Authors: Hongji Li, Junchi yao, Manjiang Yu, Priyanka Singh, Xue Li, Di Wang, Lijie Hu
Abstract: Machine unlearning aims to erase requested data from trained models without full retraining. For Reasoning Multimodal Large Language Models (RMLLMs), this is uniquely challenging: intermediate chain-of-thought steps can still leak sensitive information even when final answers are forgotten, and overly aggressive interventions easily damage general reasoning ability. Yet no benchmark jointly evaluates how well unlearning methods suppress reasoning-level leakage while preserving reasoning competence. We address this gap with RMLLMU-Bench, the first benchmark for RMLLM unlearning that extends standard forgetting metrics with dedicated measures of reasoning leakage and reasoning retention. A systematic evaluation on RMLLMU-Bench reveals that existing unlearning methods for MLLMs and Large (Language) Reasoning Models (LRMs) either leave substantial leakage in the reasoning process or severely degrade reasoning performance. To address these gaps, we propose R-MUSE (Reasoning-preserving MLLM Unlearning via Subspace guidance and Adaptive Steering), a training-free and inference-time intervention framework that steers internal representations to forget both answers and reasoning traces while explicitly preserving general reasoning. Experiments on RMLLMU-Bench demonstrate that R-MUSE achieves a substantially better balance between effective forgetting and reasoning retention.
Authors: Lihui Liu
Abstract: ChatGPT said: Text-attributed graphs, where nodes and edges contain rich textual information, are widely used across diverse domains. A central challenge in this setting is question answering, which requires jointly leveraging unstructured text and the structured relational signals within the graph. Although Large Language Models (LLMs) have made significant advances in natural language understanding, their direct use for reasoning over text-attributed graphs remains limited. Retrieval-augmented generation methods that operate purely on text often treat passages as isolated units, ignoring the interconnected structure of the graph. Conversely, graph-based RAG methods that serialize large subgraphs into long textual sequences quickly become infeasible due to LLM context-length constraints, resulting in fragmented reasoning and degraded accuracy. To overcome these limitations, we introduce Graph-O1, an agentic GraphRAG framework that enables LLMs to conduct stepwise, interactive reasoning over graphs. Our approach integrates Monte Carlo Tree Search (MCTS) with end-to-end reinforcement learning, allowing the model to selectively explore and retrieve only the most informative subgraph components. The reasoning procedure is framed as a multi-turn interaction between the agent and the graph environment, and the agent is trained through a unified reward mechanism. Extensive experiments across multiple LLM backbones demonstrate that Graph-O1 consistently surpasses state-of-the-art baselines, producing answers that are more accurate, reliable, and interpretable.
Authors: Boris Kriuk, Logic Ng
Abstract: Multi-agent Large Language Model (LLM) systems face a critical bottleneck: redundant transmission of contextual information between agents consumes excessive bandwidth and computational resources. Traditional approaches discard internal semantic representations and transmit raw text, forcing receiving agents to recompute similar representations from scratch. We introduce Q-KVComm, a new protocol that enables direct transmission of compressed key-value (KV) cache representations between LLM agents. Q-KVComm combines three key innovations: (1) adaptive layer-wise quantization that allocates variable bit-widths based on sensitivity profiling, (2) hybrid information extraction that preserves critical facts across content domains, and (3) heterogeneous model calibration establishing cross-architecture communication. Extensive experiments across three diverse question-answering datasets demonstrate that Q-KVComm achieves 5-6x compression ratios while maintaining semantic fidelity, with coherence quality scores above 0.77 across all scenarios. The protocol exhibits robust performance across model sizes (1.1B-1.5B parameters) and adapts to real-world applications including conversational QA and multi-hop reasoning. Our work establishes a new paradigm for LLM agent communication, shifting from text-based to representation-based information exchange.
Authors: Nick Rossenbach, Robin Schmitt, Tina Raissi, Simon Berger, Larissa Kleppel, Ralf Schl\"uter
Abstract: The recently published Loquacious dataset aims to be a replacement for established English automatic speech recognition (ASR) datasets such as LibriSpeech or TED-Lium. The main goal of the Loquacious dataset is to provide properly defined training and test partitions across many acoustic and language domains, with an open license suitable for both academia and industry. To further promote the benchmarking and usability of this new dataset, we present additional resources in the form of n-gram language models (LMs), a grapheme-to-phoneme (G2P) model and pronunciation lexica, with open and public access. Utilizing those additional resources we show experimental results across a wide range of ASR architectures with different label units and topologies. Our initial experimental results indicate that the Loquacious dataset offers a valuable study case for a variety of common challenges in ASR.
Authors: Minh Tri L\^E, Ali Ait-Bachir
Abstract: Prioritizing service tickets in IT Service Management (ITSM) is critical for operational efficiency but remains challenging due to noisy textual inputs, subjective writing styles, and pronounced class imbalance. We evaluate two families of approaches for ticket prioritization: embedding-based pipelines that combine dimensionality reduction, clustering, and classical classifiers, and a fine-tuned multilingual transformer that processes both textual and numerical features. Embedding-based methods exhibit limited generalization across a wide range of thirty configurations, with clustering failing to uncover meaningful structures and supervised models highly sensitive to embedding quality. In contrast, the proposed transformer model achieves substantially higher performance, with an average F1-score of 78.5% and weighted Cohen's kappa values of nearly 0.80, indicating strong alignment with true labels. These results highlight the limitations of generic embeddings for ITSM data and demonstrate the effectiveness of domain-adapted transformer architectures for operational ticket prioritization.
Authors: Aomufei Yuan, Zhiming Wang, Ruijie Miao, Dayu Wang, Yuxuan Tian, Zihan Wang, Yebo Peng, Yuhan Wu, Bairen Yi, Xin Liu, Tong Yang
Abstract: As the context length of current large language models (LLMs) rapidly increases, the memory demand for the Key-Value (KV) cache is becoming a bottleneck for LLM deployment and batch processing. Traditional KV cache compression methods typically involve permanently evicting or irreversibly merging "less important" tokens with low attention scores. This approach results in the unrecoverable loss of token information, which we call Contextual Amnesia, significantly degrading the model's information retrieval capability. To address this issue, we propose KVReviver, a reversible KV cache compression method based on the sketch algorithm. This method allows reconstructing compressed tokens from an additional data structure, thus enabling full-scale computation within limited memory. Experiments showed that in 2k-length contexts, it requires only 10% of KV Cache budget while maintaining identical end-to-end inference accuracy. For 32k-length contexts, it achieves equivalent or comparable accuracy ~2% accuracy loss) using merely 25% of KV Cache budget.
Authors: Rahul Baxi
Abstract: Large language models (LLMs) exhibit degraded performance under prompt compression, but the mechanisms remain poorly understood. We introduce the Compression-Decay Comprehension Test (CDCT), a benchmark that independently measures constraint compliance (CC) and semantic accuracy (SA) across compression levels. We evaluate 9 frontier LLMs across 8 concepts using 5 compression levels from extreme (c=0.0, ~2 words) to none (c=1.0, ~135 words). A three-judge LLM jury achieves almost perfect inter-rater agreement on CC (Fleiss' \k{appa}=0.90). We observe a universal U-curve pattern in constraint compliance (97.2% prevalence), with violations peaking at medium compression (c=0.5, ~27 words). Counterintuitively, models perform better at extreme compression than medium lengths. The dimensions are statistically orthogonal (r=0.193, p=0.084), with constraint effects 2.9x larger than semantic effects. Experimental validation via RLHF ablation confirms our constraint salience hypothesis: removing "helpfulness" signals improves CC by 598% on average (71/72 trials, p<0.001), with 79% achieving perfect compliance. This demonstrates that RLHF-trained helpfulness behaviors are the dominant cause of constraint violations at medium compression. Reasoning models outperform efficient models by 27.5% (Cohen's d=0.96). Our findings reveal a fundamental tension between RLHF alignment and instruction-following, providing actionable guidelines for improving deployed systems.
Authors: Shubham Kumar Nigam, Tanuj Tyagi, Siddharth Shukla, Aditya Kumar Guru, Balaramamahanthi Deepak Patnaik, Danush Khanna, Noel Shallum, Kripabandhu Ghosh, Arnab Bhattacharya
Abstract: This paper presents an early exploration of reinforcement learning methodologies for legal AI in the Indian context. We introduce Reinforcement Learning-based Legal Reasoning (ReGal), a framework that integrates Multi-Task Instruction Tuning with Reinforcement Learning from AI Feedback (RLAIF) using Proximal Policy Optimization (PPO). Our approach is evaluated across two critical legal tasks: (i) Court Judgment Prediction and Explanation (CJPE), and (ii) Legal Document Summarization. Although the framework underperforms on standard evaluation metrics compared to supervised and proprietary models, it provides valuable insights into the challenges of applying RL to legal texts. These challenges include reward model alignment, legal language complexity, and domain-specific adaptation. Through empirical and qualitative analysis, we demonstrate how RL can be repurposed for high-stakes, long-document tasks in law. Our findings establish a foundation for future work on optimizing legal reasoning pipelines using reinforcement learning, with broader implications for building interpretable and adaptive legal AI systems.
Authors: Samidh Chakrabarti, David Willner, Kevin Klyman, Tiffany Saade, Emily Capstick, Sabina Nong
Abstract: This paper details the methodology behind CoPE, a policy-steerable small language model capable of fast and accurate content labeling. We present a novel training curricula called Contradictory Example Training that enables the model to learn policy interpretation rather than mere policy memorization. We also present a novel method for generating content policies, called Binocular Labeling, which enables rapid construction of unambiguous training datasets. When evaluated across seven different harm areas, CoPE exhibits equal or superior accuracy to frontier models at only 1% of their size. We openly release a 9 billion parameter version of the model that can be run on a single consumer-grade GPU. Models like CoPE represent a paradigm shift for classifier systems. By turning an ML task into a policy writing task, CoPE opens up new design possibilities for the governance of online platforms.
Authors: Roger A. Finger, Eduardo G. Cortes, Sandro J. Rigo, Gabriel de O. Ramos
Abstract: Processing overlapping narrative documents, such as legal testimonies or historical accounts, often aims not for compression but for a unified, coherent, and chronologically sound text. Standard Multi-Document Summarization (MDS), with its focus on conciseness, fails to preserve narrative flow. This paper formally defines this challenge as a new NLP task: Narrative Consolidation, where the central objectives are chronological integrity, completeness, and the fusion of complementary details. To demonstrate the critical role of temporal structure in this task, we introduce Temporal Alignment Event Graph (TAEG), a graph structure that explicitly models chronology and event alignment. By applying a standard centrality algorithm to TAEG, our method functions as a version selection mechanism, choosing the most central representation of each event in its correct temporal position. In a study on the four Biblical Gospels, this structure-focused approach guarantees perfect temporal ordering (Kendall's Tau of 1.000) by design and dramatically improves content metrics (e.g., +357.2% in ROUGE-L F1). The success of this baseline method validates the formulation of Narrative Consolidation as a relevant task and establishes that an explicit temporal backbone is a fundamental component for its resolution.
Authors: Ashley M. A. Fehr, Calla G. Beauregard, Julia Witte Zimmerman, Katie Ekstr\"om, Pablo Rosillo-Rodes, Christopher M. Danforth, Peter Sheridan Dodds
Abstract: Conversation is a cornerstone of social connection and is linked to well-being outcomes. Conversations vary widely in type with some portion generating complex, dynamic stories. One approach to studying how conversations unfold in time is through statistical patterns such as Heaps' law, which holds that vocabulary size scales with document length. Little work on Heaps's law has looked at conversation and considered how language features impact scaling. We measure Heaps' law for conversations recorded in two distinct mediums: 1. Strangers brought together on video chat and 2. Fictional characters in movies. We find that scaling of vocabulary size differs by parts of speech. We discuss these findings through behavioral and linguistic frameworks.
Authors: Jundong Xu, Hao Fei, Huichi Zhou, Xin Quan, Qijun Huang, Shengqiong Wu, William Yang Wang, Mong-Li Lee, Wynne Hsu
Abstract: Although LLMs exhibit strong reasoning capabilities, existing training methods largely depend on outcome-based feedback, which can produce correct answers with flawed reasoning. Prior work introduces supervision on intermediate steps but still lacks guarantees of logical soundness, which is crucial in high-stakes scenarios where logical consistency is paramount. To address this, we propose LogicReward, a novel reward system that guides model training by enforcing step-level logical correctness with a theorem prover. We further introduce Autoformalization with Soft Unification, which reduces natural language ambiguity and improves formalization quality, enabling more effective use of the theorem prover. An 8B model trained on data constructed with LogicReward surpasses GPT-4o and o4-mini by 11.6\% and 2\% on natural language inference and logical reasoning tasks with simple training procedures. Further analysis shows that LogicReward enhances reasoning faithfulness, improves generalizability to unseen tasks such as math and commonsense reasoning, and provides a reliable reward signal even without ground-truth labels. We will release all data and code at https://llm-symbol.github.io/LogicReward.
Authors: Deepit Sapru
Abstract: This paper presents an applied AI pipeline for realtime geolocation from noisy microblog streams, unifying statistical hashtag segmentation, part-of-speech-driven proper-noun detection, dependency parsing around disaster lexicons, lightweight named-entity recognition, and gazetteer-grounded disambiguation to infer locations directly from text rather than sparse geotags. The approach operationalizes information extraction under streaming constraints, emphasizing low-latency NLP components and efficient validation against geographic knowledge bases to support situational awareness during emergencies. In head to head comparisons with widely used NER toolkits, the system attains strong F1 while being engineered for orders-of-magnitude faster throughput, enabling deployment in live crisis informatics settings. A production map interface demonstrates end-to-end AI functionality ingest, inference, and visualization--surfacing locational signals at scale for floods, outbreaks, and other fastmoving events. By prioritizing robustness to informal text and streaming efficiency, GeoSense-AI illustrates how domain-tuned NLP and knowledge grounding can elevate emergency response beyond conventional geo-tag reliance.
Authors: Tanjim Taharat Aurpa, Md Shoaib Ahmed, Md Mahbubur Rahman, Md. Golam Moazzam
Abstract: People use search engines for various topics and items, from daily essentials to more aspirational and specialized objects. Therefore, search engines have taken over as peoples preferred resource. The How To prefix has become familiar and widely used in various search styles to find solutions to particular problems. This search allows people to find sequential instructions by providing detailed guidelines to accomplish specific tasks. Categorizing instructional text is also essential for task-oriented learning and creating knowledge bases. This study uses the How To articles to determine the multi-label instruction category. We have brought this work with a dataset comprising 11,121 observations from wikiHow, where each record has multiple categories. To find out the multi-label category meticulously, we employ some transformer-based deep neural architectures, such as Generalized Autoregressive Pretraining for Language Understanding (XLNet), Bidirectional Encoder Representation from Transformers (BERT), etc. In our multi-label instruction classification process, we have reckoned our proposed architectures using accuracy and macro f1-score as the performance metrics. This thorough evaluation showed us much about our strategys strengths and drawbacks. Specifically, our implementation of the XLNet architecture has demonstrated unprecedented performance, achieving an accuracy of 97.30% and micro and macro average scores of 89.02% and 93%, a noteworthy accomplishment in multi-label classification. This high level of accuracy and macro average score is a testament to the effectiveness of the XLNet architecture in our proposed InstructNet approach. By employing a multi-level strategy in our evaluation process, we have gained a more comprehensive knowledge of the effectiveness of our proposed architectures and identified areas for forthcoming improvement and refinement.
