VISTA-R1 to interleave tool-use with agentic reasoning via multi-turn trajectory sampling and end-to-end reinforcement learning. Extensive experiments across 11 public reasoning-intensive VQA benchmarks show that VISTA-R1-8B outperforms state-of-the-art baselines with similar sizes by 9.51%-18.72%, demonstrating VISTA-Gym as an effective training ground to unlock the tool-integrated reasoning capabilities for VLMs.
Authors: Debin Meng, Chen Jin, Zheng Gao, Yanran Li, Ioannis Patras, Georgios Tzimiropoulos
Abstract: Image diversity remains a fundamental challenge for text-to-image diffusion models. Low-diversity models tend to generate repetitive outputs, increasing sampling redundancy and hindering both creative exploration and downstream applications. A primary cause is that generation often collapses toward a strong mode in the learned distribution. Existing attempts to improve diversity, such as noise resampling, prompt rewriting, or steering-based guidance, often still collapse to dominant modes or introduce distortions that degrade image quality. In light of this, we propose Token-Prompt embedding Space Optimization (TPSO), a training-free and model-agnostic module. TPSO introduces learnable parameters to explore underrepresented regions of the token embedding space, reducing the tendency of the model to repeatedly generate samples from strong modes of the learned distribution. At the same time, the prompt-level space provides a global semantic constraint that regulates distribution shifts, preventing quality degradation while maintaining high fidelity. Extensive experiments on MS-COCO and three diffusion backbones show that TPSO significantly enhances generative diversity, improving baseline performance from 1.10 to 4.18 points, without sacrificing image quality. Code will be released upon acceptance.
Authors: Miguel Carvalho, Helder Dias, Bruno Martins
Abstract: Vision-Language Models (VLMs) often struggle with tasks that require fine-grained image understanding, such as scene-text recognition or document analysis, due to perception limitations and visual fragmentation. To address these challenges, we introduce CropVLM as an external low-cost method for boosting performance, enabling VLMs to dynamically ''zoom in'' on relevant image regions, enhancing their ability to capture fine details. CropVLM is trained using reinforcement learning, without using human-labeled bounding boxes as a supervision signal, and without expensive synthetic evaluations. The model is trained once and can be paired with both open-source and proprietary VLMs to improve their performance. Our approach delivers significant improvements on tasks that require high-resolution image understanding, notably for benchmarks that are out-of-domain for the target VLM, without modifying or fine-tuning the VLM, thus avoiding catastrophic forgetting.
Authors: Chengyue Huang, Mellon M. Zhang, Robert Azarcon, Glen Chou, Zsolt Kira
Abstract: Vision-Language-Action (VLA) models inherit strong priors from pretrained Vision-Language Models (VLMs), but naive fine-tuning often disrupts these representations and harms generalization. Existing fixes -- freezing modules or applying uniform regularization -- either overconstrain adaptation or ignore the differing roles of VLA components. We present MAPS (Module-Wise Proximity Scheduling), the first robust fine-tuning framework for VLAs. Through systematic analysis, we uncover an empirical order in which proximity constraints should be relaxed to balance stability and flexibility. MAPS linearly schedules this relaxation, enabling visual encoders to stay close to their pretrained priors while action-oriented language layers adapt more freely. MAPS introduces no additional parameters or data, and can be seamlessly integrated into existing VLAs. Across MiniVLA-VQ, MiniVLA-OFT, OpenVLA-OFT, and challenging benchmarks such as SimplerEnv, CALVIN, LIBERO, as well as real-world evaluations on the Franka Emika Panda platform, MAPS consistently boosts both in-distribution and out-of-distribution performance (up to +30%). Our findings highlight empirically guided proximity to pretrained VLMs as a simple yet powerful principle for preserving broad generalization in VLM-to-VLA transfer.
Authors: Yuefei Chen, Jiang Liu, Xiaodong Lin, Ruixiang Tang
Abstract: Vision Language Models (VLMs) have recently shown significant advancements in video understanding, especially in feature alignment, event reasoning, and instruction-following tasks. However, their capability for counterfactual reasoning, inferring alternative outcomes under hypothetical conditions, remains underexplored. This capability is essential for robust video understanding, as it requires identifying underlying causal structures and reasoning about unobserved possibilities, rather than merely recognizing observed patterns. To systematically evaluate this capability, we introduce CounterVQA, a video-based benchmark featuring three progressive difficulty levels that assess different aspects of counterfactual reasoning. Through comprehensive evaluation of both state-of-the-art open-source and closed-source models, we uncover a substantial performance gap: while these models achieve reasonable accuracy on simple counterfactual questions, performance degrades significantly on complex multi-hop causal chains. To address these limitations, we develop a post-training method, CFGPT, that enhances a model's visual counterfactual reasoning ability by distilling its counterfactual reasoning capability from the language modality, yielding consistent improvements across all CounterVQA difficulty levels. Dataset and code will be further released.
Authors: Songlin Zhao, Michael Pitts, Zhuwei Qin
Abstract: The rapid advancement of large language models (LLMs) has increased the demand for domain-specialized variants in areas such as law, healthcare, and finance. However, their large size remains a barrier to deployment in resource-constrained environments, and existing compression methods either generalize poorly across domains or incur high overhead. In this work, we propose \textbf{EfficientXpert}, a lightweight domain-pruning framework that combines a propagation-aware pruning criterion (Foresight Mask) with an efficient adapter-update algorithm (Partial Brain Surgeon). Integrated into the LoRA fine-tuning process, EfficientXpert enables a one-step transformation of general pretrained models into sparse, domain-adapted experts. Across health and legal tasks, it retains up to 98% of dense-model performance at 40% sparsity, outperforming state-of-the-art methods. Further analysis reveals substantial domain-dependent structural shifts that degrade the effectiveness of general pruning masks, underscoring the need for adaptive, domain-aware pruning strategies tailored to each domain.
Authors: Xinguo Zhu, Shaohui Peng, Jiaming Guo, Yunji Chen, Qi Guo, Yuanbo Wen, Hang Qin, Ruizhi Chen, Qirui Zhou, Ke Gao, Yanjun Wu, Chen Zhao, Ling Li
Abstract: Developing high-performance GPU kernels is critical for AI and scientific computing, but remains challenging due to its reliance on expert crafting and poor portability. While LLMs offer promise for automation, both general-purpose and finetuned LLMs suffer from two fundamental and conflicting limitations: correctness and efficiency. The key reason is that existing LLM-based approaches directly generate the entire optimized low-level programs, requiring exploration of an extremely vast space encompassing both optimization policies and implementation codes. To address the challenge of exploring an intractable space, we propose Macro Thinking Micro Coding (MTMC), a hierarchical framework inspired by the staged optimization strategy of human experts. It decouples optimization strategy from implementation details, ensuring efficiency through high-level strategy and correctness through low-level implementation. Specifically, Macro Thinking employs reinforcement learning to guide lightweight LLMs in efficiently exploring and learning semantic optimization strategies that maximize hardware utilization. Micro Coding leverages general-purpose LLMs to incrementally implement the stepwise optimization proposals from Macro Thinking, avoiding full-kernel generation errors. Together, they effectively navigate the vast optimization space and intricate implementation details, enabling LLMs for high-performance GPU kernel generation. Comprehensive results on widely adopted benchmarks demonstrate the superior performance of MTMC on GPU kernel generation in both accuracy and running time. On KernelBench, MTMC achieves near 100% and 70% accuracy at Levels 1-2 and 3, over 50% than SOTA general-purpose and domain-finetuned LLMs, with up to 7.3x speedup over LLMs, and 2.2x over expert-optimized PyTorch Eager kernels. On the more challenging TritonBench, MTMC attains up to 59.64% accuracy and 34x speedup.
Authors: Craig Dickson
Abstract: Prior work has shown that fine-tuning models on a narrow domain with misaligned data can lead to broad misalignment - a phenomenon termed "emergent misalignment" (Betley et al. 2025). While all tested models were susceptible to emergent misalignment, some models showed more resistance than others. Specifically the Qwen-2.5 family proved to be relatively resistant, while GPT-4o exhibited the strongest misalignment. In this paper we evaluate if current-generation open-weights models exhibit similar resistance to the Qwen-2.5 family and measure misalignment robustness over a range of model architectures and scales. We replicate the effect across nine modern open-weights models (Gemma 3 and Qwen 3 families, 1B-32B parameters). Models fine-tuned on insecure code generation show a 0.68% misalignment rate (compared to 0.07% for base models), matching the lower end of prior open-model results but dramatically lower than GPT-4o's 20%. We identify a critical format-dependent vulnerability: requiring JSON output doubles misalignment rates compared to natural language prompts (0.96% vs 0.42%). This suggests that structural constraints may bypass safety training by reducing the model's 'degrees of freedom' to refuse. These findings confirm emergent misalignment as a reproducible phenomenon in modern open-weights models, with rates substantially lower than observed in proprietary systems.