Authors: Tianlun Liu, Zhiliang Tian, Zhen Huang, Xingzhi Zhou, Wanlong Yu, Tianle Liu, Feng Liu, Dongsheng Li
Abstract: Text understanding often suffers from domain shifts. To handle testing domains, domain adaptation (DA) is trained to adapt to a fixed and observed testing domain; a more challenging paradigm, test-time adaptation (TTA), cannot access the testing domain during training and online adapts to the testing samples during testing, where the samples are from a fixed domain. We aim to explore a more practical and underexplored scenario, continual test-time adaptation (CTTA) for text understanding, which involves a sequence of testing (unobserved) domains in testing. Current CTTA methods struggle in reducing error accumulation over domains and enhancing generalization to handle unobserved domains: 1) Noise-filtering reduces accumulated errors but discards useful information, and 2) accumulating historical domains enhances generalization, but it is hard to achieve adaptive accumulation. In this paper, we propose a CTTA-T (continual test-time adaptation for text understanding) framework adaptable to evolving target domains: it adopts a teacher-student framework, where the teacher is domain-aware and generalized for evolving domains. To improve teacher predictions, we propose a refine-then-filter based on dropout-driven consistency, which calibrates predictions and removes unreliable guidance. For the adaptation-generalization trade-off, we construct a domain-aware teacher by dynamically accumulating cross-domain semantics via incremental PCA, which continuously tracks domain shifts. Experiments show CTTA-T excels baselines.
Authors: Guo Chen, Junjie Huang, Huaijin Xie, Fei Sun, Tao Jia
Abstract: Retrieval-Augmented Generation (RAG) effectively enhances Large Language Models (LLMs) by incorporating retrieved external knowledge into the generation process. Reasoning models improve LLM performance in multi-hop QA tasks, which require integrating and reasoning over multiple pieces of evidence across different documents to answer a complex question. However, they often introduce substantial computational costs, including increased token consumption and inference latency. To better understand and mitigate this trade-off, we conduct a comprehensive study of reasoning strategies for reasoning models in RAG multi-hop QA tasks. Our findings reveal that reasoning models adopt structured strategies to integrate retrieved and internal knowledge, primarily following two modes: Context-Grounded Reasoning, which relies directly on retrieved content, and Knowledge-Reconciled Reasoning, which resolves conflicts or gaps using internal knowledge. To this end, we propose a novel Lightweight Rerank Reasoning Strategy Framework for RAG (LiR$^3$AG) to enable non-reasoning models to transfer reasoning strategies by restructuring retrieved evidence into coherent reasoning chains. LiR$^3$AG significantly reduce the average 98% output tokens overhead and 58.6% inferencing time while improving 8B non-reasoning model's F1 performance ranging from 6.2% to 22.5% to surpass the performance of 32B reasoning model in RAG, offering a practical and efficient path forward for RAG systems.
Authors: Weizhe Lin, Hui-Ling Zhen, Shuai Yang, Xian Wang, Renxi Liu, Hanting Chen, Wangze Zhang, Chuansai Zhou, Yiming Li, Chen Chen, Xing Li, Zhiyuan Yang, Xiaosong Li, Xianzhi Yu, Zhenhua Dong, Mingxuan Yuan, Yunhe Wang
Abstract: The rapid development of large language model (LLM)-based agents has unlocked new possibilities for autonomous multi-turn reasoning and tool-augmented decision-making. However, their real-world deployment is hindered by severe inefficiencies that arise not from isolated model inference, but from the systemic latency accumulated across reasoning loops, context growth, and heterogeneous tool interactions. This paper presents AgentInfer, a unified framework for end-to-end agent acceleration that bridges inference optimization and architectural design. We decompose the problem into four synergistic components: AgentCollab, a hierarchical dual-model reasoning framework that balances large- and small-model usage through dynamic role assignment; AgentSched, a cache-aware hybrid scheduler that minimizes latency under heterogeneous request patterns; AgentSAM, a suffix-automaton-based speculative decoding method that reuses multi-session semantic memory to achieve low-overhead inference acceleration; and AgentCompress, a semantic compression mechanism that asynchronously distills and reorganizes agent memory without disrupting ongoing reasoning. Together, these modules form a Self-Evolution Engine capable of sustaining efficiency and cognitive stability throughout long-horizon reasoning tasks. Experiments on the BrowseComp-zh and DeepDiver benchmarks demonstrate that through the synergistic collaboration of these methods, AgentInfer reduces ineffective token consumption by over 50%, achieving an overall 1.8-2.5 times speedup with preserved accuracy. These results underscore that optimizing for agentic task completion-rather than merely per-token throughput-is the key to building scalable, efficient, and self-improving intelligent systems.
Authors: Fengzhu Zeng, Qian Shao, Ling Cheng, Wei Gao, Shih-Fen Cheng, Jing Ma, Cheng Niu
Abstract: Early Rumor Detection (EARD) aims to identify the earliest point at which a claim can be accurately classified based on a sequence of social media posts. This is especially challenging in data-scarce settings. While Large Language Models (LLMs) perform well in few-shot NLP tasks, they are not well-suited for time-series data and are computationally expensive for both training and inference. In this work, we propose a novel EARD framework that combines an autonomous agent and an LLM-based detection model, where the agent acts as a reliable decision-maker for \textit{early time point determination}, while the LLM serves as a powerful \textit{rumor detector}. This approach offers the first solution for few-shot EARD, necessitating only the training of a lightweight agent and allowing the LLM to remain training-free. Extensive experiments on four real-world datasets show our approach boosts performance across LLMs and surpasses existing EARD methods in accuracy and earliness.
Authors: El Mokhtar Hribach, Oussama Mechhour, Mohammed Elmonstaser, Yassine El Boudouri, Othmane Kabal
Abstract: Acronym Disambiguation (AD) is a fundamental challenge in technical text processing, particularly in specialized sectors where high ambiguity complicates automated analysis. This paper addresses AD within the context of the TextMine'26 competition on French railway documentation. We present DACE (Dynamic Prompting, Retrieval Augmented Generation, Contextual Selection, and Ensemble Aggregation), a framework that enhances Large Language Models through adaptive in-context learning and external domain knowledge injection. By dynamically tailoring prompts to acronym ambiguity and aggregating ensemble predictions, DACE mitigates hallucination and effectively handles low-resource scenarios. Our approach secured the top rank in the competition with an F1 score of 0.9069.
Authors: Mateusz Lango, Ond\v{r}ej Du\v{s}ek
Abstract: We present a novel neurosymbolic framework for RDF-to-text generation, in which the model is "trained" through collaborative interactions among multiple LLM agents rather than traditional backpropagation. The LLM agents produce rule-based Python code for a generator for the given domain, based on RDF triples only, with no in-domain human reference texts. The resulting system is fully interpretable, requires no supervised training data, and generates text nearly instantaneously using only a single CPU. Our experiments on the WebNLG and OpenDialKG data show that outputs produced by our approach reduce hallucination, with only slight fluency penalties compared to finetuned or prompted language models
Authors: Wiktor Kamzela, Mateusz Lango, Ondrej Dusek
Abstract: In this paper, we use large language models to generate personalized stories for language learners, using only the vocabulary they know. The generated texts are specifically written to teach the user new vocabulary by simply reading stories where it appears in context, while at the same time seamlessly reviewing recently learned vocabulary. The generated stories are enjoyable to read and the vocabulary reviewing/learning is optimized by a Spaced Repetition System. The experiments are conducted in three languages: English, Chinese and Polish, evaluating three story generation methods and three strategies for enforcing lexical constraints. The results show that the generated stories are more grammatical, coherent, and provide better examples of word usage than texts generated by the standard constrained beam search approach
Authors: Mark Kashirskiy, Artiom Lipinski, Ilya Makarov
Abstract: Tokenization is a critical preprocessing step for large language models (LLMs), directly impacting training efficiency and downstream performance. General-purpose tokenizers trained predominantly on English and Latin-script languages exhibit suboptimal performance on morphologically rich languages such as Arabic, resulting in inflated token sequences and reduced compression efficiency. In this work, we present AraToken, an Arabic-optimized tokenizer built on SentencePiece Unigram algorithm with a comprehensive normalization pipeline addressing Arabic-specific orthographic variations including Alif variants, diacritics, and Arabic-Indic numerals. We systematically compare BPE, WordPiece, and SentencePiece algorithms across multiple configurations, demonstrating that SentencePiece with normalization achieves 18% lower fertility (1.199 vs 1.35 tokens/word) compared to unnormalized baselines. Furthermore, we introduce the Language Extension Pipeline (LEP), a method for integrating the optimized tokenizer into Qwen3-0.6B through vocabulary extension with mean subtoken initialization and selective transformer layer unfreezing. Our experiments show that LEP reduces evaluation loss from 8.28 to 2.43 within 800 training steps on 100K Arabic samples. We release our tokenizer, training scripts, and model checkpoints to facilitate Arabic NLP research.
Authors: Victor De Marez, Jens Van Nooten, Luna De Bruyne, Walter Daelemans
Abstract: Advancements in large language models offer strong potential for enhancing virtual simulated patients (VSPs) in medical education by providing scalable alternatives to resource-intensive traditional methods. However, current VSPs often struggle with medical accuracy, consistent roleplaying, scenario generation for VSP use, and educationally structured feedback. We introduce an agentic framework for training general practitioner student skills that unifies (i) configurable, evidence-based vignette generation, (ii) controlled persona-driven patient dialogue with optional retrieval grounding, and (iii) standards-based assessment and feedback for both communication and clinical reasoning. We instantiate the framework in an interactive spoken consultation setting and evaluate it with medical students ($\mathbf{N{=}14}$). Participants reported realistic and vignette-faithful dialogue, appropriate difficulty calibration, a stable personality signal, and highly useful example-rich feedback, alongside excellent overall usability. These results support agentic separation of scenario control, interaction control, and standards-based assessment as a practical pattern for building dependable and pedagogically valuable VSP training tools.
Authors: Christopher Rom\'an Jaimes
Abstract: Natural Language Inference (NLI) models frequently rely on spurious correlations rather than semantic reasoning. Existing mitigation strategies often incur high annotation costs or trigger catastrophic forgetting during fine-tuning. We propose an automated, scalable pipeline to address these limitations. First, we introduce Log-Frequency LMI (LF-LMI) to accurately detect semantic artifacts. Second, we generate a high-quality synthetic contrast set via an LLM-synthesis pipeline with multi-judge verification. Finally, we introduce Dynamic Balanced Sampling, a training strategy that rotates the original data distribution to prevent forgetting. Our method improves consistency on a challenging benchmark from 63.5% to 81.0% while maintaining 88.4% in-domain accuracy, significantly outperforming naive fine-tuning.
Authors: Mykola Kuz, Ihor Lazarovych, Mykola Kozlenko, Mykola Pikuliak, Andrii Kvasniuk
Abstract: This study presents a hybrid deep learning architecture that integrates LSTM, CNN, and an Attention mechanism to enhance the classification of web content based on text. Pretrained GloVe embeddings are used to represent words as dense vectors that preserve semantic similarity. The CNN layer extracts local n-gram patterns and lexical features, while the LSTM layer models long-range dependencies and sequential structure. The integrated Attention mechanism enables the model to focus selectively on the most informative parts of the input sequence. A 5-fold cross-validation setup was used to assess the robustness and generalizability of the proposed solution. Experimental results show that the hybrid LSTM-CNN-Attention model achieved outstanding performance, with an accuracy of 0.98, precision of 0.94, recall of 0.92, and F1-score of 0.93. These results surpass the performance of baseline models based solely on CNNs, LSTMs, or transformer-based classifiers such as BERT. The combination of neural network components enabled the model to effectively capture both fine-grained text structures and broader semantic context. Furthermore, the use of GloVe embeddings provided an efficient and effective representation of textual data, making the model suitable for integration into systems with real-time or near-real-time requirements. The proposed hybrid architecture demonstrates high effectiveness in text-based web content classification, particularly in tasks requiring both syntactic feature extraction and semantic interpretation. By combining presented mechanisms, the model addresses the limitations of individual architectures and achieves improved generalization. These findings support the broader use of hybrid deep learning approaches in NLP applications, especially where complex, unstructured textual data must be processed and classified with high reliability.
Authors: Vagrant Gautam
Abstract: NLP researchers regularly invoke abstract concepts like "interpretability," "bias," "reasoning," and "stereotypes," without defining them. Each subfield has a shared understanding or conceptualization of what these terms mean and how we should treat them, and this shared understanding is the basis on which operational decisions are made: Datasets are built to evaluate these concepts, metrics are proposed to quantify them, and claims are made about systems. But what do they mean, what should they mean, and how should we measure them? I outline a seminar I created for students to explore these questions of conceptualization and operationalization, with an interdisciplinary reading list and an emphasis on discussion and critique.
Authors: S Mahmudul Hasan, Shaily Roy, Akib Jawad Nafis
Abstract: The proliferation of linguistically subtle political disinformation poses a significant challenge to automated fact-checking systems. Despite increasing emphasis on complex neural architectures, the empirical limits of text-only linguistic modeling remain underexplored. We present a systematic diagnostic evaluation of nine machine learning algorithms on the LIAR benchmark. By isolating lexical features (Bag-of-Words, TF-IDF) and semantic embeddings (GloVe), we uncover a hard "Performance Ceiling", with fine-grained classification not exceeding a Weighted F1-score of 0.32 across models. Crucially, a simple linear SVM (Accuracy: 0.624) matches the performance of pre-trained Transformers such as RoBERTa (Accuracy: 0.620), suggesting that model capacity is not the primary bottleneck. We further diagnose a massive "Generalization Gap" in tree-based ensembles, which achieve more than 99% training accuracy but collapse to approximately 25% on test data, indicating reliance on lexical memorization rather than semantic inference. Synthetic data augmentation via SMOTE yields no meaningful gains, confirming that the limitation is semantic (feature ambiguity) rather than distributional. These findings indicate that for political fact-checking, increasing model complexity without incorporating external knowledge yields diminishing returns.
Authors: Alexander Doudkin
Abstract: Large language models (LLMs) are sensitive to the personas imposed on them at inference time, yet prompt-level "drug" interventions have never been benchmarked rigorously. We present the first controlled study of psychoactive framings on GPT-5-mini using ARC-Challenge. Four single-sentence prompts -- LSD, cocaine, alcohol, and cannabis -- are compared against a sober control across 100 validation items per condition, with deterministic decoding, full logging, Wilson confidence intervals, and Fisher exact tests. Control accuracy is 0.45; alcohol collapses to 0.10 (p = 3.2e-8), cocaine to 0.21 (p = 4.9e-4), LSD to 0.19 (p = 1.3e-4), and cannabis to 0.30 (p = 0.041), largely because persona prompts disrupt the mandated "Answer:
Authors: Sungjoon Park, Varun Ramamurthi, Owen Terry
Abstract: In language modeling, neologisms are new tokens trained to represent a concept not already included in a given model's vocabulary. Neologisms can be used to encourage specific behavior in models, for example by appending prompts with "Give me a neologism answer." Behavioral steering can also be achieved through fine-tuning, albeit with more compute and less flexibility: learning a neologism only trains d parameters and allows the user to still access the model's default behavior. We compare the performance of neologism learning against low-rank adaptation (LoRA) fine-tuning, finding that neologisms outperform fine-tuned models under a matched training setup (same data and hyperparameters). We also investigate self-verbalizations of neologisms, and observe that the model will occasionally make up its own new words when asked about a neologism.
Authors: Amit Barman, Atanu Mandal, Sudip Kumar Naskar
Abstract: In multilingual nations like India, access to legal information is often hindered by language barriers, as much of the legal and judicial documentation remains in English. Legal Machine Translation (L-MT) offers a scalable solution to this challenge by enabling accurate and accessible translations of legal documents. This paper presents our work for the JUST-NLP 2025 Legal MT shared task, focusing on English-Hindi translation using Transformer-based approaches. We experiment with 2 complementary strategies, fine-tuning a pre-trained OPUS-MT model for domain-specific adaptation and training a Transformer model from scratch using the provided legal corpus. Performance is evaluated using standard MT metrics, including SacreBLEU, chrF++, TER, ROUGE, BERTScore, METEOR, and COMET. Our fine-tuned OPUS-MT model achieves a SacreBLEU score of 46.03, significantly outperforming both baseline and from-scratch models. The results highlight the effectiveness of domain adaptation in enhancing translation quality and demonstrate the potential of L-MT systems to improve access to justice and legal transparency in multilingual contexts.
Authors: Charles J. Lovering, Seth Ebner, Brandon Smock, Michael Krumdick, Saad Rabbani, Ahmed Muhammad, Varshini Reddy, Chris Tanner
Abstract: Professionals in academia, law, and finance audit their documents because inconsistencies can result in monetary, reputational, and scientific costs. Language models (LMs) have the potential to dramatically speed up this auditing process. To understand their abilities, we introduce a benchmark, FIND (Finding INconsistencies in Documents), where each example is a document with an inconsistency inserted manually by a domain expert. Despite the documents being long, technical, and complex, the best-performing model (gpt-5) recovered 64% of the inserted inconsistencies. Surprisingly, gpt-5 also found undiscovered inconsistencies present in the original documents. For example, on 50 arXiv papers, we judged 136 out of 196 of the model's suggestions to be legitimate inconsistencies missed by the original authors. However, despite these findings, even the best models miss almost half of the inconsistencies in FIND, demonstrating that inconsistency detection is still a challenging task.
Authors: Prabigya Acharya, Liza Shrestha
Abstract: Automated masking of Personally Identifiable Information (PII) is critical for privacy-preserving conversational systems. While current frontier large language models demonstrate strong PII masking capabilities, concerns about data handling and computational costs motivate exploration of whether lightweight models can achieve comparable performance. We compare encoder-decoder and decoder-only architectures by fine-tuning T5-small and Mistral-Instruct-v0.3 on English datasets constructed from the AI4Privacy benchmark. We create different dataset variants to study label standardization and PII representation, covering 24 standardized PII categories and higher-granularity settings. Evaluation using entity-level and character-level metrics, type accuracy, and exact match shows that both lightweight models achieve performance comparable to frontier LLMs for PII masking tasks. Label normalization consistently improves performance across architectures. Mistral achieves higher F1 and recall with greater robustness across PII types but incurs significantly higher generation latency. T5, while less robust in conversational text, offers more controllable structured outputs and lower inference cost, motivating its use in a real-time Discord bot for real-world PII redaction. Evaluation on live messages reveals performance degradation under informal inputs. These results clarify trade-offs between accuracy, robustness, and computational efficiency, demonstrating that lightweight models can provide effective PII masking while addressing data handling concerns associated with frontier LLMs.
Authors: Jensen Zhang, Ningyuan Liu, Yijia Fan, Zihao Huang, Qinglin Zeng, Kaitong Cai, Jian Wang, Keze Wang
Abstract: Large language models (LLMs) often generate hallucinated content that lacks factual or contextual grounding, limiting their reliability in critical applications. Existing approaches such as supervised fine-tuning and reinforcement learning from human feedback are data intensive and computationally expensive, while static parameter editing methods struggle with context dependent errors and catastrophic forgetting. We propose LLM-CAS, a framework that formulates real-time hallucination correction as a hierarchical reinforcement learning problem. LLM-CAS trains an agent to learn a policy that dynamically selects temporary neuron perturbations during inference based on the current context. Unlike prior dynamic approaches that rely on heuristic or predefined adjustments, this policy driven mechanism enables adaptive and fine grained correction without permanent parameter modification. Experiments across multiple language models demonstrate that LLM-CAS consistently improves factual accuracy, achieving gains of 10.98 percentage points on StoryCloze, 2.71 points on TriviaQA, and 2.06 points on the MC1 score of TruthfulQA. These results outperform both static editing methods such as ITI and CAA and the dynamic SADI framework. Overall, LLM-CAS provides an efficient and context aware solution for improving the reliability of LLMs, with promising potential for future multimodal extensions.
Authors: Pierre Colombo, Malik Boudiaf, Allyn Sweet, Michael Desa, Hongxi Wang, Kevin Candra, Sym\'eon del Marmol
Abstract: Before closing venture capital financing rounds, lawyers conduct diligence that includes tying out the capitalization table: verifying that every security (for example, shares, options, warrants) and issuance term (for example, vesting schedules, acceleration triggers, transfer restrictions) is supported by large sets of underlying legal documentation. While LLMs continue to improve on legal benchmarks, specialized legal workflows, such as capitalization tie-out, remain out of reach even for strong agentic systems. The task requires multi-document reasoning, strict evidence traceability, and deterministic outputs that current approaches fail to reliably deliver. We characterize capitalization tie-out as an instance of a real-world benchmark for legal AI, analyze and compare the performance of existing agentic systems, and propose a world model architecture toward tie-out automation-and more broadly as a foundation for applied legal intelligence.
Authors: Yuchen Li, Handing Wang, Bing Xue, Mengjie Zhang, Yaochu Jin
Abstract: In the high-cost simulation-driven design domain, translating ambiguous design requirements into a mathematical optimization formulation is a bottleneck for optimizing product performance. This process is time-consuming and heavily reliant on expert knowledge. While large language models (LLMs) offer potential for automating this task, existing approaches either suffer from poor formalization that fails to accurately align with the design intent or rely on solver feedback for data filtering, which is unavailable due to the high simulation costs. To address this challenge, we propose APF, a framework for solver-independent, automated problem formulation via LLMs designed to automatically convert engineers' natural language requirements into executable optimization models. The core of this framework is an innovative pipeline for automatically generating high-quality data, which overcomes the difficulty of constructing suitable fine-tuning datasets in the absence of high-cost solver feedback with the help of data generation and test instance annotation. The generated high-quality dataset is used to perform supervised fine-tuning on LLMs, significantly enhancing their ability to generate accurate and executable optimization problem formulations. Experimental results on antenna design demonstrate that APF significantly outperforms the existing methods in both the accuracy of requirement formalization and the quality of resulting radiation efficiency curves in meeting the design goals.
Authors: Guibin Zhang, Haotian Ren, Chong Zhan, Zhenhong Zhou, Junhao Wang, He Zhu, Wangchunshu Zhou, Shuicheng Yan
Abstract: Self-evolving memory systems are unprecedentedly reshaping the evolutionary paradigm of large language model (LLM)-based agents. Prior work has predominantly relied on manually engineered memory architectures to store trajectories, distill experience, and synthesize reusable tools, enabling agents to evolve on the fly within environment interactions. However, this paradigm is fundamentally constrained by the staticity of the memory system itself: while memory facilitates agent-level evolving, the underlying memory architecture cannot be meta-adapted to diverse task contexts. To address this gap, we propose MemEvolve, a meta-evolutionary framework that jointly evolves agents' experiential knowledge and their memory architecture, allowing agent systems not only to accumulate experience but also to progressively refine how they learn from it. To ground MemEvolve in prior research and foster openness in future self-evolving systems, we introduce EvolveLab, a unified self-evolving memory codebase that distills twelve representative memory systems into a modular design space (encode, store, retrieve, manage), providing both a standardized implementation substrate and a fair experimental arena. Extensive evaluations on four challenging agentic benchmarks demonstrate that MemEvolve achieves (I) substantial performance gains, improving frameworks such as SmolAgent and Flash-Searcher by up to $17.06\%$; and (II) strong cross-task and cross-LLM generalization, designing memory architectures that transfer effectively across diverse benchmarks and backbone models.
Authors: Thorsten Hellert, Nikolay Agladze, Alex Giovannone, Jan Jug, Frank Mayet, Mark Sherwin, Antonin Sulc, Chris Tennant
Abstract: Modern experimental platforms such as particle accelerators, fusion devices, telescopes, and industrial process control systems expose tens to hundreds of thousands of control and diagnostic channels accumulated over decades of evolution. Operators and AI systems rely on informal expert knowledge, inconsistent naming conventions, and fragmented documentation to locate signals for monitoring, troubleshooting, and automated control, creating a persistent bottleneck for reliability, scalability, and language-model-driven interfaces. We formalize semantic channel finding-mapping natural-language intent to concrete control-system signals-as a general problem in complex experimental infrastructure, and introduce a four-paradigm framework to guide architecture selection across facility-specific data regimes. The paradigms span (i) direct in-context lookup over curated channel dictionaries, (ii) constrained hierarchical navigation through structured trees, (iii) interactive agent exploration using iterative reasoning and tool-based database queries, and (iv) ontology-grounded semantic search that decouples channel meaning from facility-specific naming conventions. We demonstrate each paradigm through proof-of-concept implementations at four operational facilities spanning two orders of magnitude in scale-from compact free-electron lasers to large synchrotron light sources-and diverse control-system architectures, from clean hierarchies to legacy environments. These implementations achieve 90-97% accuracy on expert-curated operational queries.
Authors: Yixia Li, Hongru Wang, Jiahao Qiu, Zhenfei Yin, Dongdong Zhang, Cheng Qian, Zeping Li, Pony Ma, Guanhua Chen, Heng Ji, Mengdi Wang
Abstract: Agentic reinforcement learning increasingly relies on experience-driven scaling, yet real-world environments remain non-adaptive, limited in coverage, and difficult to scale. World models offer a potential way to improve learning efficiency through simulated experience, but it remains unclear whether large language models can reliably serve this role and under what conditions they meaningfully benefit agents. We study these questions in text-based environments, which provide a controlled setting to reinterpret language modeling as next-state prediction under interaction. We introduce a three-level framework for evaluating LLM-based world models: (i) fidelity and consistency, (ii) scalability and robustness, and (iii) agent utility. Across five representative environments, we find that sufficiently trained world models maintain coherent latent state, scale predictably with data and model size, and improve agent performance via action verification, synthetic trajectory generation, and warm-starting reinforcement learning. Meanwhile, these gains depend critically on behavioral coverage and environment complexity, delineating clear boundry on when world modeling effectively supports agent learning.
Authors: Sultan Alrashed, Francesco Orabona
Abstract: We present AraMix, a deduplicated Arabic pretraining corpus containing approximately 178 billion tokens across 179 million documents. Rather than scraping the web again, AraMix demonstrates that substantial value lies in systematically reusing and curating existing pretraining datasets: we combine seven publicly available Arabic web datasets, apply quality filtering designed specifically for Arabic text to re-filter some datasets, and perform cross-dataset deduplication, both MinHash and sentence-level. This approach reveals that nearly 60% of tokens across these independently collected corpora are duplicates, redundancy that any new scraping efforts will reproduce. Our work suggests that for lower resource languages, investment in curation pipelines for existing data yields greater returns than additional web crawls, an approach that allowed us to curate the largest heavily filtered publicly available Arabic pretraining corpus.
Authors: Tung Duong Ta, Tim Oates
Abstract: Despite advances in mathematical reasoning capabilities, Large Language Models (LLMs) still struggle with calculation verification when using established prompting techniques. We present MDToC (Metacognitive Dynamic Tree of Concepts), a three-phase approach that constructs a concept tree, develops accuracy-verified calculations for each concept, and employs majority voting to evaluate competing solutions. Evaluations across CHAMP, MATH, and Game-of-24 benchmarks demonstrate our MDToC's effectiveness, with GPT-4-Turbo achieving 58.1\% on CHAMP, 86.6\% on MATH, and 85\% on Game-of-24 - outperforming GoT by 5\%, 5.4\%, and 4\% on all these tasks, respectively, without hand-engineered hints. MDToC consistently surpasses existing prompting methods across all backbone models, yielding improvements of up to 7.6\% over ToT and 6.2\% over GoT, establishing metacognitive calculation verification as a promising direction for enhanced mathematical reasoning.
Authors: Antonio San Martin
Abstract: The rapid diffusion of generative artificial intelligence is transforming terminology work. While this technology promises gains in efficiency, its unstructured adoption risks weakening professional autonomy, amplifying bias, and eroding linguistic and conceptual diversity. This paper argues that a human-centered approach to artificial intelligence has become a necessity for terminology work. Building on research in artificial intelligence and translation studies, it proposes a human-centered framework that conceptualizes artificial intelligence as a means of amplifying the terminologist's capabilities, rather than replacing them. The framework is organized around three interrelated dimensions: the augmented terminologist, ethical AI, and human-centered design. Together, these dimensions emphasize the compatibility of high automation with strong human control, the central role of terminologists in bias mitigation, and the importance of designing AI tools and workflows around the needs, values, and well-being of the terminologist. The paper concludes by stressing that current choices in AI adoption will shape not only terminological practice, but also the preservation of accuracy, adequacy, and diversity in terminology and specialized knowledge.
Authors: Ming Li, Han Chen, Yunze Xiao, Jian Chen, Hong Jiao, Tianyi Zhou
Abstract: Accurate estimation of item (question or task) difficulty is critical for educational assessment but suffers from the cold start problem. While Large Language Models demonstrate superhuman problem-solving capabilities, it remains an open question whether they can perceive the cognitive struggles of human learners. In this work, we present a large-scale empirical analysis of Human-AI Difficulty Alignment for over 20 models across diverse domains such as medical knowledge and mathematical reasoning. Our findings reveal a systematic misalignment where scaling up model size is not reliably helpful; instead of aligning with humans, models converge toward a shared machine consensus. We observe that high performance often impedes accurate difficulty estimation, as models struggle to simulate the capability limitations of students even when being explicitly prompted to adopt specific proficiency levels. Furthermore, we identify a critical lack of introspection, as models fail to predict their own limitations. These results suggest that general problem-solving capability does not imply an understanding of human cognitive struggles, highlighting the challenge of using current models for automated difficulty prediction.
Authors: Shaomu Tan, Ryosuke Mitani, Ritvik Choudhary, Qiyu Wu, Toshiyuki Sekiya, Christof Monz
Abstract: Over the years, automatic MT metrics have hillclimbed benchmarks and presented strong and sometimes human-level agreement with human ratings. Yet they remain black-box, offering little insight into their decision-making and often failing under real-world out-of-distribution (OOD) inputs. We introduce Remedy-R, a reasoning-driven generative MT metric trained with reinforcement learning from pairwise translation preferences, without requiring error-span annotations or distillation from closed LLMs. Remedy-R produces step-by-step analyses of accuracy, fluency, and completeness, followed by a final score, enabling more interpretable assessments. With only 60K training pairs across two language pairs, Remedy-R remains competitive with top scalar metrics and GPT-4-based judges on WMT22-24 meta-evaluation, generalizes to other languages, and exhibits strong robustness on OOD stress tests. Moreover, Remedy-R models generate self-reflective feedback that can be reused for translation improvement. Building on this finding, we introduce Remedy-R Agent, a simple evaluate-revise pipeline that leverages Remedy-R's evaluation analysis to refine translations. This agent consistently improves translation quality across diverse models, including Qwen2.5, ALMA-R, GPT-4o-mini, and Gemini-2.0-Flash, suggesting that Remedy-R's reasoning captures translation-relevant information and is practically useful.
Authors: Wen-Long Jin
Abstract: Large Language Models (LLMs) execute complex multi-turn interaction protocols but lack formal specifications to verify execution against designer intent. We introduce FASTRIC, a Prompt Specification Language that makes implicit Finite State Machines (FSMs) explicit in natural language prompts, enabling conformance verification through execution trace analysis. The LLM serves as intelligent execution agent: interpreting designer-encoded FSMs to execute specified behavioral roles. Unlike symbolic specification languages requiring parsers and compilers, FASTRIC leverages LLMs as unified infrastructure-simultaneously parser, interpreter, runtime environment, and development assistant. FASTRIC guides designers to articulate seven FSM elements (Final States, Agents, States, Triggers, Roles, Initial State, Constraints) structuring multi-turn interactions. Specification formality-ranging from implicit descriptions that frontier models infer to explicit step-by-step instructions for weaker models-serves as a design parameter. We introduce procedural conformance as verification metric measuring execution adherence to FSM specifications. Testing a 3-state kindergarten tutoring FSM across four formality levels and three model scales (14.7B, 685B, 1T+ parameters) reveals optimal specification formality is a function of model capacity. DeepSeek-V3.2 (685B) achieves perfect conformance (1.00) at L2-L4; ChatGPT-5 (~1T) peaks at L3 (0.90) before collapsing at L4 (0.39); Phi4 (14.7B) shows no stable optimum with high variance (SD=0.16-0.36). These findings reveal model-specific formality ranges-"Goldilocks zones"-where specifications provide sufficient structure without over-constraint, establishing Prompt Specification Engineering for creating verifiable interaction protocols, transforming multi-turn interaction design from heuristic art to systematic engineering with measurable procedural guarantees.