Authors: Areeb Ahmad, Abhinav Joshi, Ashutosh Modi
Abstract: Transformer-based language models exhibit complex and distributed behavior, yet their internal computations remain poorly understood. Existing mechanistic interpretability methods typically treat attention heads and multilayer perceptron layers (MLPs) (the building blocks of a transformer architecture) as indivisible units, overlooking possibilities of functional substructure learned within them. In this work, we introduce a more fine-grained perspective that decomposes these components into orthogonal singular directions, revealing superposed and independent computations within a single head or MLP. We validate our perspective on widely used standard tasks like Indirect Object Identification (IOI), Gender Pronoun (GP), and Greater Than (GT), showing that previously identified canonical functional heads, such as the name mover, encode multiple overlapping subfunctions aligned with distinct singular directions. Nodes in a computational graph, that are previously identified as circuit elements show strong activation along specific low-rank directions, suggesting that meaningful computations reside in compact subspaces. While some directions remain challenging to interpret fully, our results highlight that transformer computations are more distributed, structured, and compositional than previously assumed. This perspective opens new avenues for fine-grained mechanistic interpretability and a deeper understanding of model internals.
Authors: Abhinav Joshi, Divyanshu Bhatt, Ashutosh Modi
Abstract: Large Language Models (LLMs) show strong generalization across diverse tasks, yet the internal decision-making processes behind their predictions remain opaque. In this work, we study the geometry of hidden representations in LLMs through the lens of \textit{intrinsic dimension} (ID), focusing specifically on decision-making dynamics in a multiple-choice question answering (MCQA) setting. We perform a large-scale study, with 28 open-weight transformer models and estimate ID across layers using multiple estimators, while also quantifying per-layer performance on MCQA tasks. Our findings reveal a consistent ID pattern across models: early layers operate on low-dimensional manifolds, middle layers expand this space, and later layers compress it again, converging to decision-relevant representations. Together, these results suggest LLMs implicitly learn to project linguistic inputs onto structured, low-dimensional manifolds aligned with task-specific decisions, providing new geometric insights into how generalization and reasoning emerge in language models.
Authors: Chang Gao, Chujie Zheng, Xiong-Hui Chen, Kai Dang, Shixuan Liu, Bowen Yu, An Yang, Shuai Bai, Jingren Zhou, Junyang Lin
Abstract: Reinforcement learning (RL) plays an increasingly important role in enhancing the reasoning capabilities of large language models (LLMs), yet stable and performant policy optimization remains challenging. Token-level importance ratios often exhibit high variance-a phenomenon exacerbated in Mixture-of-Experts models-leading to unstable updates. Existing group-based policy optimization methods, such as GSPO and GRPO, alleviate this problem via hard clipping, making it difficult to maintain both stability and effective learning. We propose Soft Adaptive Policy Optimization (SAPO), which replaces hard clipping with a smooth, temperature-controlled gate that adaptively attenuates off-policy updates while preserving useful learning signals. Compared with GSPO and GRPO, SAPO is both sequence-coherent and token-adaptive. Like GSPO, SAPO maintains sequence-level coherence, but its soft gating forms a continuous trust region that avoids the brittle hard clipping band used in GSPO. When a sequence contains a few highly off-policy tokens, GSPO suppresses all gradients for that sequence, whereas SAPO selectively down-weights only the offending tokens and preserves the learning signal from the near-on-policy ones, improving sample efficiency. Relative to GRPO, SAPO replaces hard token-level clipping with smooth, temperature-controlled scaling, enabling more informative and stable updates. Empirical results on mathematical reasoning benchmarks indicate that SAPO exhibits improved training stability and higher Pass@1 performance under comparable training budgets. Moreover, we employ SAPO to train the Qwen3-VL model series, demonstrating that SAPO yields consistent performance gains across diverse tasks and different model sizes. Overall, SAPO provides a more reliable, scalable, and effective optimization strategy for RL training of LLMs.
Authors: Yi-Hao Peng, Jeffrey P. Bigham, Jason Wu
Abstract: Generative models, such as large language models and text-to-image diffusion models, are increasingly used to create visual designs like user interfaces (UIs) and presentation slides. Finetuning and benchmarking these generative models have often relied on datasets of human-annotated design preferences. Yet, due to the subjective and highly personalized nature of visual design, preference varies widely among individuals. In this paper, we study this problem by introducing DesignPref, a dataset of 12k pairwise comparisons of UI design generation annotated by 20 professional designers with multi-level preference ratings. We found that among trained designers, substantial levels of disagreement exist (Krippendorff's alpha = 0.25 for binary preferences). Natural language rationales provided by these designers indicate that disagreements stem from differing perceptions of various design aspect importance and individual preferences. With DesignPref, we demonstrate that traditional majority-voting methods for training aggregated judge models often do not accurately reflect individual preferences. To address this challenge, we investigate multiple personalization strategies, particularly fine-tuning or incorporating designer-specific annotations into RAG pipelines. Our results show that personalized models consistently outperform aggregated baseline models in predicting individual designers' preferences, even when using 20 times fewer examples. Our work provides the first dataset to study personalized visual design evaluation and support future research into modeling individual design taste.
Authors: Yuwei Niu, Weiyang Jin, Jiaqi Liao, Chaoran Feng, Peng Jin, Bin Lin, Zongjian Li, Bin Zhu, Weihao Yu, Li Yuan
Abstract: Recent years have witnessed significant progress in Unified Multimodal Models, yet a fundamental question remains: Does understanding truly inform generation? To investigate this, we introduce UniSandbox, a decoupled evaluation framework paired with controlled, synthetic datasets to avoid data leakage and enable detailed analysis. Our findings reveal a significant understanding-generation gap, which is mainly reflected in two key dimensions: reasoning generation and knowledge transfer. Specifically, for reasoning generation tasks, we observe that explicit Chain-of-Thought (CoT) in the understanding module effectively bridges the gap, and further demonstrate that a self-training approach can successfully internalize this ability, enabling implicit reasoning during generation. Additionally, for knowledge transfer tasks, we find that CoT assists the generative process by helping retrieve newly learned knowledge, and also discover that query-based architectures inherently exhibit latent CoT-like properties that affect this transfer. UniSandbox provides preliminary insights for designing future unified architectures and training strategies that truly bridge the gap between understanding and generation. Code and data are available at https://github.com/PKU-YuanGroup/UniSandBox
Authors: Peter Zeng, Hannah Stortz, Eric Sclafani, Alina Shabaeva, Maria Elizabeth Garza, Daniel Greeson Owen Rambow
Abstract: We present Gram2Vec, a grammatical style embedding system that embeds documents into a higher dimensional space by extracting the normalized relative frequencies of grammatical features present in the text. Compared to neural approaches, Gram2Vec offers inherent interpretability based on how the feature vectors are generated. In this paper, we use authorship verification and AI detection as two applications to show how Gram2Vec can be used. For authorship verification, we use the features from Gram2Vec to explain why a pair of documents is by the same or by different authors. We also demonstrate how Gram2Vec features can be used to train a classifier for AI detection, outperforming machine learning models trained on a comparable set of Biber features.
Authors: Isack Lee, Haebin Seong
Abstract: Although large language models (LLMs) demonstrate impressive proficiency in various tasks, they present potential safety risks, such as `jailbreaks', where malicious inputs can coerce LLMs into generating harmful content bypassing safety alignments. In this paper, we delve into the ethical biases in LLMs and examine how those biases could be exploited for jailbreaks. Notably, these biases result in a jailbreaking success rate in GPT-4o models that differs by 20\% between non-binary and cisgender keywords and by 16\% between white and black keywords, even when the other parts of the prompts are identical. We introduce the concept of BiasJailbreak, highlighting the inherent risks posed by these safety-induced biases. BiasJailbreak generates biased keywords automatically by asking the target LLM itself, and utilizes the keywords to generate harmful output. Additionally, we propose an efficient defense method BiasDefense, which prevents jailbreak attempts by injecting defense prompts prior to generation. BiasDefense stands as an appealing alternative to Guard Models, such as Llama-Guard, that require additional inference cost after text generation. Our findings emphasize that ethical biases in LLMs can actually lead to generating unsafe output, and suggest a method to make the LLMs more secure and unbiased. To enable further research and improvements, we open-source our code and artifacts of BiasJailbreak, providing the community with tools to better understand and mitigate safety-induced biases in LLMs.
Authors: Shuyang Jiang, Yusheng Liao, Zhe Chen, Ya Zhang, Yanfeng Wang, Yu Wang
Abstract: Medical language models face critical barriers to real-world clinical reasoning applications. However, mainstream efforts, which fall short in task coverage, lack fine-grained supervision for intermediate reasoning steps, and rely on proprietary systems, are still far from a versatile, credible and efficient language model for clinical reasoning usage. To this end, we propose MedS3, a self-evolving framework that imparts robust reasoning capabilities to small, deployable models. Starting with 8,000 curated instances sampled via a curriculum strategy across five medical domains and 16 datasets, we use a small base policy model to conduct Monte Carlo Tree Search (MCTS) for constructing rule-verifiable reasoning trajectories. Self-explored reasoning trajectories ranked by node values are used to bootstrap the policy model via reinforcement fine-tuning and preference learning. Moreover, we introduce a soft dual process reward model that incorporates value dynamics: steps that degrade node value are penalized, enabling fine-grained identification of reasoning errors even when the final answer is correct. Experiments on eleven benchmarks show that MedS3 outperforms the previous state-of-the-art medical model by +6.45 accuracy points and surpasses 32B-scale general-purpose reasoning models by +8.57 points. Additional empirical analysis further demonstrates that MedS3 achieves robust and faithful reasoning behavior.