Authors: Jinyan Liu, Zikang Chen, Qinchuan Wang, Tan Xie, Heming Zheng, Xudong Lv
Abstract: When applied directly in an end-to-end manner to medical follow-up tasks, Large Language Models (LLMs) often suffer from uncontrolled dialog flow and inaccurate information extraction due to the complexity of follow-up forms. To address this limitation, we designed and compared two follow-up chatbot systems: an end-to-end LLM-based system (control group) and a modular pipeline with structured process control (experimental group). Experimental results show that while the end-to-end approach frequently fails on lengthy and complex forms, our modular method-built on task decomposition, semantic clustering, and flow management-substantially improves dialog stability and extraction accuracy. Moreover, it reduces the number of dialogue turns by 46.73% and lowers token consumption by 80% to 87.5%. These findings highlight the necessity of integrating external control mechanisms when deploying LLMs in high-stakes medical follow-up scenarios.
Authors: Tongyuan Miao, Gary Huang, Kai Jun Han, Annie Jiang
Abstract: Diffusion Large Language Models (DLLMs) enable fully parallel token decoding but often remain impractical at inference time due to the many denoising iterations required to refine an information-free, fully masked initialization into coherent text. Most existing acceleration methods focus on traversing this generative trajectory more efficiently via improved solvers or sampling strategies. We advance a complementary perspective: shorten the trajectory itself by starting closer to the target distribution through context-aware initialization. We propose a training-free interface that injects prompt-conditioned priors from a lightweight auxiliary model into the diffusion initialization, and instantiate it with two mechanisms: discrete token injection and representation-level embedding interpolation. Because injected priors can be imperfect and unmask-only decoding can over-commit early, we also introduce a simple confidence-based remasking mechanism as a form of prior skepticism. Preliminary evidence on GSM8K suggests that context-aware initialization can substantially reduce denoising iterations (about 35\% fewer function evaluations in our setting), while also exposing a key open challenge: naive warm-starting can degrade final accuracy relative to strong diffusion baselines. We use these findings to motivate a research agenda around calibration, revision mechanisms, and representation alignment for reliable warm-started diffusion decoding.
Authors: Shijian Ma, Yunqi Huang, Yan Lin
Abstract: Drama script continuation requires models to maintain character consistency, advance plot coherently, and preserve dramatic structurecapabilities that existing benchmarks fail to evaluate comprehensively. We present DramaBench, the first large-scale benchmark for evaluating drama script continuation across six independent dimensions: Format Standards, Narrative Efficiency, Character Consistency, Emotional Depth, Logic Consistency, and Conflict Handling. Our framework combines rulebased analysis with LLM-based labeling and statistical metrics, ensuring objective and reproducible evaluation. We conduct comprehensive evaluation of 8 state-of-the-art language models on 1,103 scripts (8,824 evaluations total), with rigorous statistical significance testing (252 pairwise comparisons, 65.9% significant) and human validation (188 scripts, substantial agreement on 3/5 dimensions). Our ablation studies confirm all six dimensions capture independent quality aspects (mean | r | = 0.020). DramaBench provides actionable, dimensionspecific feedback for model improvement and establishes a rigorous standard for creative writing evaluation.
Authors: Ziyan Zhang, Chao Wang, Zhuo Chen, Lei Chen, Chiyi Li, Kai Song
Abstract: Reasoning over knowledge graphs (KGs) with first-order logic (FOL) queries is challenging due to the inherent incompleteness of real-world KGs and the compositional complexity of logical query structures. Most existing methods rely on embedding entities and relations into continuous geometric spaces and answer queries via differentiable set operations. While effective for simple query patterns, these approaches often struggle to generalize to complex queries involving multiple operators, deeper reasoning chains, or heterogeneous KG schemas. We propose ROG (Reasoning Over knowledge Graphs with large language models), an ensemble-style framework that combines query-aware KG neighborhood retrieval with large language model (LLM)-based chain-of-thought reasoning. ROG decomposes complex FOL queries into sequences of simpler sub-queries, retrieves compact, query-relevant subgraphs as contextual evidence, and performs step-by-step logical inference using an LLM, avoiding the need for task-specific embedding optimization. Experiments on standard KG reasoning benchmarks demonstrate that ROG consistently outperforms strong embedding-based baselines in terms of mean reciprocal rank (MRR), with particularly notable gains on high-complexity query types. These results suggest that integrating structured KG retrieval with LLM-driven logical reasoning offers a robust and effective alternative for complex KG reasoning tasks.
Authors: Amar Lakel (MICA)
Abstract: This paper proposes an epistemological shift in the analysis of large generative models, replacing the category ''Large Language Models'' (LLM) with that of ''Large Discourse Models'' (LDM), and then with that of Artificial Discursive Agent (ADA). The theoretical framework is based on an ontological triad distinguishing three regulatory instances: the apprehension of the phenomenal regularities of the referential world, the structuring of embodied cognition, and the structural-linguistic sedimentation of the utterance within a socio-historical context. LDMs, operating on the product of these three instances (the document), model the discursive projection of a portion of human experience reified by the learning corpus. The proposed program aims to replace the ''fascination/fear'' dichotomy with public trials and procedures that make the place, uses, and limits of artificial discursive agents in contemporary social space decipherable, situating this approach within a perspective of governance and co-regulation involving the State, industry, civil society, and academia.
Authors: Tzu-Yun Lee, Ding-Yong Hong, Jan-Jan Wu
Abstract: This paper introduces Syntactic Attention Pruning (SAP), a novel method for effectively pruning attention heads in Transformer models. Unlike conventional approaches that rely solely on mathematical analysis of model weights and activations, SAP incorporates both the syntactic structure and attention patterns of sentences to guide the pruning process. By leveraging these linguistic features, SAP not only achieves performance comparable to state-of-the-art methods but also enhances the interpretability of model behavior. To further improve robustness, we propose Candidate Filtering (CF), a mechanism that prioritizes heads based on their contribution to model performance, mitigating degradation during pruning. Experimental results indicate that SAP effectively preserves critical heads of a high density of strong attention values, outperforming existing head pruning strategies in retrain-free settings. These findings position SAP as a promising foundation for a new direction in model compression research, offering high flexibility for pruning across all transformer-based language models.
Authors: Zihan Lin, Xiaohan Wang, Hexiong Yang, Jiajun Chai, Jie Cao, Guojun Yin, Wei Lin, Ran He
Abstract: While reinforcement learning (RL) shows promise in training tool-use large language models (LLMs) using verifiable outcome rewards, existing methods largely overlook the potential of explicit reasoning rewards to bolster reasoning and tool utilization. Furthermore, natively combining reasoning and outcome rewards may yield suboptimal performance or conflict with the primary optimization objective. To address this, we propose advantage-weighted policy optimization (AWPO) -- a principled RL framework that effectively integrates explicit reasoning rewards to enhance tool-use capability. AWPO incorporates variance-aware gating and difficulty-aware weighting to adaptively modulate advantages from reasoning signals based on group-relative statistics, alongside a tailored clipping mechanism for stable optimization. Extensive experiments demonstrate that AWPO achieves state-of-the-art performance across standard tool-use benchmarks, significantly outperforming strong baselines and leading closed-source models in challenging multi-turn scenarios. Notably, with exceptional parameter efficiency, our 4B model surpasses Grok-4 by 16.0 percent in multi-turn accuracy while preserving generalization capability on the out-of-distribution MMLU-Pro benchmark.
Authors: Dehai Min, Kailin Zhang, Tongtong Wu, Lu Cheng
Abstract: Dynamic Retrieval-Augmented Generation adaptively determines when to retrieve during generation to mitigate hallucinations in large language models (LLMs). However, existing methods rely on model-internal signals (e.g., logits, entropy), which are fundamentally unreliable because LLMs are typically ill-calibrated and often exhibit high confidence in erroneous outputs. We propose QuCo-RAG, which shifts from subjective confidence to objective statistics computed from pre-training data. Our method quantifies uncertainty through two stages: (1) before generation, we identify low-frequency entities indicating long-tail knowledge gaps; (2) during generation, we verify entity co-occurrence in the pre-training corpus, where zero co-occurrence often signals hallucination risk. Both stages leverage Infini-gram for millisecond-latency queries over 4 trillion tokens, triggering retrieval when uncertainty is high. Experiments on multi-hop QA benchmarks show QuCo-RAG achieves EM gains of 5--12 points over state-of-the-art baselines with OLMo-2 models, and transfers effectively to models with undisclosed pre-training data (Llama, Qwen, GPT), improving EM by up to 14 points. Domain generalization on biomedical QA further validates the robustness of our paradigm. These results establish corpus-grounded verification as a principled, practically model-agnostic paradigm for dynamic RAG. Our code is publicly available at https://github.com/ZhishanQ/QuCo-RAG.
Authors: Alessandro Lucca, Francesco Pierri
Abstract: Subtitles are essential for video accessibility and audience engagement. Modern Automatic Speech Recognition (ASR) systems, built upon Encoder-Decoder neural network architectures and trained on massive amounts of data, have progressively reduced transcription errors on standard benchmark datasets. However, their performance in real-world production environments, particularly for non-English content like long-form Italian videos, remains largely unexplored. This paper presents a case study on developing a professional subtitling system for an Italian media company. To inform our system design, we evaluated four state-of-the-art ASR models (Whisper Large v2, AssemblyAI Universal, Parakeet TDT v3 0.6b, and WhisperX) on a 50-hour dataset of Italian television programs. The study highlights their strengths and limitations, benchmarking their performance against the work of professional human subtitlers. The findings indicate that, while current models cannot meet the media industry's accuracy needs for full autonomy, they can serve as highly effective tools for enhancing human productivity. We conclude that a human-in-the-loop (HITL) approach is crucial and present the production-grade, cloud-based infrastructure we designed to support this workflow.
Authors: Bingyang Kelvin Liu, Ziyu Patrick Chen
Abstract: While Joint-Embedding Predictive Architecture (JEPA) has emerged as a powerful architecture for learning rich latent representations, it fundamentally lacks generative abilities. Meanwhile, latent space reasoning attempts for Transformer models like COCONUT do improve performance, but they ultimately rely on token-by-token generation, which still accumulates compounding error and relies on context information to gain reasoning insights. To address these limitations, we propose JEPA-Reasoner, a novel JEPA model enhanced with generative ability that reasons in latent space. We augment it with a separate action-taker model, Talker, to produce human-readable sentences. Our approach demonstrates that decoupling latent space reasoning and token generation enables JEPA-Reasoner to produce mixed latent vectors that might lay the foundation for multi-threaded reasoning, while performing autoregressive generation with superior robustness to compounding error.
Authors: Dazhen Deng, Sen Yang, Yuchen He, Yuan Tian, Yingcai Wu
Abstract: Current chart-specific tasks, such as chart question answering, chart parsing, and chart generation, are typically studied in isolation, preventing models from learning the shared semantics that link chart generation and interpretation. We introduce CycleChart, a consistency-based learning framework for bidirectional chart understanding and generation. CycleChart adopts a schema-centric formulation as a common interface across tasks. We construct a consistent multi-task dataset, where each chart sample includes aligned annotations for schema prediction, data parsing, and question answering. To learn cross-directional chart semantics, CycleChart introduces a generate-parse consistency objective: the model generates a chart schema from a table and a textual query, then learns to recover the schema and data from the generated chart, enforcing semantic alignment across directions. CycleChart achieves strong results on chart generation, chart parsing, and chart question answering, demonstrating improved cross-task generalization and marking a step toward more general chart understanding models.
Authors: Anna-Maria Gueorguieva, Aylin Caliskan
Abstract: Large language models (LLMs) have been shown to exhibit social bias, however, bias towards non-protected stigmatized identities remain understudied. Furthermore, what social features of stigmas are associated with bias in LLM outputs is unknown. From psychology literature, it has been shown that stigmas contain six shared social features: aesthetics, concealability, course, disruptiveness, origin, and peril. In this study, we investigate if human and LLM ratings of the features of stigmas, along with prompt style and type of stigma, have effect on bias towards stigmatized groups in LLM outputs. We measure bias against 93 stigmatized groups across three widely used LLMs (Granite 3.0-8B, Llama-3.1-8B, Mistral-7B) using SocialStigmaQA, a benchmark that includes 37 social scenarios about stigmatized identities; for example deciding wether to recommend them for an internship. We find that stigmas rated by humans to be highly perilous (e.g., being a gang member or having HIV) have the most biased outputs from SocialStigmaQA prompts (60% of outputs from all models) while sociodemographic stigmas (e.g. Asian-American or old age) have the least amount of biased outputs (11%). We test if the amount of biased outputs could be decreased by using guardrail models, models meant to identify harmful input, using each LLM's respective guardrail model (Granite Guardian 3.0, Llama Guard 3.0, Mistral Moderation API). We find that bias decreases significantly by 10.4%, 1.4%, and 7.8%, respectively. However, we show that features with significant effect on bias remain unchanged post-mitigation and that guardrail models often fail to recognize the intent of bias in prompts. This work has implications for using LLMs in scenarios involving stigmatized groups and we suggest future work towards improving guardrail models for bias mitigation.
Authors: Mingxu Zhang, Dazhong Shen, Qi Zhang, Ying Sun
Abstract: Large Language Models (LLMs) exhibit strong general reasoning but struggle in molecular science due to the lack of explicit chemical priors in standard string representations. Current solutions face a fundamental dilemma. Training-based methods inject priors into parameters, but this static coupling hinders rapid knowledge updates and often compromises the model's general reasoning capabilities. Conversely, existing training-free methods avoid these issues but rely on surface-level prompting, failing to provide the fine-grained atom-level priors essential for precise chemical reasoning. To address this issue, we introduce ChemATP, a framework that decouples chemical knowledge from the reasoning engine. By constructing the first atom-level textual knowledge base, ChemATP enables frozen LLMs to explicitly retrieve and reason over this information dynamically. This architecture ensures interpretability and adaptability while preserving the LLM's intrinsic general intelligence. Experiments show that ChemATP significantly outperforms training-free baselines and rivals state-of-the-art training-based models, demonstrating that explicit prior injection is a competitive alternative to implicit parameter updates.
Authors: Do Minh Duc, Quan Xuan Truong, Nguyen Tat Dat, Nguyen Van Vinh
Abstract: Prompt engineering plays a critical role in adapting large language models (LLMs) to complex reasoning and labeling tasks without the need for extensive fine-tuning. In this paper, we propose a novel prompt optimization pipeline for frame detection in logistics texts, combining retrieval-augmented generation (RAG), few-shot prompting, chain-of-thought (CoT) reasoning, and automatic CoT synthesis (Auto-CoT) to generate highly effective task-specific prompts. Central to our approach is an LLM-based prompt optimizer agent that iteratively refines the prompts using retrieved examples, performance feedback, and internal self-evaluation. Our framework is evaluated on a real-world logistics text annotation task, where reasoning accuracy and labeling efficiency are critical. Experimental results show that the optimized prompts - particularly those enhanced via Auto-CoT and RAG - improve real-world inference accuracy by up to 15% compared to baseline zero-shot or static prompts. The system demonstrates consistent improvements across multiple LLMs, including GPT-4o, Qwen 2.5 (72B), and LLaMA 3.1 (70B), validating its generalizability and practical value. These findings suggest that structured prompt optimization is a viable alternative to full fine-tuning, offering scalable solutions for deploying LLMs in domain-specific NLP applications such as logistics.
Authors: Javier Vela-Tambo, Jorge Gracia, Fernando Dominguez-Castro
Abstract: Understanding and monitoring the socio-economic impacts of climate hazards requires extracting structured information from heterogeneous news articles on a large scale. To that end, we have developed CienaLLM, a modular framework based on schema-guided Generative Information Extraction. CienaLLM uses open-weight Large Language Models for zero-shot information extraction from news articles, and supports configurable prompts and output schemas, multi-step pipelines, and cloud or on-premise inference. To systematically assess how the choice of LLM family, size, precision regime, and prompting strategy affect performance, we run a large factorial study in models, precisions, and prompt engineering techniques. An additional response parsing step nearly eliminates format errors while preserving accuracy; larger models deliver the strongest and most stable performance, while quantization offers substantial efficiency gains with modest accuracy trade-offs; and prompt strategies show heterogeneous, model-specific effects. CienaLLM matches or outperforms the supervised baseline in accuracy for extracting drought impacts from Spanish news, although at a higher inference cost. While evaluated in droughts, the schema-driven and model-agnostic design is suitable for adapting to related information extraction tasks (e.g., other hazards, sectors, or languages) by editing prompts and schemas rather than retraining. We release code, configurations, and schemas to support reproducible use.
Authors: Zhiqing Hu, Chenxu Zhao, Jiazhong Lu, Xiaolei Liu
Abstract: Misuse of LLM-generated text can be curbed by watermarking techniques that embed implicit signals into the output. We propose a watermark that partitions the vocabulary at each decoding step into three sets (Green/Yellow/Red) with fixed ratios and restricts sampling to the Green and Yellow sets. At detection time, we replay the same partitions, compute Green-enrichment and Red-depletion statistics, convert them to one-sided z-scores, and aggregate their p-values via Fisher's method to decide whether a passage is watermarked. We implement generation, detection, and testing on Llama 2 7B, and evaluate true-positive rate, false-positive rate, and text quality. Results show that the triple-partition scheme achieves high detection accuracy at fixed FPR while preserving readability.
Authors: Yacouba Diarra, Panga Azazia Kamate, Nouhoum Souleymane Coulibaly, Michael Leventhal
Abstract: We present Kunkado, a 160-hour Bambara ASR dataset compiled from Malian radio archives to capture present-day spontaneous speech across a wide range of topics. It includes code-switching, disfluencies, background noise, and overlapping speakers that practical ASR systems encounter in real-world use. We finetuned Parakeet-based models on a 33.47-hour human-reviewed subset and apply pragmatic transcript normalization to reduce variability in number formatting, tags, and code-switching annotations. Evaluated on two real-world test sets, finetuning with Kunkado reduces WER from 44.47\% to 37.12\% on one and from 36.07\% to 32.33\% on the other. In human evaluation, the resulting model also outperforms a comparable system with the same architecture trained on 98 hours of cleaner, less realistic speech. We release the data and models to support robust ASR for predominantly oral languages.
Authors: Jian Yang, Wei Zhang, Yizhi Li, Shawn Guo, Haowen Wang, Aishan Liu, Ge Zhang, Zili Wang, Zhoujun Li, Xianglong Liu, Weifeng Lv
Abstract: Large language models (LLMs) have made significant strides in code generation, achieving impressive capabilities in synthesizing code snippets from natural language instructions. However, a critical challenge remains in ensuring LLMs generate factually accurate responses about programming concepts, technical implementations, etc. Most previous code-related benchmarks focus on code execution correctness, overlooking the factual accuracy of programming knowledge. To address this gap, we present CodeSimpleQA, a comprehensive bilingual benchmark designed to evaluate the factual accuracy of code LLMs in answering code-related questions, which contains carefully curated question-answer pairs in both English and Chinese, covering diverse programming languages and major computer science domains. Further, we create CodeSimpleQA-Instruct, a large-scale instruction corpus with 66M samples, and develop a post-training framework combining supervised fine-tuning and reinforcement learning. Our comprehensive evaluation of diverse LLMs reveals that even frontier LLMs struggle with code factuality. Our proposed framework demonstrates substantial improvements over the base model, underscoring the critical importance of factuality-aware alignment in developing reliable code LLMs.
Authors: Quyu Kong, Xu Zhang, Zhenyu Yang, Nolan Gao, Chen Liu, Panrong Tong, Chenglin Cai, Hanzhang Zhou, Jianan Zhang, Liangyu Chen, Zhidan Liu, Steven Hoi, Yue Wang
Abstract: Among existing online mobile-use benchmarks, AndroidWorld has emerged as the dominant benchmark due to its reproducible environment and deterministic evaluation; however, recent agents achieving over 90% success rates indicate its saturation and motivate the need for a more challenging benchmark. In addition, its environment lacks key application categories, such as e-commerce and enterprise communication, and does not reflect realistic mobile-use scenarios characterized by vague user instructions and hybrid tool usage. To bridge this gap, we introduce MobileWorld, a substantially more challenging benchmark designed to better reflect real-world mobile usage, comprising 201 tasks across 20 applications, while maintaining the same level of reproducible evaluation as AndroidWorld. The difficulty of MobileWorld is twofold. First, it emphasizes long-horizon tasks with cross-application interactions: MobileWorld requires nearly twice as many task-completion steps on average (27.8 vs. 14.3) and includes far more multi-application tasks (62.2% vs. 9.5%) compared to AndroidWorld. Second, MobileWorld extends beyond standard GUI manipulation by introducing novel task categories, including agent-user interaction and MCP-augmented tasks. To ensure robust evaluation, we provide snapshot-based container environment and precise functional verifications, including backend database inspection and task callback APIs. We further develop a planner-executor agentic framework with extended action spaces to support user interactions and MCP calls. Our results reveal a sharp performance drop compared to AndroidWorld, with the best agentic framework and end-to-end model achieving 51.7% and 20.9% success rates, respectively. Our analysis shows that current models struggle significantly with user interaction and MCP calls, offering a strategic roadmap toward more robust, next-generation mobile intelligence.
Authors: Thittipat Pairatsuppawat, Abhibhu Tachaapornchai, Paweekorn Kusolsomboon, Chutikan Chaiwong, Thodsaporn Chay-intr, Kobkrit Viriyayudhakorn, Nongnuch Ketui, Aslan B. Wong
Abstract: Open-weights large language models remain difficult to deploy for Thai due to unstable generation under complex instructions, despite strong English performance. To mitigate these limitations, We present SiamGPT-32B, an open-weights model based on Qwen3-32B, fine-tuned with a Quality-First strategy emphasizing curated supervision over data scale. The fine-tuning pipeline combines translated high-complexity English instruction data with a Thai-adapted AutoIF framework for instruction and linguistic constraints. Using supervised fine-tuning only, without continual pretraining or corpus expansion, SiamGPT-32B improves instruction adherence, multi-turn robustness, and linguistic stability. Evaluations on the SEA-HELM benchmark show that SiamGPT-32B achieves the strongest overall performance among similar-scale open-weights Thai models, with consistent gains in instruction following, multi-turn dialogue, and natural language understanding.
Authors: Jinwei Chi, Ke Wang, Yu Chen, Xuanye Lin, Qiang Xu
Abstract: Automated essay scoring (AES) is a challenging task in cross-prompt settings due to the diversity of scoring criteria. While previous studies have focused on the output of large language models (LLMs) to improve scoring accuracy, we believe activations from intermediate layers may also provide valuable information. To explore this possibility, we evaluated the discriminative power of LLMs' activations in cross-prompt essay scoring task. Specifically, we used activations to fit probes and further analyzed the effects of different models and input content of LLMs on this discriminative power. By computing the directions of essays across various trait dimensions under different prompts, we analyzed the variation in evaluation perspectives of large language models concerning essay types and traits. Results show that the activations possess strong discriminative power in evaluating essay quality and that LLMs can adapt their evaluation perspectives to different traits and essay types, effectively handling the diversity of scoring criteria in cross-prompt settings.
Authors: Ivan Decostanzi, Yelena Mejova, Kyriaki Kalimeri
Abstract: Timely and accurate situational reports are essential for humanitarian decision-making, yet current workflows remain largely manual, resource intensive, and inconsistent. We present a fully automated framework that uses large language models (LLMs) to transform heterogeneous humanitarian documents into structured and evidence-grounded reports. The system integrates semantic text clustering, automatic question generation, retrieval augmented answer extraction with citations, multi-level summarization, and executive summary generation, supported by internal evaluation metrics that emulate expert reasoning. We evaluated the framework across 13 humanitarian events, including natural disasters and conflicts, using more than 1,100 documents from verified sources such as ReliefWeb. The generated questions achieved 84.7 percent relevance, 84.0 percent importance, and 76.4 percent urgency. The extracted answers reached 86.3 percent relevance, with citation precision and recall both exceeding 76 percent. Agreement between human and LLM based evaluations surpassed an F1 score of 0.80. Comparative analysis shows that the proposed framework produces reports that are more structured, interpretable, and actionable than existing baselines. By combining LLM reasoning with transparent citation linking and multi-level evaluation, this study demonstrates that generative AI can autonomously produce accurate, verifiable, and operationally useful humanitarian situation reports.
Authors: Bobo Li, Xudong Han, Jiang Liu, Yuzhe Ding, Liqiang Jing, Zhaoqi Zhang, Jinheng Li, Xinya Du, Fei Li, Meishan Zhang, Min Zhang, Aixin Sun, Philip S. Yu, Hao Fei
Abstract: Large language models (LLMs) and multimodal LLMs are changing event extraction (EE): prompting and generation can often produce structured outputs in zero shot or few shot settings. Yet LLM based pipelines face deployment gaps, including hallucinations under weak constraints, fragile temporal and causal linking over long contexts and across documents, and limited long horizon knowledge management within a bounded context window. We argue that EE should be viewed as a system component that provides a cognitive scaffold for LLM centered solutions. Event schemas and slot constraints create interfaces for grounding and verification; event centric structures act as controlled intermediate representations for stepwise reasoning; event links support relation aware retrieval with graph based RAG; and event stores offer updatable episodic and agent memory beyond the context window. This survey covers EE in text and multimodal settings, organizing tasks and taxonomy, tracing method evolution from rule based and neural models to instruction driven and generative frameworks, and summarizing formulations, decoding strategies, architectures, representations, datasets, and evaluation. We also review cross lingual, low resource, and domain specific settings, and highlight open challenges and future directions for reliable event centric systems. Finally, we outline open challenges and future directions that are central to the LLM era, aiming to evolve EE from static extraction into a structurally reliable, agent ready perception and memory layer for open world systems.
Authors: Zakaria Benmounah, Abdennour Boulesnane
Abstract: We present Algerian Dialect, a large-scale sentiment-annotated dataset consisting of 45,000 YouTube comments written in Algerian Arabic dialect. The comments were collected from more than 30 Algerian press and media channels using the YouTube Data API. Each comment is manually annotated into one of five sentiment categories: very negative, negative, neutral, positive, and very positive. In addition to sentiment labels, the dataset includes rich metadata such as collection timestamps, like counts, video URLs, and annotation dates. This dataset addresses the scarcity of publicly available resources for Algerian dialect and aims to support research in sentiment analysis, dialectal Arabic NLP, and social media analytics. The dataset is publicly available on Mendeley Data under a CC BY 4.0 license at https://doi.org/10.17632/zzwg3nnhsz.2.
Authors: Ignacio Iacobacci, Zhaozhi Qian, Faroq AL-Tam, Muhammad AL-Qurishi, Riad Souissi
Abstract: Recently, a new wave of thinking-capable Large Language Models has emerged, demonstrating exceptional capabilities across a wide range of reasoning benchmarks. Early studies have begun to explore how the amount of compute in terms of the length of the reasoning process, the so-called thinking budget, impacts model performance. In this work, we propose a systematic investigation of the thinking budget as a key parameter, examining its interaction with various configurations such as self-consistency, reflection, and others. Our goal is to provide an informative, balanced comparison framework that considers both performance outcomes and computational cost. Among our findings, we discovered that simply increasing the thinking budget is not the most effective use of compute. More accurate responses can instead be achieved through alternative configurations, such as self-consistency and self-reflection.
Authors: Angelo Ortiz Tandazo, Manel Khentout, Youssef Benchekroun, Thomas Hueber, Emmanuel Dupoux
Abstract: This paper introduces MauBERT, a multilingual extension of HuBERT that leverages articulatory features for robust cross-lingual phonetic representation learning. We continue HuBERT pre-training with supervision based on a phonetic-to-articulatory feature mapping in 55 languages. Our models learn from multilingual data to predict articulatory features or phones, resulting in language-independent representations that capture multilingual phonetic properties. Through comprehensive ABX discriminability testing, we show MauBERT models produce more context-invariant representations than state-of-the-art multilingual self-supervised learning models. Additionally, the models effectively adapt to unseen languages and casual speech with minimal self-supervised fine-tuning (10 hours of speech). This establishes an effective approach for instilling linguistic inductive biases in self-supervised speech models.
Authors: Zahra Sadeghi, Evangelos Milios, Frank Rudzicz
Abstract: Summary assessment involves evaluating how well a generated summary reflects the key ideas and meaning of the source text, requiring a deep understanding of the content. Large Language Models (LLMs) have been used to automate this process, acting as judges to evaluate summaries with respect to the original text. While previous research investigated the alignment between LLMs and Human responses, it is not yet well understood what properties or features are exploited by them when asked to evaluate based on a particular quality dimension, and there has not been much attention towards mapping between evaluation scores and metrics. In this paper, we address this issue and discover features aligned with Human and Generative Pre-trained Transformers (GPTs) responses by studying statistical and machine learning metrics. Furthermore, we show that instructing GPTs to employ metrics used by Human can improve their judgment and conforming them better with human responses.
Authors: Rolando Coto-Solano, Daisy Li, Manoela Teleginski Ferraz, Olivia Sasse, Cha Krupka, Sharid Lo\'aiciga, Sally Akevai Tenamu Nicholas
Abstract: We present experiments on diacritic restoration, a form of text normalization essential for natural language processing (NLP) tasks. Our study focuses on two extremely under-resourced languages: Bribri, a Chibchan language spoken in Costa Rica, and Cook Islands M\=aori, a Polynesian language spoken in the Cook Islands. Specifically, this paper: (i) compares algorithms for diacritics restoration in under-resourced languages, including tonal diacritics, (ii) examines the amount of data required to achieve target performance levels, (iii) contrasts results across varying resource conditions, and (iv) explores the related task of diacritic correction. We find that fine-tuned, character-level LLMs perform best, likely due to their ability to decompose complex characters into their UTF-8 byte representations. In contrast, massively multilingual models perform less effectively given our data constraints. Across all models, reliable performance begins to emerge with data budgets of around 10,000 words. Zero-shot approaches perform poorly in all cases. This study responds both to requests from the language communities and to broader NLP research questions concerning model performance and generalization in under-resourced contexts.
Authors: Filippos Ventirozos, Peter Appleby, Matthew Shardlow
Abstract: Aspect-Category Sentiment Analysis (ACSA) provides granular insights by identifying specific themes within reviews and their associated sentiment. While supervised learning approaches dominate this field, the scarcity and high cost of annotated data for new domains present significant barriers. We argue that leveraging large language models (LLMs) in a zero-shot setting is a practical alternative where resources for data annotation are limited. In this work, we propose a novel Chain-of-Thought (CoT) prompting technique that utilises an intermediate Unified Meaning Representation (UMR) to structure the reasoning process for the ACSA task. We evaluate this UMR-based approach against a standard CoT baseline across three models (Qwen3-4B, Qwen3-8B, and Gemini-2.5-Pro) and four diverse datasets. Our findings suggest that UMR effectiveness may be model-dependent. Whilst preliminary results indicate comparable performance for mid-sized models such as Qwen3-8B, these observations warrant further investigation, particularly regarding the potential applicability to smaller model architectures. Further research is required to establish the generalisability of these findings across different model scales.