Authors: Jiaheng Liu, Dawei Zhu, Zhiqi Bai, Yancheng He, Huanxuan Liao, Haoran Que, Zekun Wang, Chenchen Zhang, Ge Zhang, Jiebin Zhang, Yuanxing Zhang, Zhuo Chen, Hangyu Guo, Shilong Li, Ziqiang Liu, Yong Shan, Yifan Song, Jiayi Tian, Wenhao Wu, Zhejian Zhou, Ruijie Zhu, Junlan Feng, Yang Gao, Shizhu He, Zhoujun Li, Tianyu Liu, Fanyu Meng, Wenbo Su, Yingshui Tan, Zili Wang, Jian Yang, Wei Ye, Bo Zheng, Wangchunshu Zhou, Wenhao Huang, Sujian Li, Zhaoxiang Zhang
Abstract: Efficient processing of long contexts has been a persistent pursuit in Natural Language Processing. With the growing number of long documents, dialogues, and other textual data, it is important to develop Long Context Language Models (LCLMs) that can process and analyze extensive inputs in an effective and efficient way. In this paper, we present a comprehensive survey on recent advances in long-context modeling for large language models. Our survey is structured around three key aspects: how to obtain effective and efficient LCLMs, how to train and deploy LCLMs efficiently, and how to evaluate and analyze LCLMs comprehensively. For the first aspect, we discuss data strategies, architectural designs, and workflow approaches oriented with long context processing. For the second aspect, we provide a detailed examination of the infrastructure required for LCLM training and inference. For the third aspect, we present evaluation paradigms for long-context comprehension and long-form generation, as well as behavioral analysis and mechanism interpretability of LCLMs. Beyond these three key aspects, we thoroughly explore the diverse application scenarios where existing LCLMs have been deployed and outline promising future development directions. This survey provides an up-to-date review of the literature on long-context LLMs, which we wish to serve as a valuable resource for both researchers and engineers. An associated GitHub repository collecting the latest papers and repos is available at: \href{https://github.com/LCLM-Horizon/A-Comprehensive-Survey-For-Long-Context-Language-Modeling}{\color[RGB]{175,36,67}{LCLM-Horizon}}.
URLs: https://github.com/LCLM-Horizon/A-Comprehensive-Survey-For-Long-Context-Language-Modeling
Authors: Jing Bi, Susan Liang, Xiaofei Zhou, Pinxin Liu, Junjia Guo, Yunlong Tang, Luchuan Song, Chao Huang, Ali Vosoughi, Guangyu Sun, Jinxi He, Jiarui Wu, Shu Yang, Daoan Zhang, Chen Chen, Lianggong Bruce Wen, Zhang Liu, Jiebo Luo, Chenliang Xu
Abstract: Reasoning is central to human intelligence, enabling structured problem-solving across diverse tasks. Recent advances in large language models (LLMs) have greatly enhanced their reasoning abilities in arithmetic, commonsense, and symbolic domains. However, effectively extending these capabilities into multimodal contexts-where models must integrate both visual and textual inputs-continues to be a significant challenge. Multimodal reasoning introduces complexities, such as handling conflicting information across modalities, which require models to adopt advanced interpretative strategies. Addressing these challenges involves not only sophisticated algorithms but also robust methodologies for evaluating reasoning accuracy and coherence. This paper offers a concise yet insightful overview of reasoning techniques in both textual and multimodal LLMs. Through a thorough and up-to-date comparison, we clearly formulate core reasoning challenges and opportunities, highlighting practical methods for post-training optimization and test-time inference. Our work provides valuable insights and guidance, bridging theoretical frameworks and practical implementations, and sets clear directions for future research.
Authors: Zhihao Zhang, Yiran Zhang, Xiyue Zhou, Liting Huang, Imran Razzak, Preslav Nakov, Usman Naseem
Abstract: Infodemics and health misinformation have significant negative impact on individuals and society, exacerbating confusion and increasing hesitancy in adopting recommended health measures. Recent advancements in generative AI, capable of producing realistic, human like text and images, have significantly accelerated the spread and expanded the reach of health misinformation, resulting in an alarming surge in its dissemination. To combat the infodemics, most existing work has focused on developing misinformation datasets from social media and fact checking platforms, but has faced limitations in topical coverage, inclusion of AI generation, and accessibility of raw content. To address these issues, we present MM Health, a large scale multimodal misinformation dataset in the health domain consisting of 34,746 news article encompassing both textual and visual information. MM Health includes human-generated multimodal information (5,776 articles) and AI generated multimodal information (28,880 articles) from various SOTA generative AI models. Additionally, We benchmarked our dataset against three tasks (reliability checks, originality checks, and fine-grained AI detection) demonstrating that existing SOTA models struggle to accurately distinguish the reliability and origin of information. Our dataset aims to support the development of misinformation detection across various health scenarios, facilitating the detection of human and machine generated content at multimodal levels.
Authors: Marvin Limpijankit, Yanda Chen, Melanie Subbiah, Nicholas Deas, Kathleen McKeown
Abstract: LLMs can be unpredictable, as even slight alterations to the prompt can cause the output to change in unexpected ways. Thus, the ability of models to accurately explain their behavior is critical, especially in high-stakes settings. One approach for evaluating explanations is counterfactual simulatability, how well an explanation allows users to infer the model's output on related counterfactuals. Counterfactual simulatability has been previously studied for yes/no question answering tasks. We provide a general framework for extending this method to generation tasks, using news summarization and medical suggestion as example use cases. We find that while LLM explanations do enable users to better predict LLM outputs on counterfactuals in the summarization setting, there is significant room for improvement for medical suggestion. Furthermore, our results suggest that the evaluation for counterfactual simulatability may be more appropriate for skill-based tasks as opposed to knowledge-based tasks.
Authors: Yiran Zhang, Mo Wang, Xiaoyang Li, Kaixuan Ren, Chencheng Zhu, Usman Naseem
Abstract: Despite impressive advances in large language models (LLMs), existing benchmarks often focus on single-turn or single-step tasks, failing to capture the kind of iterative reasoning required in real-world settings. To address this limitation, we introduce TurnBench, a novel benchmark that evaluates multi-turn, multi-step reasoning through an interactive code-breaking task inspired by the "Turing Machine Board Game." In each episode, a model must uncover hidden logical or arithmetic rules by making sequential guesses, receiving structured feedback, and integrating clues across multiple rounds. This dynamic setup requires models to reason over time, adapt based on past information, and maintain consistency across steps-capabilities underexplored in current benchmarks. TurnBench includes two modes: Classic, which tests standard reasoning, and Nightmare, which introduces increased complexity and requires robust inferential chains. To support fine-grained analysis, we provide ground-truth annotations for intermediate reasoning steps. Our evaluation of state-of-the-art LLMs reveals significant gaps: the best model achieves 84% accuracy in Classic mode, but performance drops to 18% in Nightmare mode. In contrast, human participants achieve 100% in both, underscoring the challenge TurnBench poses to current models. By incorporating feedback loops and hiding task rules, TurnBench reduces contamination risks and provides a rigorous testbed for diagnosing and advancing multi-step, multi-turn reasoning in LLMs.
Authors: Nelvin Tan, Zian Seng, Liang Zhang, Yu-Ching Shih, Dong Yang, Amol Salunkhe
Abstract: Large language models (LLMs) have shown impressive capabilities on numerous natural language processing tasks. However, LLMs still struggle with numerical question answering for financial documents that include tabular and textual data. Recent works have showed the effectiveness of critic agents (i.e., self-correction) for this task given oracle labels. Building upon this framework, this paper examines the effectiveness of the traditional critic agent when oracle labels are not available, and show, through experiments, that this critic agent's performance deteriorates in this scenario. With this in mind, we present an improved critic agent, along with the calculator agent which outperforms the previous state-of-the-art approach (program-of-thought) and is safer. Furthermore, we investigate how our agents interact with each other, and how this interaction affects their performance.
Authors: Devesh Pant, Rishi Raj Grandhe, Vipin Samaria, Mukul Paul, Sudhir Kumar, Saransh Khanna, Jatin Agrawal, Jushaan Singh Kalra, Akhil VSSG, Satish V Khalikar, Vipin Garg, Himanshu Chauhan, Pranay Verma, Neha Khandelwal, Soma S Dhavala, Minesh Mathew
Abstract: Early detection of disease outbreaks is crucial to ensure timely intervention by the health authorities. Due to the challenges associated with traditional indicator-based surveillance, monitoring informal sources such as online media has become increasingly popular. However, owing to the number of online articles getting published everyday, manual screening of the articles is impractical. To address this, we propose Health Sentinel. It is a multi-stage information extraction pipeline that uses a combination of ML and non-ML methods to extract events-structured information concerning disease outbreaks or other unusual health events-from online articles. The extracted events are made available to the Media Scanning and Verification Cell (MSVC) at the National Centre for Disease Control (NCDC), Delhi for analysis, interpretation and further dissemination to local agencies for timely intervention. From April 2022 till date, Health Sentinel has processed over 300 million news articles and identified over 95,000 unique health events across India of which over 3,500 events were shortlisted by the public health experts at NCDC as potential outbreaks.