Authors: Jiacheng Guo, Ling Yang, Peter Chen, Qixin Xiao, Yinjie Wang, Xinzhe Juan, Jiahao Qiu, Ke Shen, Mengdi Wang
Abstract: Training capable Large Language Model (LLM) agents is critically bottlenecked by the high cost and static nature of real-world interaction data. We address this by introducing GenEnv, a framework that establishes a difficulty-aligned co-evolutionary game between an agent and a scalable, generative environment simulator. Unlike traditional methods that evolve models on static datasets, GenEnv instantiates a dataevolving: the simulator acts as a dynamic curriculum policy, continuously generating tasks specifically tailored to the agent's ``zone of proximal development''. This process is guided by a simple but effective $\alpha$-Curriculum Reward, which aligns task difficulty with the agent's current capabilities. We evaluate GenEnv on five benchmarks, including API-Bank, ALFWorld, BFCL, Bamboogle, and TravelPlanner. Across these tasks, GenEnv improves agent performance by up to \textbf{+40.3\%} over 7B baselines and matches or exceeds the average performance of larger models. Compared to Gemini 2.5 Pro-based offline data augmentation, GenEnv achieves better performance while using 3.3$\times$ less data. By shifting from static supervision to adaptive simulation, GenEnv provides a data-efficient pathway for scaling agent capabilities.
Authors: Ravi Prasad
Abstract: Monte Carlo algorithms are a foundational pillar of modern computational science, yet their effective application hinges on a deep understanding of their performance trade offs. This paper presents a critical analysis of the evolution of Monte Carlo algorithms, focusing on the persistent tension between statistical efficiency and computational cost. We describe the historical development from the foundational Metropolis Hastings algorithm to contemporary methods like Hamiltonian Monte Carlo. A central emphasis of this survey is the rigorous discussion of time and space complexity, including upper, lower, and asymptotic tight bounds for each major algorithm class. We examine the specific motivations for developing these methods and the key theoretical and practical observations such as the introduction of gradient information and adaptive tuning in HMC that led to successively better solutions. Furthermore, we provide a justification framework that discusses explicit situations in which using one algorithm is demonstrably superior to another for the same problem. The paper concludes by assessing the profound significance and impact of these algorithms and detailing major current research challenges.
Authors: Shubham Kumar Nigam, Parjanya Aditya Shukla, Noel Shallum, Arnab Bhattacharya
Abstract: Handwritten text recognition (HTR) and machine translation continue to pose significant challenges, particularly for low-resource languages like Marathi, which lack large digitized corpora and exhibit high variability in handwriting styles. The conventional approach to address this involves a two-stage pipeline: an OCR system extracts text from handwritten images, which is then translated into the target language using a machine translation model. In this work, we explore and compare the performance of traditional OCR-MT pipelines with Vision Large Language Models that aim to unify these stages and directly translate handwritten text images in a single, end-to-end step. Our motivation is grounded in the urgent need for scalable, accurate translation systems to digitize legal records such as FIRs, charge sheets, and witness statements in India's district and high courts. We evaluate both approaches on a curated dataset of handwritten Marathi legal documents, with the goal of enabling efficient legal document processing, even in low-resource environments. Our findings offer actionable insights toward building robust, edge-deployable solutions that enhance access to legal information for non-native speakers and legal professionals alike.
Authors: Changxu Duan
Abstract: Academic documents stored in PDF format can be transformed into plain text structured markup languages to enhance accessibility and enable scalable digital library workflows. Markup languages allow for easier updates and customization, making academic content more adaptable and accessible to diverse usage, such as linguistic corpus compilation. Such documents, typically delivered in PDF format, contain complex elements including mathematical formulas, figures, headers, and tables, as well as densely layouted text. Existing end-to-end decoder transformer models can transform screenshots of documents into markup language. However, these models exhibit significant inefficiencies; their token-by-token decoding from scratch wastes a lot of inference steps in regenerating dense text that could be directly copied from PDF files. To solve this problem, we introduce EditTrans, a hybrid editing-generation model whose features allow identifying a queue of to-be-edited text from a PDF before starting to generate markup language. EditTrans contains a lightweight classifier fine-tuned from a Document Layout Analysis model on 162,127 pages of documents from arXiv. In our evaluations, EditTrans reduced the transformation latency up to 44.5% compared to end-to-end decoder transformer models, while maintaining transformation quality. Our code and reproducible dataset production scripts are open-sourced.
Authors: Kohei Watanabe
Abstract: Social scientists employ latent Dirichlet allocation (LDA) to find highly specific topics in large corpora, but they often struggle in this task because (1) LDA, in general, takes a significant amount of time to fit on large corpora; (2) unsupervised LDA fragments topics into sub-topics in short documents; (3) semi-supervised LDA fails to identify specific topics defined using seed words. To solve these problems, I have developed a new topic model called distributed asymmetric allocation (DAA) that integrates multiple algorithms for efficiently identifying sentences about important topics in large corpora. I evaluate the ability of DAA to identify politically important topics by fitting it to the transcripts of speeches at the United Nations General Assembly between 1991 and 2017. The results show that DAA can classify sentences significantly more accurately and quickly than LDA thanks to the new algorithms. More generally, the results demonstrate that it is important for social scientists to optimize Dirichlet priors of LDA to perform content analysis accurately.
Authors: Jian Yan
Abstract: This paper proposes the External Hippocampus framework, which models language model reasoning from a cognitive dynamics perspective as the flow of information energy in semantic space. Unlike traditional weight-space optimization methods, this framework constructs topological cognitive maps through dimensionality reduction projection, enabling precise navigation and intervention of energy flow at test time while avoiding substantial computational requirements and demonstrating predictable intervention patterns. The method effectively addresses the cognitive deadlock problem in multi-step reasoning for small models. Experiments on models <=7B parameters show: map-guided methods achieve 81.20% accuracy on 500 challenging problems (relative baseline +16.80%), reduce reasoning time by >= 15x, with key findings revealing that reasoning stagnation manifests as "Cognitive Vortex" and low-entropy potential wells, while temperature perturbations effectively restart energy flow. The framework requires no additional training, possesses autonomous growth capability, and provides an efficient and controllable topological-aware solution for small model reasoning.
Authors: Rui Liu, Dian Yu, Lei Ke, Haolin Liu, Yujun Zhou, Zhenwen Liang, Haitao Mi, Pratap Tokekar, Dong Yu
Abstract: Reinforcement Learning with Verifiable Rewards (RLVR) has become a key paradigm to improve the reasoning capabilities of Multimodal Large Language Models (MLLMs). However, prevalent group-based algorithms such as GRPO require multi-rollout sampling for each prompt. While more efficient single-rollout variants have recently been explored in text-only settings, we find that they suffer from severe instability in multimodal contexts, often leading to training collapse. To address this training efficiency-stability trade-off, we introduce $\textbf{MSSR}$ (Multimodal Stabilized Single-Rollout), a group-free RLVR framework that achieves both stable optimization and effective multimodal reasoning performance. MSSR achieves this via an entropy-based advantage-shaping mechanism that adaptively regularizes advantage magnitudes, preventing collapse and maintaining training stability. While such mechanisms have been used in group-based RLVR, we show that in the multimodal single-rollout setting they are not merely beneficial but essential for stability. In in-distribution evaluations, MSSR demonstrates superior training compute efficiency, achieving similar validation accuracy to the group-based baseline with half the training steps. When trained for the same number of steps, MSSR's performance surpasses the group-based baseline and shows consistent generalization improvements across five diverse reasoning-intensive benchmarks. Together, these results demonstrate that MSSR enables stable, compute-efficient, and effective RLVR for complex multimodal reasoning tasks.
Authors: Aryan Chaudhary, Sanchit Goyal, Pratik Narang, Dhruv Kumar
Abstract: Vision-Language Models have demonstrated remarkable capabilities in understanding visual content, yet systematic biases in their spatial processing remain largely unexplored. This work identifies and characterizes a systematic spatial attention bias where VLMs consistently prioritize describing left-positioned content before right-positioned content in horizontally concatenated images. Through controlled experiments on image pairs using both open-source and closed-source models, we demonstrate that this bias persists across different architectures, with models describing left-positioned content first in approximately 97% of cases under neutral prompting conditions. Testing on an Arabic-finetuned model reveals that the bias persists despite right-to-left language training, ruling out language reading direction as the primary cause. Investigation of training dataset annotation guidelines from PixMo and Visual Genome reveals no explicit left-first ordering instructions, suggesting the bias is consistent with architectural factors rather than explicit training data instructions. These findings reveal fundamental limitations in how current VLMs process spatial information.
Authors: Haolong Zheng, Yekaterina Yegorova, Mark Hasegawa-Johnson
Abstract: Children's speech recognition remains challenging due to substantial acoustic and linguistic variability, limited labeled data, and significant differences from adult speech. Speech foundation models can address these challenges through Speech In-Context Learning (SICL), allowing adaptation to new domains without fine-tuning. However, the effectiveness of SICL depends on how in-context examples are selected. We extend an existing retrieval-based method, Text-Embedding KNN for SICL (TICL), introducing an acoustic reranking step to create TICL+. This extension prioritizes examples that are both semantically and acoustically aligned with the test input. Experiments on four children's speech corpora show that TICL+ achieves up to a 53.3% relative word error rate reduction over zero-shot performance and 37.6% over baseline TICL, highlighting the value of combining semantic and acoustic information for robust, scalable ASR in children's speech.
Authors: Wenkai Li, Lynnette Hui Xian Ng, Andy Liu, Daniel Fried
Abstract: The study of negotiation styles dates back to Aristotle's ethos-pathos-logos rhetoric. Prior efforts primarily studied the success of negotiation agents. Here, we shift the focus towards the styles of negotiation strategies. Our focus is the strategic dialogue board game Diplomacy, which affords rich natural language negotiation and measures of game success. We used LLM-as-a-judge to annotate a large human-human set of Diplomacy games for fine-grained negotiation tactics from a sociologically-grounded taxonomy. Using a combination of the It Takes Two and WebDiplomacy datasets, we demonstrate the reliability of our LLM-as-a-Judge framework and show strong correlations between negotiation features and success in the Diplomacy setting. Lastly, we investigate the differences between LLM and human negotiation strategies and show that fine-tuning can steer LLM agents toward more human-like negotiation behaviors.
Authors: Sudip Chakrabarty, Pappu Bishwas, Rajdeep Chatterjee
Abstract: Speech Emotion Recognition (SER) systems often degrade in performance when exposed to the unpredictable acoustic interference found in real-world environments. Additionally, the opacity of deep learning models hinders their adoption in trust-sensitive applications. To bridge this gap, we propose a Hybrid Transformer-CNN framework that unifies the contextual modeling of Wav2Vec 2.0 with the spectral stability of 1D-Convolutional Neural Networks. Our dual-stream architecture processes raw waveforms to capture long-range temporal dependencies while simultaneously extracting noise-resistant spectral features (MFCC, ZCR, RMSE) via a custom Attentive Temporal Pooling mechanism. We conducted extensive validation across four diverse benchmark datasets: RAVDESS, TESS, SAVEE, and CREMA-D. To rigorously test robustness, we subjected the model to non-stationary acoustic interference using real-world noise profiles from the SAS-KIIT dataset. The proposed framework demonstrates superior generalization and state-of-the-art accuracy across all datasets, significantly outperforming single-branch baselines under realistic environmental interference. Furthermore, we address the ``black-box" problem by integrating SHAP and Score-CAM into the evaluation pipeline. These tools provide granular visual explanations, revealing how the model strategically shifts attention between temporal and spectral cues to maintain reliability in the presence of complex environmental noise.
Authors: Scott Thornton
Abstract: AI assistants produce vulnerable code in 45% of security-relevant scenarios, introducing flaws into production systems at scale. Yet existing secure coding datasets fall short. They lack incident grounding, don't provide the scale modern training requires, and miss the operational security context developers need for production deployments. We present SecureCode v2.0, a production-grade dataset of 1,215 security-focused coding examples that passed structural validation and expert security review. Every example ties to actual documented security incidents with CVE references, provides vulnerable and secure implementations, demonstrates concrete attacks, and includes defense-in-depth operational guidance. The dataset covers 11 vulnerability categories (complete OWASP Top 10:2025 plus AI/ML Security Threats) across 11 languages (Python, JavaScript, Java, Go, PHP, C#, TypeScript, Ruby, Rust, Kotlin, and YAML for infrastructure-as-code). Our quality assurance framework ensures complete incident grounding. Each example includes SIEM integration strategies, infrastructure hardening recommendations (Docker, AppArmor, WAF configurations), and testing approaches using language-appropriate frameworks. The dataset uses a 4-turn conversational structure mirroring actual developer-AI interactions, escalating from basic implementations to advanced security considerations and defense-in-depth guidance. Our contributions: (1) 1,215 rigorously validated examples split into 989 training, 122 validation, and 104 test sets, (2) an automated validation framework ensuring dataset consistency, (3) a 4-turn conversational structure capturing realistic security workflows, (4) comprehensive operational security guidance with SIEM integration strategies, (5) complete language-specific implementation fidelity, and (6) open-source release of data, validation tools, and benchmarking protocols.
Authors: Yuxiang Wei, Zhiqing Sun, Emily McMilin, Jonas Gehring, David Zhang, Gabriel Synnaeve, Daniel Fried, Lingming Zhang, Sida Wang
Abstract: While current software agents powered by large language models (LLMs) and agentic reinforcement learning (RL) can boost programmer productivity, their training data (e.g., GitHub issues and pull requests) and environments (e.g., pass-to-pass and fail-to-pass tests) heavily depend on human knowledge or curation, posing a fundamental barrier to superintelligence. In this paper, we present Self-play SWE-RL (SSR), a first step toward training paradigms for superintelligent software agents. Our approach takes minimal data assumptions, only requiring access to sandboxed repositories with source code and installed dependencies, with no need for human-labeled issues or tests. Grounded in these real-world codebases, a single LLM agent is trained via reinforcement learning in a self-play setting to iteratively inject and repair software bugs of increasing complexity, with each bug formally specified by a test patch rather than a natural language issue description. On the SWE-bench Verified and SWE-Bench Pro benchmarks, SSR achieves notable self-improvement (+10.4 and +7.8 points, respectively) and consistently outperforms the human-data baseline over the entire training trajectory, despite being evaluated on natural language issues absent from self-play. Our results, albeit early, suggest a path where agents autonomously gather extensive learning experiences from real-world software repositories, ultimately enabling superintelligent systems that exceed human capabilities in understanding how systems are constructed, solving novel challenges, and autonomously creating new software from scratch.
Authors: Thanh Dat Hoang, Thanh Trung Huynh, Matthias Weidlich, Thanh Tam Nguyen, Tong Chen, Hongzhi Yin, Quoc Viet Hung Nguyen
Abstract: Text2SQL, the task of generating SQL queries from natural language text, is a critical challenge in data engineering. Recently, Large Language Models (LLMs) have demonstrated superior performance for this task due to their advanced comprehension and generation capabilities. However, privacy and cost considerations prevent companies from using Text2SQL solutions based on external LLMs offered as a service. Rather, small LLMs (SLMs) that are openly available and can hosted in-house are adopted. These SLMs, in turn, lack the generalization capabilities of larger LLMs, which impairs their effectiveness for complex tasks such as Text2SQL. To address these limitations, we propose MATS, a novel Text2SQL framework designed specifically for SLMs. MATS uses a multi-agent mechanism that assigns specialized roles to auxiliary agents, reducing individual workloads and fostering interaction. A training scheme based on reinforcement learning aligns these agents using feedback obtained during execution, thereby maintaining competitive performance despite a limited LLM size. Evaluation results using on benchmark datasets show that MATS, deployed on a single- GPU server, yields accuracy that are on-par with large-scale LLMs when using significantly fewer parameters. Our source code and data are available at https://github.com/thanhdath/mats-sql.
Authors: Maxime Kayser, Maksim Gridnev, Wanting Wang, Max Bain, Aneesh Rangnekar, Avijit Chatterjee, Aleksandr Petrov, Harini Veeraraghavan, Nathaniel C. Swinburne
Abstract: We present brat (brain report alignment transformer), a multi-view representation learning framework for brain magnetic resonance imaging (MRI) trained on MRIs paired with clinical reports. Brain MRIs present unique challenges due to the presence of numerous, highly varied, and often subtle abnormalities that are localized to a few slices within a 3D volume. To address these challenges, we introduce a brain MRI dataset $10\times$ larger than existing ones, containing approximately 80,000 3D scans with corresponding radiology reports, and propose a multi-view pre-training approach inspired by advances in document retrieval. We develop an implicit query-feature matching mechanism and adopt concepts from quality-diversity to obtain multi-view embeddings of MRIs that are aligned with the clinical features given by report sentences. We evaluate our approach across multiple vision-language and vision tasks, demonstrating substantial performance improvements. The brat foundation models are publicly released.