Authors: Chuanyang Zheng, Jiankai Sun, Yihang Gao, Yuehao Wang, Peihao Wang, Jing Xiong, Liliang Ren, Hao Cheng, Janardhan Kulkarni, Yelong Shen, Atlas Wang, Mac Schwager, Anderson Schneider, Xiaodong Liu, Jianfeng Gao
Abstract: The attention mechanism is a core component of the Transformer architecture. Various methods have been developed to compute attention scores, including multi-head attention (MHA), multi-query attention, group-query attention and so on. We further analyze the MHA and observe that its performance improves as the number of attention heads increases, provided the hidden size per head remains sufficiently large. Therefore, increasing both the head count and hidden size per head with minimal parameter overhead can lead to significant performance gains at a low cost. Motivated by this insight, we introduce Simulated Attention Score (SAS), which maintains a compact model size while simulating a larger number of attention heads and hidden feature dimension per head. This is achieved by projecting a low-dimensional head representation into a higher-dimensional space, effectively increasing attention capacity without increasing parameter count. Beyond the head representations, we further extend the simulation approach to feature dimension of the key and query embeddings, enhancing expressiveness by mimicking the behavior of a larger model while preserving the original model size. To control the parameter cost, we also propose Parameter-Efficient Attention Aggregation (PEAA). Comprehensive experiments on a variety of datasets and tasks demonstrate the effectiveness of the proposed SAS method, achieving significant improvements over different attention variants.
Authors: Samuel Ackerman, Gal Amram, Ora Nova Fandina, Eitan Farchi, Shmulik Froimovich, Raviv Gal, Wesam Ibraheem, Avi Ziv
Abstract: Large Language Models (LLMs) are increasingly used as evaluators in natural language processing tasks, a paradigm known as LLM-as-a-Judge (LaaJ). The analysis of a LaaJ software, commonly refereed to as meta-evaluation, pose significant challenges in domain-specific contexts. In such domains, in contrast to general domains, annotated data is scarce and expert evaluation is costly. As a result, meta-evaluation is often performed using metrics that have not been validated for the specific domain in which they are applied. Therefore, it becomes difficult to determine which metrics effectively identify LaaJ quality, and further, what threshold indicates sufficient evaluator performance. In this work, we introduce LaaJMeter, a simulation-based framework for controlled meta-evaluation of LaaJs. LaaJMeter enables engineers to generate synthetic data representing virtual models and judges, allowing systematic analysis of evaluation metrics under realistic conditions. This helps practitioners validate LaaJs for specific tasks: they can test whether their metrics correctly distinguish between high and low quality (virtual) LaaJs, and estimate appropriate thresholds for evaluator adequacy. We demonstrate the utility of LaaJMeter in a code translation task involving a legacy programming language, showing how different metrics vary in sensitivity to evaluator quality. Our results highlight the limitations of common metrics and the importance of principled metric selection. LaaJMeter provides a scalable and extensible solution for assessing LaaJs in low-resource settings, contributing to the broader effort to ensure trustworthy and reproducible evaluation in NLP.
Authors: Yibo Li, Miao Xiong, Jiaying Wu, Bryan Hooi
Abstract: Large Language Models (LLMs) are increasingly deployed in high-stakes domains such as science, law, and healthcare, where accurate expressions of uncertainty are essential for reliability and trust. However, current LLMs are often observed to generate incorrect answers with high confidence, a phenomenon known as "overconfidence". Recent efforts have focused on calibrating LLMs' verbalized confidence: i.e., their expressions of confidence in text form, such as "I am 80% confident that...". Existing approaches either rely on prompt engineering or fine-tuning with heuristically generated uncertainty estimates, both of which have limited effectiveness and generalizability. Motivated by the notion of proper scoring rules for calibration in classical machine learning models, we introduce ConfTuner, a simple and efficient fine-tuning method that introduces minimal overhead and does not require ground-truth confidence scores or proxy confidence estimates. ConfTuner relies on a new loss function, tokenized Brier score, which we theoretically prove to be a proper scoring rule, intuitively meaning that it "correctly incentivizes the model to report its true probability of being correct". ConfTuner improves calibration across diverse reasoning tasks and generalizes to black-box models such as GPT-4o. Our results further show that better-calibrated confidence enables downstream gains in self-correction and model cascade, advancing the development of trustworthy LLM systems. The code is available at https://github.com/liushiliushi/ConfTuner.
Authors: Bowen Qin, Chen Yue, Fang Yin, Hui Wang, JG Yao, Jiakang Liu, Jing-Shu Zheng, Miguel Hu Chen, Richeng Xuan, Shibei Meng, Shiqi Zhou, Teng Dai, Tong-Shuai Ren, Wei Cui, Xi Yang, Xialin Du, Xiaojing Xu, Xue Sun, Xuejing Li, Yaming Liu, Yesheng Liu, Ying Liu, Yonghua Lin, Yu Zhao, Yunduo Zhang, Yuwen Luo, Zheqi He, Zhiyuan He, Zhongyuan Wang
Abstract: We conduct a moderate-scale contamination-free (to some extent) evaluation of current large reasoning models (LRMs) with some preliminary findings. We also release ROME, our evaluation benchmark for vision language models intended to test reasoning from visual clues. We attach links to the benchmark, evaluation data, and other updates on this website: https://flageval-baai.github.io/LRM-Eval/
Authors: Pengfei Wang, Baolin Sun, Xuemei Dong, Yaxun Dai, Hongwei Yuan, Mengdie Chu, Yingqi Gao, Xiang Qi, Peng Zhang, Ying Yan
Abstract: State-of-the-art (SOTA) Text-to-SQL methods still lag significantly behind human experts on challenging benchmarks like BIRD. Current approaches that explore test-time scaling lack an orchestrated strategy and neglect the model's internal reasoning process. To bridge this gap, we introduce Agentar-Scale-SQL, a novel framework leveraging scalable computation to improve performance. Agentar-Scale-SQL implements an Orchestrated Test-Time Scaling strategy that synergistically combines three distinct perspectives: i) Internal Scaling via RL-enhanced Intrinsic Reasoning, ii) Sequential Scaling through Iterative Refinement, and iii) Parallel Scaling using Diverse Synthesis and Tournament Selection. Agentar-Scale-SQL is a general-purpose framework designed for easy adaptation to new databases and more powerful language models. Extensive experiments show that Agentar-Scale-SQL achieves SOTA performance on the BIRD benchmark, reaching 81.67% execution accuracy on the test set and ranking first on the official leaderboard, demonstrating an effective path toward human-level performance.
Authors: Yida Xue, Mingjun Mao, Xiangyuan Ru, Yuqi Zhu, Baochang Ren, Shuofei Qiao, Mengru Wang, Shumin Deng, Xinyu An, Ningyu Zhang, Ying Chen, Huajun Chen
Abstract: We introduce OceanGym, the first comprehensive benchmark for ocean underwater embodied agents, designed to advance AI in one of the most demanding real-world environments. Unlike terrestrial or aerial domains, underwater settings present extreme perceptual and decision-making challenges, including low visibility, dynamic ocean currents, making effective agent deployment exceptionally difficult. OceanGym encompasses eight realistic task domains and a unified agent framework driven by Multi-modal Large Language Models (MLLMs), which integrates perception, memory, and sequential decision-making. Agents are required to comprehend optical and sonar data, autonomously explore complex environments, and accomplish long-horizon objectives under these harsh conditions. Extensive experiments reveal substantial gaps between state-of-the-art MLLM-driven agents and human experts, highlighting the persistent difficulty of perception, planning, and adaptability in ocean underwater environments. By providing a high-fidelity, rigorously designed platform, OceanGym establishes a testbed for developing robust embodied AI and transferring these capabilities to real-world autonomous ocean underwater vehicles, marking a decisive step toward intelligent agents capable of operating in one of Earth's last unexplored frontiers. The code and data are available at https://github.com/OceanGPT/OceanGym.
Authors: Gregory Hok Tjoan Go, Khang Ly, Anders S{\o}gaard, Amin Tabatabaei, Maarten de Rijke, Xinyi Chen
Abstract: The rapid growth of scientific publications has made it increasingly difficult to keep literature reviews comprehensive and up-to-date. Though prior work has focused on automating retrieval and screening, the writing phase of systematic reviews remains largely under-explored, especially with regard to readability and factual accuracy. To address this, we present LiRA (Literature Review Agents), a multi-agent collaborative workflow which emulates the human literature review process. LiRA utilizes specialized agents for content outlining, subsection writing, editing, and reviewing, producing cohesive and comprehensive review articles. Evaluated on SciReviewGen and a proprietary ScienceDirect dataset, LiRA outperforms current baselines such as AutoSurvey and MASS-Survey in writing and citation quality, while maintaining competitive similarity to human-written reviews. We further evaluate LiRA in real-world scenarios using document retrieval and assess its robustness to reviewer model variation. Our findings highlight the potential of agentic LLM workflows, even without domain-specific tuning, to improve the reliability and usability of automated scientific writing.
Authors: Jizhan Fang, Xinle Deng, Haoming Xu, Ziyan Jiang, Yuqi Tang, Ziwen Xu, Shumin Deng, Yunzhi Yao, Mengru Wang, Shuofei Qiao, Huajun Chen, Ningyu Zhang
Abstract: Despite their remarkable capabilities, Large Language Models (LLMs) struggle to effectively leverage historical interaction information in dynamic and complex environments. Memory systems enable LLMs to move beyond stateless interactions by introducing persistent information storage, retrieval, and utilization mechanisms. However, existing memory systems often introduce substantial time and computational overhead. To this end, we introduce a new memory system called LightMem, which strikes a balance between the performance and efficiency of memory systems. Inspired by the Atkinson-Shiffrin model of human memory, LightMem organizes memory into three complementary stages. First, cognition-inspired sensory memory rapidly filters irrelevant information through lightweight compression and groups information according to their topics. Next, topic-aware short-term memory consolidates these topic-based groups, organizing and summarizing content for more structured access. Finally, long-term memory with sleep-time update employs an offline procedure that decouples consolidation from online inference. On LongMemEval and LoCoMo, using GPT and Qwen backbones, LightMem consistently surpasses strong baselines, improving QA accuracy by up to 7.7% / 29.3%, reducing total token usage by up to 38x / 20.9x and API calls by up to 30x / 55.5x, while purely online test-time costs are even lower, achieving up to 106x / 117x token reduction and 159x / 310x fewer API calls. The code is available at https://github.com/zjunlp/LightMem.
Authors: Mehrdad Ghassabi, Sadra Hakim, Hamidreza Baradaran Kashani, Pedram Rostami
Abstract: Enhancing reasoning capabilities in small language models is critical for specialized applications such as medical question answering, particularly in underrepresented languages like Persian. In this study, we employ Reinforcement Learning with AI Feedback (RLAIF) and Direct preference optimization (DPO) to improve the reasoning skills of a general-purpose Persian language model. To achieve this, we translated a multiple-choice medical question-answering dataset into Persian and used RLAIF to generate rejected-preferred answer pairs, which are essential for DPO training. By prompting both teacher and student models to produce Chain-of-Thought (CoT) reasoning responses, we compiled a dataset containing correct and incorrect reasoning trajectories. This dataset, comprising 2 million tokens in preferred answers and 2.5 million tokens in rejected ones, was used to train a baseline model, significantly enhancing its medical reasoning capabilities in Persian. Remarkably, the resulting model outperformed its predecessor, gaokerena-V, which was trained on approximately 57 million tokens, despite leveraging a much smaller dataset. These results highlight the efficiency and effectiveness of reasoning-focused training approaches in developing domain-specific language models with limited data availability.
Authors: Yusheng Liao, Chaoyi Wu, Junwei Liu, Shuyang Jiang, Pengcheng Qiu, Haowen Wang, Yun Yue, Shuai Zhen, Jian Wang, Qianrui Fan, Jinjie Gu, Ya Zhang, Yanfeng Wang, Yu Wang, Weidi Xie
Abstract: Electronic Health Records (EHRs) contain rich yet complex information, and their automated analysis is critical for clinical decision-making. Despite recent advances of large language models (LLMs) in clinical workflows, their ability to analyze EHRs remains limited due to narrow task coverage and lack of EHR-oriented reasoning capabilities. This paper aims to bridge the gap, specifically, we present EHR-Ins, a large-scale, comprehensive EHR reasoning instruction dataset, comprising 300k high-quality reasoning cases and 4M non-reasoning cases across 42 distinct EHR tasks. Its core innovation is a thinking-graph-driven framework that enables to generate high-quality reasoning data at scale. Based on it, we develop EHR-R1, a series of reasoning-enhanced LLMs with up to 72B parameters tailored for EHR analysis. Through a multi-stage training paradigm, including domain adaptation, reasoning enhancement, and reinforcement learning, EHR-R1 systematically acquires domain knowledge and diverse reasoning capabilities, enabling accurate and robust EHR analysis. Lastly, we introduce EHR-Bench, a new benchmark curated from MIMIC-IV, spanning 42 tasks, to comprehensively assess reasoning and prediction across EHR scenarios. In experiments, we show that the resulting EHR-R1 consistently outperforms state-of-the-art commercial and open-source LLMs (including DeepSeek-V3 and GPT-4o), surpassing GPT-4o by over 30 points on MIMIC-Bench and achieving a 10\% higher zero-shot AUROC on EHRSHOT. Collectively, EHR-Ins, EHR-R1, and EHR-Bench have significantly advanced the development for more reliable and clinically relevant EHR analysis.
Authors: Mo El-Haj, Paul Rayson
Abstract: We introduce AraFinNews, the largest publicly available Arabic financial news dataset to date, comprising 212,500 article-headline pairs spanning a decade of reporting from 2015 to 2025. Designed as an Arabic counterpart to major English summarisation corpora such as CNN/DailyMail, AraFinNews provides a realistic benchmark for evaluating domain-specific language understanding and generation in financial contexts. Using this resource, we investigate the impact of domain specificity on abstractive summarisation of Arabic financial texts with large language models (LLMs). In particular, we evaluate transformer-based models: mT5, AraT5, and the domain-adapted FinAraT5 to examine how financial-domain pretraining influences accuracy, numerical reliability, and stylistic alignment with professional reporting. Experimental results show that domain-adapted models generate more coherent summaries, especially in their handling of quantitative and entity-centric information. These findings highlight the importance of domain-specific adaptation for improving narrative fluency in Arabic financial summarisation. The dataset is freely available for non-commercial research at https://github.com/ArabicNLP-uk/AraFinNews.
Authors: Nicol\`o Pagan, Petter T\"ornberg, Christopher A. Bail, Anik\'o Hann\'ak, Christopher Barrie
Abstract: Large language models (LLMs) are increasingly used in the social sciences to simulate human behavior, based on the assumption that they can generate realistic, human-like text. Yet this assumption remains largely untested. Existing validation efforts rely heavily on human-judgment-based evaluations -- testing whether humans can distinguish AI from human output -- despite evidence that such judgments are blunt and unreliable. As a result, the field lacks robust tools for assessing the realism of LLM-generated text or for calibrating models to real-world data. This paper makes two contributions. First, we introduce a computational Turing test: a validation framework that integrates aggregate metrics (BERT-based detectability and semantic similarity) with interpretable linguistic features (stylistic markers and topical patterns) to assess how closely LLMs approximate human language within a given dataset. Second, we systematically compare nine open-weight LLMs across five calibration strategies -- including fine-tuning, stylistic prompting, and context retrieval -- benchmarking their ability to reproduce user interactions on X (formerly Twitter), Bluesky, and Reddit. Our findings challenge core assumptions in the literature. Even after calibration, LLM outputs remain clearly distinguishable from human text, particularly in affective tone and emotional expression. Instruction-tuned models underperform their base counterparts, and scaling up model size does not enhance human-likeness. Crucially, we identify a trade-off: optimizing for human-likeness often comes at the cost of semantic fidelity, and vice versa. These results provide a much-needed scalable framework for validation and calibration in LLM simulations -- and offer a cautionary note about their current limitations in capturing human communication.
Authors: Jiarui Liu, Kaustubh Dhole, Yingheng Wang, Haoyang Wen, Sarah Zhang, Haitao Mao, Gaotang Li, Neeraj Varshney, Jingguo Liu, Xiaoman Pan
Abstract: Deductive reasoning is the process of deriving conclusions strictly from the given premises, without relying on external knowledge. We define honesty in this setting as a model's ability to respond only when the conclusion is logically entailed by the premises, and to abstain otherwise. However, current language models often fail to reason honestly, producing unwarranted answers when the input is insufficient. To study this challenge, we formulate honest deductive reasoning as multi-step tasks where models must either derive the correct conclusion or abstain. We curate two datasets from graph structures, one for linear algebra and one for logical inference, and introduce unanswerable cases by randomly perturbing an edge in half of the instances. We find that prompting and existing training methods, including GRPO with or without supervised fine-tuning initialization, struggle on these tasks. In particular, GRPO optimize only for final task outcomes, leaving models vulnerable to collapse when negative rewards dominate early training. To address this, we propose ACNCHOR, a reinforcement learning method that injects ground truth trajectories into rollouts, preventing early training collapse. Our results demonstrate that this method stabilizes learning and significantly improves the overall reasoning performance, underscoring the importance of training dynamics for enabling honest deductive reasoning in language models.