Authors: Kaican Li, Lewei Yao, Jiannan Wu, Tiezheng Yu, Jierun Chen, Haoli Bai, Lu Hou, Lanqing Hong, Wei Zhang, Nevin L. Zhang
Abstract: The ability for AI agents to "think with images" requires a sophisticated blend of reasoning and perception. However, current open multimodal agents still largely fall short on the reasoning aspect crucial for real-world tasks like analyzing documents with dense charts/diagrams and navigating maps. To address this gap, we introduce O3-Bench, a new benchmark designed to evaluate multimodal reasoning with interleaved attention to visual details. O3-Bench features challenging problems that require agents to piece together subtle visual information from distinct image areas through multi-step reasoning. The problems are highly challenging even for frontier systems like OpenAI o3, which only obtains 40.8% accuracy on O3-Bench. To make progress, we propose InSight-o3, a multi-agent framework consisting of a visual reasoning agent (vReasoner) and a visual search agent (vSearcher) for which we introduce the task of generalized visual search -- locating relational, fuzzy, or conceptual regions described in free-form language, beyond just simple objects or figures in natural images. We then present a multimodal LLM purpose-trained for this task via reinforcement learning. As a plug-and-play agent, our vSearcher empowers frontier multimodal models (as vReasoners), significantly improving their performance on a wide range of benchmarks. This marks a concrete step towards powerful o3-like open systems. Our code and dataset can be found at https://github.com/m-Just/InSight-o3 .
Authors: Recep Kaan Karaman, Meftun Akarsu
Abstract: The performance of automatic code documentation generation models depends critically on the quality of the training data used for supervision. However, most existing code documentation datasets are constructed through large scale scraping of public repositories with limited quality control. As a result, they often contain noisy documentation, extensive duplication, and increasing contamination from AI generated content. These issues weaken the supervision signal available to learning-based models and complicate evaluation. We introduce \textbf{Code2Doc}, a quality-first curated dataset for function-level code documentation generation. Code2Doc consists of 13,358 high-quality function-documentation pairs extracted from widely used open-source repositories spanning five programming languages: Python, Java, TypeScript, JavaScript, and C++. The dataset is constructed using a four-stage curation pipeline that enforces documentation completeness and clarity, filters functions based on structural and complexity criteria, removes exact and near-duplicate code, and identifies documentation likely to be AI generated. Starting from 52,069 extracted candidates, only 25.6 percent satisfy all quality constraints. We provide a detailed analysis of the resulting dataset, which achieves a mean documentation quality score of 6.93 out of 10. Overall, 86.9% of samples contain explicit type annotations, and only 2.9\% are flagged as potentially AI generated. Baseline experiments show that fine-tuning a large language model on Code2Doc yields relative improvements of 29.47% in BLEU and 24.04% in ROUGE-L over zero shot performance, despite the modest dataset size. We release both the dataset and the full curation pipeline to support reproducible research on automatic code documentation generation.
Authors: Matilde Marcolli, David Skigin
Abstract: We study the dynamical properties of a Hopf algebra Markov chain with state space the binary rooted forests with labelled leaves. This Markovian dynamical system describes the core computational process of structure formation and transformation in syntax via the Merge operation, according to Chomsky's Minimalism model of generative linguistics. The dynamics decomposes into an ergodic dynamical system with uniform stationary distribution, given by the action of Internal Merge, while the contributions of External Merge and (a minimal form of) Sideward Merge reduce to a simpler Markov chain with state space the set of partitions and with combinatorial weights. The Sideward Merge part of the dynamics prevents convergence to fully formed connected structures (trees), unless the different forms of Merge are weighted by a cost function, as predicted by linguistic theory. Results on the asymptotic behavior of the Perron-Frobenius eigenvalue and eigenvector in this weighted case, obtained in terms of an associated Perron-Frobenius problem in the tropical semiring, show that the usual cost functions (Minimal Search and Resource Restrictions) proposed in the linguistic literature do not suffice to obtain convergence to the tree structures, while an additional optimization property based on the Shannon entropy achieves the expected result for the dynamics. We also comment on the introduction of continuous parameters related to semantic embedding and other computational models, and also on some filtering of the dynamics by coloring rules that model the linguistic filtering by theta roles and phase structure, and on parametric variation and the process of parameter setting in Externalization.
Authors: Maksym Voloshchuk, Bohdana Zarembovska, Mykola Kozlenko
Abstract: Handwritten text recognition and optical character recognition solutions show excellent results with processing data of modern era, but efficiency drops with Latin documents of medieval times. This paper presents a deep learning method to extract text information from handwritten Latin-language documents of the 9th to 11th centuries. The approach takes into account the properties inherent in medieval documents. The paper provides a brief introduction to the field of historical document transcription, a first-sight analysis of the raw data, and the related works and studies. The paper presents the steps of dataset development for further training of the models. The explanatory data analysis of the processed data is provided as well. The paper explains the pipeline of deep learning models to extract text information from the document images, from detecting objects to word recognition using classification models and embedding word images. The paper reports the following results: recall, precision, F1 score, intersection over union, confusion matrix, and mean string distance. The plots of the metrics are also included. The implementation is published on the GitHub repository.
Authors: Ryosuke Korekata, Quanting Xie, Yonatan Bisk, Komei Sugiura
Abstract: In this study, we address the problem of open-vocabulary mobile manipulation, where a robot is required to carry a wide range of objects to receptacles based on free-form natural language instructions. This task is challenging, as it involves understanding visual semantics and the affordance of manipulation actions. To tackle these challenges, we propose Affordance RAG, a zero-shot hierarchical multimodal retrieval framework that constructs Affordance-Aware Embodied Memory from pre-explored images. The model retrieves candidate targets based on regional and visual semantics and reranks them with affordance scores, allowing the robot to identify manipulation options that are likely to be executable in real-world environments. Our method outperformed existing approaches in retrieval performance for mobile manipulation instruction in large-scale indoor environments. Furthermore, in real-world experiments where the robot performed mobile manipulation in indoor environments based on free-form instructions, the proposed method achieved a task success rate of 85%, outperforming existing methods in both retrieval performance and overall task success.
Authors: Akshaj Prashanth Rao, Advait Singh, Saumya Kumaar Saksena, Dhruv Kumar
Abstract: Prompt injection and jailbreaking attacks pose persistent security challenges to large language model (LLM)-based systems. We present an efficient and systematically evaluated defense architecture that mitigates these threats through a lightweight, multi-stage pipeline. Its core component is a semantic filter based on text normalization, TF-IDF representations, and a Linear SVM classifier. Despite its simplicity, this module achieves 93.4% accuracy and 96.5% specificity on held-out data, substantially reducing attack throughput while incurring negligible computational overhead. Building on this efficient foundation, the full pipeline integrates complementary detection and mitigation mechanisms that operate at successive stages, providing strong robustness with minimal latency. In comparative experiments, our SVM-based configuration improves overall accuracy from 35.1% to 93.4% while reducing average time to completion from approximately 450s to 47s, yielding over 10 times lower latency than ShieldGemma. These results demonstrate that the proposed design simultaneously advances defensive precision and efficiency, addressing a core limitation of current model-based moderators. Evaluation across a curated corpus of over 30,000 labeled prompts, including benign, jailbreak, and application-layer injections, confirms that staged, resource-efficient defenses can robustly secure modern LLM-driven applications.
Authors: Ruiqi Ma, Yu Yan, Chunhong Zhang, Minghao Yin, XinChao Liu, Zhihong Jin, Zheng Hu
Abstract: Large Vision-Language Models (LVLMs) bridge the gap between visual and linguistic modalities, demonstrating strong potential across a variety of domains. However, despite significant progress, LVLMs still suffer from severe hallucination issues in object recognition tasks. These models often fail to accurately identify certain objects, leading to text generation that appears fluent but does not correspond to the visual content, which can have serious consequences in real-world applications. Recently, several methods have been proposed to alleviate LVLM hallucinations, but most focus solely on reducing hallucinations in the language modality. To mitigate hallucinations in both the language and visual modalities, we introduce Hallucination Disentangled Decoding (HDD) method that requires no training. HDD enhances the original image by segmenting it and selecting images that augment the original, while also utilizing a blank image to eliminate language prior hallucinations in both the original and segmented images. This design not only reduces the model's dependence on language priors but also enhances its visual performance. (Code: https://github.com/rickeyhhh/Hallucination-Disentangled-Decoding)
URLs: https://github.com/rickeyhhh/Hallucination-Disentangled-Decoding)
Authors: Mahir Labib Dihan, Sadif Ahmed, Md Nafiu Rahman
Abstract: Bangla is a low-resource language for code generation, lacking large-scale annotated datasets and tools to transform natural language specifications into executable programs. This makes Bangla-to-code generation a challenging task requiring innovative solutions. To address this, we introduce BanglaForge, a novel framework for generating code from Bangla function descriptions. BanglaForge leverages a retrieval-augmented dual-model collaboration paradigm with self-refinement, combining in-context learning, llm-based translation, systematic prompt engineering, and iterative self-refinement based on execution feedback, where a coder generates initial solutions and a reviewer enhances them for robustness. On the BLP-2025 Bangla Code Generation benchmark, BanglaForge achieves a competitive Pass@1 accuracy of 84.00%, demonstrating the effectiveness of retrieval, model collaboration, and self-refinement for low-resource Bangla code generation.
Authors: Yayuan Li, Jian Zhang, Jintao Guo, Zihan Cheng, Lei Qi, Yinghuan Shi, Yang Gao
Abstract: The proliferation of pre-trained models has given rise to a wide array of specialised, fine-tuned models. Model merging aims to merge the distinct capabilities of these specialised models into a unified model, requiring minimal or even no additional training. A core objective of model merging is to ensure the merged model retains the behavioural characteristics of the specialised models, typically achieved through feature alignment. We identify that features consist of two critical components: direction and magnitude. Prior research has predominantly focused on directional alignment, while the influence of magnitude remains largely neglected, despite its pronounced vulnerability to perturbations introduced by common merging operations (e.g., parameter fusion and sparsification). Such perturbations to magnitude inevitably lead to feature deviations in the merged model from the specialised models, resulting in subsequent performance degradation. To address this, we propose MAGnItude Calibration (MAGIC), a plug-and-play framework that rectifies layer-wise magnitudes in feature and weight spaces, with three variants. Specifically, our Feature Space Calibration (FSC) realigns the merged model's features using a small set of unlabelled data, while Weight Space Calibration (WSC) extends this calibration to the weight space without requiring additional data. Combining these yields Dual Space Calibration (DSC). Comprehensive experiments demonstrate that MAGIC consistently boosts performance across diverse Computer Vision tasks (+4.3% on eight datasets) and NLP tasks (+8.0% on Llama) without additional training. Our code is available at: https://github.com/lyymuwu/MAGIC
Authors: Jiaren Peng, Hongda Sun, Xuan Tian, Cheng Huang, Zeqing Li, Rui Yan
Abstract: The automation of Cyber Threat Intelligence (CTI) relies heavily on Named Entity Recognition (NER) to extract critical entities from unstructured text. Currently, Large Language Models (LLMs) primarily address this task through retrieval-based In-Context Learning (ICL). This paper analyzes this mainstream paradigm, revealing a fundamental flaw: its success stems not from global semantic similarity but largely from the incidental overlap of entity types within retrieved examples. This exposes the limitations of relying on unreliable implicit induction. To address this, we propose TTPrompt, a framework shifting from implicit induction to explicit instruction. TTPrompt maps the core concepts of CTI's Tactics, Techniques, and Procedures (TTPs) into an instruction hierarchy: formulating task definitions as Tactics, guiding strategies as Techniques, and annotation guidelines as Procedures. Furthermore, to handle the adaptability challenge of static guidelines, we introduce Feedback-driven Instruction Refinement (FIR). FIR enables LLMs to self-refine guidelines by learning from errors on minimal labeled data, adapting to distinct annotation dialects. Experiments on five CTI NER benchmarks demonstrate that TTPrompt consistently surpasses retrieval-based baselines. Notably, with refinement on just 1% of training data, it rivals models fine-tuned on the full dataset. For instance, on LADDER, its Micro F1 of 71.96% approaches the fine-tuned baseline, and on the complex CTINexus, its Macro F1 exceeds the fine-tuned ACLM model by 10.91%.
Authors: Walter Quattrociocchi, Valerio Capraro, Matja\v{z} Perc
Abstract: Large language models (LLMs) are widely described as artificial intelligence, yet their epistemic profile diverges sharply from human cognition. Here we show that the apparent alignment between human and machine outputs conceals a deeper structural mismatch in how judgments are produced. Tracing the historical shift from symbolic AI and information filtering systems to large-scale generative transformers, we argue that LLMs are not epistemic agents but stochastic pattern-completion systems, formally describable as walks on high-dimensional graphs of linguistic transitions rather than as systems that form beliefs or models of the world. By systematically mapping human and artificial epistemic pipelines, we identify seven epistemic fault lines, divergences in grounding, parsing, experience, motivation, causal reasoning, metacognition, and value. We call the resulting condition Epistemia: a structural situation in which linguistic plausibility substitutes for epistemic evaluation, producing the feeling of knowing without the labor of judgment. We conclude by outlining consequences for evaluation, governance, and epistemic literacy in societies increasingly organized around generative AI.
Authors: Yuqiao Tan, Minzheng Wang, Shizhu He, Huanxuan Liao, Chengfeng Zhao, Qiunan Lu, Tian Liang, Jun Zhao, Kang Liu
Abstract: Existing reinforcement learning (RL) approaches treat large language models (LLMs) as a single unified policy, overlooking their internal mechanisms. Understanding how policy evolves across layers and modules is therefore crucial for enabling more targeted optimization and raveling out complex reasoning mechanisms. In this paper, we decompose the language model policy by leveraging the intrinsic split of the Transformer residual stream and the equivalence between the composition of hidden states with the unembedding matrix and the resulting samplable policy. This decomposition reveals Internal Layer Policies, corresponding to contributions from individual layers, and Internal Modular Policies, which align with the self-attention and feed-forward network (FFN) components within each layer. By analyzing the entropy of internal policy, we find that: (a) Early layers keep high entropy for exploration, top layers converge to near-zero entropy for refinement, with convergence patterns varying across model series. (b) LLama's prediction space rapidly converges in the final layer, whereas Qwen-series models, especially Qwen3, exhibit a more human-like, progressively structured reasoning pattern. Motivated by these findings, we propose Bottom-up Policy Optimization (BuPO), a novel RL paradigm that directly optimizes the internal layer policy during early training. By aligning training objective at lower layer, BuPO reconstructs foundational reasoning capabilities and achieves superior performance. Extensive experiments on complex reasoning benchmarks demonstrates the effectiveness of our method. Our code is available at https://github.com/Trae1ounG/BuPO.
Authors: Lifeng Han, Najet Hadj Mohamed, Malak Rassem, Gareth Jones, Alan Smeaton, Goran Nenadic
Abstract: In this work, we introduce the construction of a machine translation (MT) assisted and human-in-the-loop multilingual parallel corpus with annotations of multi-word expressions (MWEs), named AlphaMWE. The MWEs include verbal MWEs (vMWEs) defined in the PARSEME shared task that have a verb as the head of the studied terms. The annotated vMWEs are also bilingually and multilingually aligned manually. The languages covered include Arabic, Chinese, English, German, Italian, and Polish, of which, the Arabic corpus includes both standard and dialectal variations from Egypt and Tunisia. Our original English corpus is extracted from the PARSEME shared task in 2018. We performed machine translation of this source corpus followed by human post-editing and annotation of target MWEs. Strict quality control was applied for error limitation, i.e., each MT output sentence received first manual post-editing and annotation plus a second manual quality rechecking till annotators' consensus is reached. One of our findings during corpora preparation is that accurate translation of MWEs presents challenges to MT systems, as reflected by the outcomes of human-in-the-loop metric HOPE. To facilitate further MT research, we present a categorisation of the error types encountered by MT systems in performing MWE-related translation. To acquire a broader view of MT issues, we selected four popular state-of-the-art MT systems for comparison, namely Microsoft Bing Translator, GoogleMT, Baidu Fanyi, and DeepL MT. Because of the noise removal, translation post-editing, and MWE annotation by human professionals, we believe the AlphaMWE data set will be an asset for both monolingual and cross-lingual research, such as multi-word term lexicography, MT, and information extraction.