Authors: Jos\'e Pombal, Maya D'Eon, Nuno M. Guerreiro, Pedro Henrique Martins, Ant\'onio Farinhas, Ricardo Rei
Abstract: Demand for mental health support through AI chatbots is surging, though current systems present several limitations, like sycophancy or overvalidation, and reinforcement of maladaptive beliefs. A core obstacle to the creation of better systems is the scarcity of benchmarks that capture the complexity of real therapeutic interactions. Most existing benchmarks either only test clinical knowledge through multiple-choice questions or assess single responses in isolation. To bridge this gap, we present MindEval, a framework designed in collaboration with Ph.D-level Licensed Clinical Psychologists for automatically evaluating language models in realistic, multi-turn mental health therapy conversations. Through patient simulation and automatic evaluation with LLMs, our framework balances resistance to gaming with reproducibility via its fully automated, model-agnostic design. We begin by quantitatively validating the realism of our simulated patients against human-generated text and by demonstrating strong correlations between automatic and human expert judgments. Then, we evaluate 12 state-of-the-art LLMs and show that all models struggle, scoring below 4 out of 6, on average, with particular weaknesses in problematic AI-specific patterns of communication. Notably, reasoning capabilities and model scale do not guarantee better performance, and systems deteriorate with longer interactions or when supporting patients with severe symptoms. We release all code, prompts, and human evaluation data.
Authors: Linxiao Cao, Ruitao Wang, Jindong Li, Zhipeng Zhou, Menglin Yang
Abstract: Retrieval-augmented generation (RAG) enables large language models (LLMs) to access external knowledge, helping mitigate hallucinations and enhance domain-specific expertise. Graph-based RAG enhances structural reasoning by introducing explicit relational organization that enables information propagation across semantically connected text units. However, these methods typically rely on Euclidean embeddings that capture semantic similarity but lack a geometric notion of hierarchical depth, limiting their ability to represent abstraction relationships inherent in complex knowledge graphs. To capture both fine-grained semantics and global hierarchy, we propose HyperbolicRAG, a retrieval framework that integrates hyperbolic geometry into graph-based RAG. HyperbolicRAG introduces three key designs: (1) a depth-aware representation learner that embeds nodes within a shared Poincare manifold to align semantic similarity with hierarchical containment, (2) an unsupervised contrastive regularization that enforces geometric consistency across abstraction levels, and (3) a mutual-ranking fusion mechanism that jointly exploits retrieval signals from Euclidean and hyperbolic spaces, emphasizing cross-space agreement during inference. Extensive experiments across multiple QA benchmarks demonstrate that HyperbolicRAG outperforms competitive baselines, including both standard RAG and graph-augmented baselines.
Authors: Tianyu Fu, Haofeng Huang, Xuefei Ning, Genghan Zhang, Boju Chen, Tianqi Wu, Hongyi Wang, Zixiao Huang, Shiyao Li, Shengen Yan, Guohao Dai, Huazhong Yang, Yu Wang
Abstract: Sliding-window attention offers a hardware-efficient solution to the memory and throughput challenges of Large Language Models (LLMs) in long-context scenarios. Existing methods typically employ a single window length across all attention heads and input sizes. However, this uniform approach fails to capture the heterogeneous attention patterns inherent in LLMs, ignoring their distinct accuracy-latency trade-offs. To address this challenge, we propose *Mixture of Attention Spans* (MoA), which automatically tailors distinct sliding-window length configurations to different heads and layers. MoA constructs and navigates a search space of various window lengths and their scaling rules relative to input sizes. It profiles the model, evaluates potential configurations, and pinpoints the optimal length configurations for each head. MoA adapts to varying input sizes, revealing that some attention heads expand their focus to accommodate longer inputs, while other heads consistently concentrate on fixed-length local contexts. Experiments show that MoA increases the effective context length by 3.9x with the same average sliding-window length, boosting retrieval accuracy by 1.5-7.1x over the uniform-window baseline across Vicuna-{7B, 13B} and Llama3-{8B, 70B} models. Moreover, MoA narrows the performance gap with full attention, reducing the maximum relative performance drop from 9%-36% to within 5% across three long-context understanding benchmarks. MoA achieves a 1.2-1.4x GPU memory reduction, boosting decode throughput by 6.6-8.2x and 1.7-1.9x over FlashAttention2 and vLLM, with minimal performance impact. Our code is available at: https://github.com/thu-nics/MoA
Authors: Farhad Nooralahzadeh, Yi Zhang, Jonathan Furst, Kurt Stockinger
Abstract: International enterprises, organizations, and hospitals collect large amounts of multi-modal data stored in databases, text documents, images, and videos. While there has been recent progress in the separate fields of multi-modal data exploration as well as in database systems that automatically translate natural language questions to database query languages, the research challenge of querying both structured databases and unstructured modalities (e.g., texts, images) in natural language remains largely unexplored. In this paper, we propose M$^2$EX -a system that enables multi-modal data exploration via language agents. Our approach is based on the following research contributions: (1) Our system is inspired by a real-world use case that enables users to explore multi-modal information systems. (2) M$^2$EX leverages an LLM-based agentic AI framework to decompose a natural language question into subtasks such as text-to-SQL generation and image analysis and to orchestrate modality-specific experts in an efficient query plan. (3) Experimental results on multi-modal datasets, encompassing relational data, text, and images, demonstrate that our system outperforms state-of-the-art multi-modal exploration systems, excelling in both accuracy and various performance metrics, including query latency, API costs, and planning efficiency, thanks to the more effective utilization of the reasoning capabilities of LLMs.
Authors: Tai-Danae Bradley, Juan Pablo Vigneaux
Abstract: The purpose of this article is twofold. Firstly, we use the next-token probabilities given by a language model to explicitly define a category of texts in natural language enriched over the unit interval, in the sense of Bradley, Terilla, and Vlassopoulos. We consider explicitly the terminating conditions for text generation and determine when the enrichment itself can be interpreted as a probability over texts. Secondly, we compute the M\"obius function and the magnitude of an associated generalized metric space of texts. The magnitude function of that space is a sum over texts (prompts) of the $t$-logarithmic (Tsallis) entropies of the next-token probability distributions associated with each prompt, plus the cardinality of the model's possible outputs. A suitable evaluation of the magnitude function's derivative recovers a sum of Shannon entropies, which justifies seeing magnitude as a partition function. Following Leinster and Shulman, we also express the magnitude function of the generalized metric space as an Euler characteristic of magnitude homology and provide an explicit description of the zeroeth and first magnitude homology groups.
Authors: Anton Alyakin, Jaden Stryker, Daniel Alexander Alber, Jin Vivian Lee, Karl L. Sangwon, Brandon Duderstadt, Akshay Save, David Kurland, Spencer Frome, Shrutika Singh, Jeff Zhang, Eunice Yang, Ki Yun Park, Cordelia Orillac, Aly A. Valliani, Sean Neifert, Albert Liu, Aneek Patel, Christopher Livia, Darryl Lau, Ilya Laufer, Peter A. Rozman, Eveline Teresa Hidalgo, Howard Riina, Rui Feng, Todd Hollon, Yindalon Aphinyanaphongs, John G. Golfinos, Laura Snyder, Eric Leuthardt, Douglas Kondziolka, Eric Karl Oermann
Abstract: General-purpose VLMs demonstrate impressive capabilities, but their opaque training on uncurated internet data poses critical limitations for high-stakes decision-making, such as in neurosurgery. We present CNS-Obsidian, a neurosurgical VLM trained on peer-reviewed literature, and demonstrate its clinical utility versus GPT-4o in a real-world setting. We compiled 23,984 articles from Neurosurgery Publications journals, yielding 78,853 figures and captions. Using GPT-4o and Claude Sonnet-3.5, we converted these into 263,064 training samples across three formats: instruction fine-tuning, multiple-choice questions, and differential diagnosis. We trained CNS-Obsidian, a fine-tune of the 34-billion parameter LLaVA-Next model. In a blinded, randomized trial at NYU Langone Health (Aug 30-Nov 30, 2024), neurosurgery consultations were assigned to either CNS-Obsidian or a HIPAA-compliant GPT-4o endpoint as diagnostic co-pilot after consultations. Primary outcomes were diagnostic helpfulness and accuracy, assessed via user ratings and presence of correct diagnosis within the VLM-provided differential. CNS-Obsidian matched GPT-4o on synthetic questions (76.13% vs 77.54%, p=0.235), but only achieved 46.81% accuracy on human-generated questions versus GPT-4o's 65.70% (p<10-15). In the randomized trial, 70 consultations were evaluated (32 CNS-Obsidian, 38 GPT-4o) from 959 total consults (7.3% utilization). CNS-Obsidian received positive ratings in 40.62% of cases versus 57.89% for GPT-4o (p=0.230). Both models included correct diagnosis in approximately 60% of cases (59.38% vs 65.79%, p=0.626). Domain-specific VLMs trained on curated scientific literature can approach frontier model performance despite being orders of magnitude smaller and less expensive to train. This establishes a transparent framework for scientific communities to build specialized AI models.
Authors: Vlad Hondru, Eduard Hogea, Darian Onchis, Radu Tudor Ionescu
Abstract: The ever growing realism and quality of generated videos makes it increasingly harder for humans to spot deepfake content, who need to rely more and more on automatic deepfake detectors. However, deepfake detectors are also prone to errors, and their decisions are not explainable, leaving humans vulnerable to deepfake-based fraud and misinformation. To this end, we introduce ExDDV, the first dataset and benchmark for Explainable Deepfake Detection in Video. ExDDV comprises around 5.4K real and deepfake videos that are manually annotated with text descriptions (to explain the artifacts) and clicks (to point out the artifacts). We evaluate a number of vision-language models on ExDDV, performing experiments with various fine-tuning and in-context learning strategies. Our results show that text and click supervision are both required to develop robust explainable models for deepfake videos, which are able to localize and describe the observed artifacts. Our novel dataset and code to reproduce the results are available at https://github.com/vladhondru25/ExDDV.
Authors: Junjie Xing, Yeye He, Mengyu Zhou, Haoyu Dong, Shi Han, Lingjiao Chen, Dongmei Zhang, Surajit Chaudhuri, H. V. Jagadish
Abstract: Tables and table-based use cases play a crucial role in many important real-world applications, such as spreadsheets, databases, and computational notebooks, which traditionally require expert-level users like data engineers, data analysts, and database administrators to operate. Although LLMs have shown remarkable progress in working with tables (e.g., in spreadsheet and database copilot scenarios), comprehensive benchmarking of such capabilities remains limited. In contrast to an extensive and growing list of NLP benchmarks, evaluations of table-related tasks are scarce, and narrowly focus on tasks like NL-to-SQL and Table-QA, overlooking the broader spectrum of real-world tasks that professional users face. This gap limits our understanding and model progress in this important area. In this work, we introduce MMTU, a large-scale benchmark with over 28K questions across 25 real-world table tasks, designed to comprehensively evaluate models ability to understand, reason, and manipulate real tables at the expert-level. These tasks are drawn from decades' worth of computer science research on tabular data, with a focus on complex table tasks faced by professional users. We show that MMTU require a combination of skills -- including table understanding, reasoning, and coding -- that remain challenging for today's frontier models, where even frontier reasoning models like OpenAI GPT-5 and DeepSeek R1 score only around 69\% and 57\% respectively, suggesting significant room for improvement. We highlight key findings in our evaluation using MMTU and hope that this benchmark drives further advances in understanding and developing foundation models for structured data processing and analysis. Our code and data are available at https://github.com/MMTU-Benchmark/MMTU and https://huggingface.co/datasets/MMTU-benchmark/MMTU.
URLs: https://github.com/MMTU-Benchmark/MMTU, https://huggingface.co/datasets/MMTU-benchmark/MMTU.
Authors: Ruichen Xu, Kexin Chen
Abstract: Modern large language models (LLMs) demonstrate exceptional performance on knowledge-intensive tasks, yet the theoretical mechanisms underlying knowledge acquisition (storage and memorization) during pre-training and extraction (retrieval and recall) during inference after fine-tuning remain poorly understood. Although prior theoretical studies have explored these processes through analyses of training dynamics, they overlook critical components essential for a comprehensive theory: (1) the multi-layer perceptron (MLP), empirically identified as the primary module for knowledge storage; (2) out-of-distribution (OOD) adaptivity, which enables LLMs to generalize to unseen scenarios post-pre-training; and (3) next-token prediction, the standard autoregressive objective that encodes knowledge as conditional probabilities. In this work, we introduce, to the best of our knowledge, the first theoretical framework that addresses these limitations by examining the training dynamics of one-layer transformers. Under regularity assumptions, we establish that: (i) transformers attain near-optimal training loss during pre-training, demonstrating effective knowledge acquisition; (ii) given a sufficiently large fine-tuning dataset and appropriate data multiplicity conditions, transformers achieve low generalization error on factual knowledge acquired during pre-training but not revisited in fine-tuning, indicating robust knowledge extraction; and (iii) violation of these conditions leads to elevated generalization error, manifesting as hallucinations. Our analysis encompasses both full fine-tuning and low-rank fine-tuning, yielding insights into the efficacy of practical low-rank adaptation methods. We validate our theoretical findings through experiments on synthetic datasets and the real-world PopQA benchmark, employing GPT-2 and Llama-3.2-1B models.
Authors: Mingxuan Cui, Yilan Jiang, Duo Zhou, Cheng Qian, Yuji Zhang, Qiong Wang
Abstract: Drug shortages pose critical risks to patient care and healthcare systems worldwide, yet the effectiveness of regulatory interventions remains poorly understood due to information asymmetries in pharmaceutical supply chains. We propose \textbf{ShortageSim}, addresses this challenge by providing the first simulation framework that evaluates the impact of regulatory interventions on competition dynamics under information asymmetry. Using Large Language Model (LLM)-based agents, the framework models the strategic decisions of drug manufacturers and institutional buyers, in response to shortage alerts given by the regulatory agency. Unlike traditional game theory models that assume perfect rationality and complete information, ShortageSim simulates heterogeneous interpretations on regulatory announcements and the resulting decisions. Experiments on self-processed dataset of historical shortage events show that ShortageSim reduces the resolution lag for production disruption cases by up to 84\%, achieving closer alignment to real-world trajectories than the zero-shot baseline. Our framework confirms the effect of regulatory alert in addressing shortages and introduces a new method for understanding competition in multi-stage environments under uncertainty. We open-source ShortageSim and a dataset of 2,925 FDA shortage events, providing a novel framework for future research on policy design and testing in supply chains under information asymmetry.
Authors: Kai Zhang, Corey D Barrett, Jangwon Kim, Lichao Sun, Tara Taghavi, Krishnaram Kenthapadi
Abstract: Agentic systems offer a potential path to solve complex clinical tasks through collaboration among specialized agents, augmented by tool use and external knowledge bases. Nevertheless, for chest X-ray (CXR) interpretation, prevailing methods remain limited: (i) reasoning is frequently neither clinically interpretable nor aligned with guidelines, reflecting mere aggregation of tool outputs; (ii) multimodal evidence is insufficiently fused, yielding text-only rationales that are not visually grounded; and (iii) systems rarely detect or resolve cross-tool inconsistencies and provide no principled verification mechanisms. To bridge the above gaps, we present RadAgents, a multi-agent framework that couples clinical priors with task-aware multimodal reasoning and encodes a radiologist-style workflow into a modular, auditable pipeline. In addition, we integrate grounding and multimodal retrieval-augmentation to verify and resolve context conflicts, resulting in outputs that are more reliable, transparent, and consistent with clinical practice.
Authors: Yulei Qin, Xiaoyu Tan, Zhengbao He, Gang Li, Haojia Lin, Zongyi Li, Zihan Xu, Yuchen Shi, Siqi Cai, Renting Rui, Shaofei Cai, Yuzheng Cai, Xuan Zhang, Sheng Ye, Ke Li, Xing Sun
Abstract: Reinforcement learning (RL) is the dominant paradigm for sharpening strategic tool use capabilities of LLMs on long-horizon, sparsely-rewarded agent tasks, yet it faces a fundamental challenge of exploration-exploitation trade-off. Existing studies stimulate exploration through the lens of policy entropy, but such mechanical entropy maximization is prone to RL instability due to the multi-turn distribution shifting. In this paper, we target the progressive exploration-exploitation balance under the guidance of the agent's own experiences without succumbing to either entropy collapsing or runaway divergence. We propose SPEAR, a self-imitation learning (SIL) recipe for training agentic LLMs. It extends the vanilla SIL, where a replay buffer stores good experience for off-policy update, by gradually steering the policy entropy across stages. Specifically, the proposed curriculum scheduling harmonizes intrinsic reward shaping and self-imitation to 1) expedite exploration via frequent tool interactions at the beginning, and 2) strengthen exploitation of successful tactics upon convergence towards familiarity with the environment. We also combine bag-of-tricks of industrial RL optimizations for a strong baseline Dr.BoT to demonstrate our effectiveness. In ALFWorld and WebShop, SPEAR increases the success rates of GRPO/GiGPO/Dr.BoT by up to 16.1%/5.1%/8.6% and 20.7%/11.8%/13.9%, respectively. In AIME24 and AIME25, SPEAR boosts Dr.BoT by up to 3.8% and 6.1%, respectively. Such gains incur only 10%-25% extra theoretical complexity and negligible runtime overhead in practice, demonstrating the plug-and-play scalability of SPEAR.
Authors: Tian Guo, Emmanuel Hauptmann
Abstract: In quantitative investing, return prediction supports various tasks, including stock selection, portfolio optimization, and risk management. Quantitative factors, such as valuation, quality, and growth, capture various characteristics of stocks. Unstructured data, like news and transcripts, has attracted growing attention, driven by recent advances in large language models (LLMs). This paper examines effective methods for leveraging multimodal factors and newsflow in return prediction and stock selection. First, we introduce a fusion learning framework to learn a unified representation from factors and newsflow representations generated by an LLM. Within this framework, we compare three methods of different architectural complexities: representation combination, representation summation, and attentive representations. Next, building on the limitation of fusion learning observed in empirical comparison, we explore the mixture model that adaptively combines predictions made by single modalities and their fusion. To mitigate the training instability of the mixture model, we introduce a decoupled training approach with theoretical insights. Finally, our experiments on real investment universes yield several insights into effective multimodal modeling of factors and news for stock return prediction and selection.
Authors: Zhida Zhao, Talas Fu, Yifan Wang, Lijun Wang, Huchuan Lu
Abstract: Despite remarkable progress in driving world models, their potential for autonomous systems remains largely untapped: the world models are mostly learned for world simulation and decoupled from trajectory planning. While recent efforts aim to unify world modeling and planning in a single framework, the synergistic facilitation mechanism of world modeling for planning still requires further exploration. In this work, we introduce a new driving paradigm named Policy World Model (PWM), which not only integrates world modeling and trajectory planning within a unified architecture, but is also able to benefit planning using the learned world knowledge through the proposed action-free future state forecasting scheme. Through collaborative state-action prediction, PWM can mimic the human-like anticipatory perception, yielding more reliable planning performance. To facilitate the efficiency of video forecasting, we further introduce a dynamically enhanced parallel token generation mechanism, equipped with a context-guided tokenizer and an adaptive dynamic focal loss. Despite utilizing only front camera input, our method matches or exceeds state-of-the-art approaches that rely on multi-view and multi-modal inputs. Code and model weights will be released at https://github.com/6550Zhao/Policy-World-Model.
Authors: Faria Huq, Elijah L. Claggett, Hirokazu Shirado
Abstract: Group segregation or cohesion can emerge from micro-level communication, and AI-assisted messaging may shape this process. Here, we report a preregistered online experiment (N = 557 across 60 sessions) in which participants discussed controversial political topics over multiple rounds and could freely change groups. Some participants received real-time message suggestions from a large language model (LLM), either personalized to their stance (individual assistance) or incorporating their group members' perspectives (relational assistance). We find that small variations in AI-mediated communication cascade into macro-level differences in group composition. Participants with individual assistance send more messages and show greater stance-based clustering, whereas those with relational assistance use more receptive language and form more heterogeneous ties. Hybrid expressive processes-jointly produced by humans and AI-can reshape collective organization. The patterns of structural division and cohesion depend on how AI incorporates users' interaction context.
Authors: Andrea Cera Palatsi, Samuel Martin-Gutierrez, Ana S. Cardenal, Max Pellert
Abstract: Large language models (LLMs) are increasingly used both to make decisions in domains such as health, education and law, and to simulate human behavior. Yet how closely LLMs mirror actual human decision-making remains poorly understood. This gap is critical: misalignment could produce harmful outcomes in practical applications, while failure to replicate human behavior renders LLMs ineffective for social simulations. Here, we address this gap by developing a digital twin of game-theoretic experiments and introducing a systematic prompting and probing framework for machine-behavioral evaluation. Testing three open-source models (Llama, Mistral and Qwen), we find that Llama reproduces human cooperation patterns with high fidelity, capturing human deviations from rational choice theory, while Qwen aligns closely with Nash equilibrium predictions. Notably, we achieved population-level behavioral replication without persona-based prompting, simplifying the simulation process. Extending beyond the original human-tested games, we generate and preregister testable hypotheses for novel game configurations outside the original parameter grid. Our findings demonstrate that appropriately calibrated LLMs can replicate aggregate human behavioral patterns and enable systematic exploration of unexplored experimental spaces, offering a complementary approach to traditional research in the social and behavioral sciences that generates new empirical predictions about human social decision-making.
Authors: Eric Xue, Ruiyi Zhang, Zijun Zhang, Pengtao Xie
Abstract: Transformer models are foundational to natural language processing (NLP) applications, yet remain vulnerable to backdoor attacks introduced through poisoned data, which implant hidden behaviors during training. To strengthen the ability to prevent such compromises, recent research has focused on designing increasingly stealthy attacks to stress-test existing defenses, pairing backdoor behaviors with stylized artifact or token-level perturbation triggers. However, this trend diverts attention from the harder and more realistic case: making the model respond to semantic triggers such as specific names or entities, where a successful backdoor could manipulate outputs tied to real people or events in deployed systems. Motivated by this growing disconnect, we introduce SteganoBackdoor, bringing stealth techniques back into line with practical threat models. Leveraging innocuous properties from natural-language steganography, SteganoBackdoor applies a gradient-guided data optimization process to transform semantic trigger seeds into steganographic carriers that embed a high backdoor payload, remain fluent, and exhibit no representational resemblance to the trigger. Across diverse experimental settings, SteganoBackdoor achieves over 99% attack success at an order-of-magnitude lower data-poisoning rate than prior approaches while maintaining unparalleled evasion against a comprehensive suite of data-level defenses. By revealing this practical and covert attack, SteganoBackdoor highlights an urgent blind spot in current defenses and demands immediate attention to adversarial data defenses and real-world threat modeling.
Authors: Jing Bi, Filippos Bellos, Junjia Guo, Yayuan Li, Chao Huang, Yolo Y. Tang, Luchuan Song, Susan Liang, Zhongfei Mark Zhang, Jason J. Corso, Chenliang Xu
Abstract: Test-time thinking (that is, generating explicit intermediate reasoning chains) is known to boost performance in large language models and has recently shown strong gains for large vision language models (LVLMs). However, despite these promising results, there is still no systematic analysis of how thinking actually affects visual reasoning. We provide the first such analysis with a large scale, controlled comparison of thinking for LVLMs, evaluating ten variants from the InternVL3.5 and Qwen3-VL families on MMMU-val under generous token budgets and multi pass decoding. We show that more thinking is not always better; long chains often yield long wrong trajectories that ignore the image and underperform the same models run in standard instruct mode. A deeper analysis reveals that certain short lookback phrases, which explicitly refer back to the image, are strongly enriched in successful trajectories and correlate with better visual grounding. Building on this insight, we propose uncertainty guided lookback, a training free decoding strategy that combines an uncertainty signal with adaptive lookback prompts and breadth search. Our method improves overall MMMU performance, delivers the largest gains in categories where standard thinking is weak, and outperforms several strong decoding baselines, setting a new state of the art under fixed model families and token budgets. We further show that this decoding strategy generalizes, yielding consistent improvements on five additional benchmarks, including two broad multimodal suites and math focused visual reasoning datasets.
Authors: Myung Ho Kim
Abstract: Large language model agents suffer from fundamental architectural problems: entangled reasoning and execution, memory volatility, and uncontrolled action sequences. We introduce Structured Cognitive Loop (SCL), a modular architecture that explicitly separates agent cognition into five phases: Retrieval, Cognition, Control, Action, and Memory (R-CCAM). At the core of SCL is Soft Symbolic Control, an adaptive governance mechanism that applies symbolic constraints to probabilistic inference, preserving neural flexibility while restoring the explainability and controllability of classical symbolic systems. Through empirical validation on multi-step conditional reasoning tasks, we demonstrate that SCL achieves zero policy violations, eliminates redundant tool calls, and maintains complete decision traceability. These results address critical gaps in existing frameworks such as ReAct, AutoGPT, and memory-augmented approaches. Our contributions are threefold: (1) we situate SCL within the taxonomy of hybrid intelligence, differentiating it from prompt-centric and memory-only approaches; (2) we formally define Soft Symbolic Control and contrast it with neuro-symbolic AI; and (3) we derive three design principles for trustworthy agents: modular decomposition, adaptive symbolic governance, and transparent state management. We provide a complete open-source implementation demonstrating the R-CCAM loop architecture, alongside a live GPT-4o-powered travel planning agent. By connecting expert system principles with modern LLM capabilities, this work offers a practical and theoretically grounded path toward reliable, explainable, and governable AI agents.
Authors: Nikita Neveditsin, Pawan Lingras, Vijay Mago
Abstract: Clustering short text embeddings is a foundational task in natural language processing, yet remains challenging due to the need to specify the number of clusters in advance. We introduce a scalable spectral method that estimates the number of clusters directly from the structure of the Laplacian eigenspectrum, constructed using cosine similarities and guided by an adaptive sampling strategy. This sampling approach enables our estimator to efficiently scale to large datasets without sacrificing reliability. To support intrinsic evaluation of cluster quality without ground-truth labels, we propose the Cohesion Ratio, a simple and interpretable evaluation metric that quantifies how much intra-cluster similarity exceeds the global similarity background. It has an information-theoretic motivation inspired by mutual information, and in our experiments it correlates closely with extrinsic measures such as normalized mutual information and homogeneity. Extensive experiments on six short-text datasets and four modern embedding models show that standard algorithms like K-Means and HAC, when guided by our estimator, significantly outperform popular parameter-light methods such as HDBSCAN, OPTICS, and Leiden. These results demonstrate the practical value of our spectral estimator and Cohesion Ratio for unsupervised organization and evaluation of short text data. Implementation of our estimator of k and Cohesion Ratio, along with code for reproducing the experiments, is available at https://anonymous.4open.science/r/towards_clustering-0C2E.
URLs: https://anonymous.4open.science/r/towards_clustering-0C2E.