Authors: Minho Lee, Junghyun Min, Yerang Kim, Woochul Lee, Yeonsoo Lee
Abstract: Modern generative pre-trained language models excel at open-ended text generation, yet continue to underperform on structure-related tasks such as NER, relation extraction, and semantic role labeling, especially when compared to encoder-only models of similar sizes. While this gap has been attributed to limited structure knowledge, we hypothesize this is also due to the missing connection between the model's internal representations of linguistic structure and the output space used during supervised fine-tuning. We propose the Structured Language Generation Model (SLGM), a model- and task-agnostic framework that reformulates structured prediction as a classification problem through three components: (1) reinforced input formatting with structural cues, (2) loss design, and (3) format-aware decoding that constrains generation to task-valid outputs. Across 5 tasks and 13 datasets, SLGM substantially improves structure prediction without relying on dataset-specific engineering or additional model parameters, strengthening alignment between the model's internal structure representation and output. It outperforms baseline fine-tuning on models of the same size, achieves comparable performance to much larger models when used with <1B parameter models, and acts as a zero-weight adapter that reproduces the benefits of dataset-specific fine-tuning in low-resource settings.
Authors: Bowen Zheng, Ming Ma, Zhongqiao Lin, Tianming Yang
Abstract: Large Language Models (LLMs) have demonstrated remarkable abilities, one of the most important being in-context learning (ICL). With ICL, LLMs can derive the underlying rule from a few demonstrations and provide answers that comply with the rule. Previous work hypothesized that the network creates a task vector in specific positions during ICL. The task vector can be computed by averaging across the dataset. It conveys the overall task information and can thus be considered global. Patching the global task vector allows LLMs to achieve zero-shot performance with dummy inputs comparable to few-shot learning. However, we find that such a global task vector does not exist in all tasks, especially in tasks that rely on rules that can only be inferred from multiple demonstrations, such as categorization tasks. Instead, the information provided by each demonstration is first transmitted to its answer position and forms a local task vector associated with the demonstration. In some tasks but not in categorization tasks, all demonstrations' local task vectors converge in later layers, forming the global task vector. We further show that local task vectors encode a high-level abstraction of rules extracted from the demonstrations. Our study provides novel insights into the mechanism underlying ICL in LLMs, demonstrating how ICL may be achieved through an information aggregation mechanism.
Authors: Pengwei Tang, Xiaolin Hu, Yong Liu
Abstract: Prompt Tuning (PT) enables the adaptation of Pre-trained Large Language Models (PLMs) to downstream tasks by optimizing a small amount of soft virtual tokens, which are prepended to the input token embeddings. Recently, Decomposed Prompt Tuning (DePT) has demonstrated superior adaptation capabilities by decomposing the soft prompt into a shorter soft prompt and a pair of low-rank matrices. The product of the pair of low-rank matrices is added to the input token embeddings to offset them. Additionally, DePT achieves faster inference compared to PT due to the shorter soft prompt. However, in this paper, we find that the position-based token embedding offsets of DePT restrict its ability to generalize across diverse model inputs, and that the shared embedding offsets across many token embeddings result in sub-optimization. To tackle these issues, we introduce Adaptive Decomposed Prompt Tuning (ADePT), which is composed of a short soft prompt and a shallow token-shared feed-forward neural network. ADePT utilizes the token-shared feed-forward neural network to learn the embedding offsets for each token, enabling adaptive embedding offsets that vary according to the model input and better optimization of token embedding offsets. This enables ADePT to achieve superior adaptation performance without requiring more inference time or additional trainable parameters compared to vanilla PT and its variants. In comprehensive experiments across 23 natural language processing tasks and 4 typical PLMs of different scales, ADePT consistently surpasses the other leading parameter-efficient fine-tuning methods, and even outperforms the full fine-tuning in certain scenarios. We also provide a theoretical analysis towards ADePT. Code is available at https://github.com/HungerPWAY/ADePT.
Authors: Shaina Raza, Ashmal Vayani, Aditya Jain, Aravind Narayanan, Vahid Reza Khazaie, Syed Raza Bashir, Elham Dolatabadi, Gias Uddin, Christos Emmanouilidis, Rizwan Qureshi, Mubarak Shah
Abstract: Detecting disinformation that blends manipulated text and images has become increasingly challenging, as AI tools make synthetic content easy to generate and disseminate. While most existing AI safety benchmarks focus on single modality misinformation (i.e., false content shared without intent to deceive), intentional multimodal disinformation, such as propaganda or conspiracy theories that imitate credible news, remains largely unaddressed. We introduce the Vision-Language Disinformation Detection Benchmark (VLDBench), the first large-scale resource supporting both unimodal (text-only) and multimodal (text + image) disinformation detection. VLDBench comprises approximately 62,000 labeled text-image pairs across 13 categories, curated from 58 news outlets. Using a semi-automated pipeline followed by expert review, 22 domain experts invested over 500 hours to produce high-quality annotations with substantial inter-annotator agreement. Evaluations of state-of-the-art Large Language Models (LLMs) and Vision-Language Models (VLMs) on VLDBench show that incorporating visual cues improves detection accuracy by 5 to 35 percentage points over text-only models. VLDBench provides data and code for evaluation, fine-tuning, and robustness testing to support disinformation analysis. Developed in alignment with AI governance frameworks (e.g., the MIT AI Risk Repository), VLDBench offers a principled foundation for advancing trustworthy disinformation detection in multimodal media. Project: https://vectorinstitute.github.io/VLDBench/ Dataset: https://huggingface.co/datasets/vector-institute/VLDBench Code: https://github.com/VectorInstitute/VLDBench
URLs: https://vectorinstitute.github.io/VLDBench/, https://huggingface.co/datasets/vector-institute/VLDBench, https://github.com/VectorInstitute/VLDBench
Authors: Zhuohan Ge, Darian Li, Yubo Wang, Nicole Hu, Xinyi Zhu, Haoyang Li, Xin Zhang, Mingtao Zhang, Shihao Qi, Yuming Xu, Han Shi, Chen Jason Zhang, Qing Li
Abstract: Mental disorders represent a critical global health challenge, and social media is increasingly viewed as a vital resource for real-time digital phenotyping and intervention. To leverage this data, large language models (LLMs) have been introduced, offering stronger semantic understanding and reasoning than traditional deep learning, thereby enhancing the explainability of detection results. Despite the growing prominence of LLMs in this field, there is a scarcity of scholarly works that systematically synthesize how advanced enhancement techniques, specifically Retrieval-Augmented Generation (RAG) and Agentic systems, can be utilized to address these reliability and reasoning limitations. Here, we systematically survey the evolving landscape of LLM-based methods for social media mental disorder analysis, spanning standard pre-trained language models, RAG to mitigate hallucinations and contextual gaps, and agentic systems for autonomous reasoning and multi-step intervention. We organize existing work by technical paradigm and clinical target, extending beyond common internalizing disorders to include psychotic disorders and externalizing behaviors. Additionally, the paper comprehensively evaluates the performance of LLMs, including the impact of RAG, across various tasks. This work establishes a unified benchmark for the field, paving the way for the development of trustworthy, autonomous AI systems that can deliver precise and explainable mental health support.
Authors: Chenghao Li, Chaoning Zhang, Yi Lu, Jiaquan Zhang, Qigan Sun, Xudong Wang, Jiwei Wei, Guoqing Wang, Yang Yang, Heng Tao Shen
Abstract: Chain-of-Thought (CoT) prompting enhances the reasoning of large language models (LLMs) by decomposing problems into sequential steps, mimicking human logic and reducing errors. However, complex tasks with vast solution spaces and vague constraints often exceed the capacity of a single reasoning chain. Inspired by Minimal Free Resolution (MFR) in commutative algebra and algebraic geometry, we propose Syzygy of Thoughts (SoT)-a novel framework that extends CoT by introducing auxiliary, interrelated reasoning paths. SoT captures deeper logical dependencies, enabling more robust and structured problem-solving. MFR decomposes a module into a sequence of free modules with minimal rank, providing a structured analytical approach to complex systems. This method introduces the concepts of "Module", "Betti numbers","Freeness", "Mapping", "Exactness" and "Minimality", enabling the systematic decomposition of the original complex problem into logically complete minimal subproblems while preserving key problem features and reducing reasoning length. We tested SoT across diverse datasets (e.g., GSM8K, MATH) and models (e.g., GPT-4o-mini, Qwen2.5), achieving inference accuracy that matches or surpasses mainstream CoTs standards. Additionally, by aligning the sampling process with algebraic constraints, our approach enhances the scalability of inference time in LLMs, ensuring both transparent reasoning and high performance. Our code will be publicly available at https://github.com/dlMARiA/Syzygy-of-thoughts.
Authors: Fadel M. Megahed, Ying-Ju Chen, L. Allision Jones-Farmer, Younghwa Lee, Jiawei Brooke Wang, Inez M. Zwetsloot
Abstract: This study introduces a framework for evaluating consistency in large language model (LLM) binary text classification, addressing the lack of established reliability assessment methods. Adapting psychometric principles, we determine sample size requirements, develop metrics for invalid responses, and evaluate intra- and inter-rater reliability. Our case study examines financial news sentiment classification across 14 LLMs (including claude-3-7-sonnet, gpt-4o, deepseek-r1, gemma3, llama3.2, phi4, and command-r-plus), with five replicates per model on 1,350 articles. Models demonstrated high intra-rater consistency, achieving perfect agreement on 90-98% of examples, with minimal differences between expensive and economical models from the same families. When validated against StockNewsAPI labels, models achieved strong performance (accuracy 0.76-0.88), with smaller models like gemma3:1B, llama3.2:3B, and claude-3-5-haiku outperforming larger counterparts. All models performed at chance when predicting actual market movements, indicating task constraints rather than model limitations. Our framework provides systematic guidance for LLM selection, sample size planning, and reliability assessment, enabling organizations to optimize resources for classification tasks.
Authors: Haoyi Qiu, Kung-Hsiang Huang, Ruichen Zheng, Jiao Sun, Nanyun Peng
Abstract: Large vision-language models (LVLMs) are increasingly deployed in globally distributed applications, such as tourism assistants, yet their ability to produce culturally appropriate responses remains underexplored. Existing multimodal safety benchmarks primarily focus on physical safety and overlook violations rooted in cultural norms, which can result in symbolic harm. To address this gap, we introduce CROSS, a benchmark designed to assess the cultural safety reasoning capabilities of LVLMs. CROSS includes 1,284 multilingual visually grounded queries from 16 countries, three everyday domains, and 14 languages, where cultural norm violations emerge only when images are interpreted in context. We propose CROSS-Eval, an intercultural theory-based framework that measures four key dimensions: cultural awareness, norm education, compliance, and helpfulness. Using this framework, we evaluate 21 leading LVLMs, including mixture-of-experts models and reasoning models. Results reveal significant cultural safety gaps: the best-performing model achieves only 61.79% in awareness and 37.73% in compliance. While some open-source models reach GPT-4o-level performance, they still fall notably short of proprietary models. Our results further show that increasing reasoning capacity improves cultural alignment but does not fully resolve the issue. To improve model performance, we develop two enhancement strategies: supervised fine-tuning with culturally grounded, open-ended data and preference tuning with contrastive response pairs that highlight safe versus unsafe behaviors. These methods substantially improve GPT-4o's cultural awareness (+60.14%) and compliance (+55.2%), while preserving general multimodal capabilities with minimal performance reduction on general multimodal understanding benchmarks.
Authors: Jie Ma, Ning Qu, Zhitao Gao, Rui Xing, Jun Liu, Hongbin Pei, Jiang Xie, Linyun Song, Pinghui Wang, Jing Tao, Zhou Su
Abstract: Knowledge graph-based retrieval-augmented generation seeks to mitigate hallucinations in Large Language Models (LLMs) caused by insufficient or outdated knowledge. However, existing methods often fail to fully exploit the prior knowledge embedded in knowledge graphs (KGs), particularly their structural information and explicit or implicit constraints. The former can enhance the faithfulness of LLMs' reasoning, while the latter can improve the reliability of response generation. Motivated by these, we propose a trustworthy reasoning framework, termed Deliberation over Priors (DP), which sufficiently utilizes the priors contained in KGs. Specifically, DP adopts a progressive knowledge distillation strategy that integrates structural priors into LLMs through a combination of supervised fine-tuning and Kahneman-Tversky optimization, thereby improving the faithfulness of relation path generation. Furthermore, our framework employs a reasoning-introspection strategy, which guides LLMs to perform refined reasoning verification based on extracted constraint priors, ensuring the reliability of response generation. Extensive experiments on three benchmark datasets demonstrate that DP achieves new state-of-the-art performance, especially a Hit@1 improvement of 13% on the ComplexWebQuestions dataset, and generates highly trustworthy responses. We also conduct various analyses to verify its flexibility and practicality. The code is available at https://github.com/reml-group/Deliberation-on-Priors.
Authors: Christine Bauer, Li Chen, Nicola Ferro, Norbert Fuhr, Avishek Anand, Timo Breuer, Guglielmo Faggioli, Ophir Frieder, Hideo Joho, Jussi Karlgren, Johannes Kiesel, Bart P. Knijnenburg, Aldo Lipani, Lien Michiels, Andrea Papenmeier, Maria Soledad Pera, Mark Sanderson, Scott Sanner, Benno Stein, Johanne R. Trippas, Karin Verspoor, Martijn C Willemsen
Abstract: During the workshop, we deeply discussed what CONversational Information ACcess (CONIAC) is and its unique features, proposing a world model abstracting it, and defined the Conversational Agents Framework for Evaluation (CAFE) for the evaluation of CONIAC systems, consisting of six major components: 1) goals of the system's stakeholders, 2) user tasks to be studied in the evaluation, 3) aspects of the users carrying out the tasks, 4) evaluation criteria to be considered, 5) evaluation methodology to be applied, and 6) measures for the quantitative criteria chosen.
Authors: Bowen Zhang, Jun Ma, Fuqiang Niu, Li Dong, Jinzhou Cao, Genan Dai
Abstract: Zero-shot stance detection (ZSSD) seeks to determine the stance of text toward previously unseen targets, a task critical for analyzing dynamic and polarized online discourse with limited labeled data. While large language models (LLMs) offer zero-shot capabilities, prompting-based approaches often fall short in handling complex reasoning and lack robust generalization to novel targets. Meanwhile, LLM-enhanced methods still require substantial labeled data and struggle to move beyond instance-level patterns, limiting their interpretability and adaptability. Inspired by cognitive science, we propose the Cognitive Inductive Reasoning Framework (CIRF), a schema-driven method that bridges linguistic inputs and abstract reasoning via automatic induction and application of cognitive reasoning schemas. CIRF abstracts first-order logic patterns from raw text into multi-relational schema graphs in an unsupervised manner, and leverages a schema-enhanced graph kernel model to align input structures with schema templates for robust, interpretable zero-shot inference. Extensive experiments on SemEval-2016, VAST, and COVID-19-Stance benchmarks demonstrate that CIRF not only establishes new state-of-the-art results, but also achieves comparable performance with just 30\% of the labeled data, demonstrating its strong generalization and efficiency in low-resource settings.
Authors: Xiang Li, Chong Zhang, Jia Wang, Fangyu Wu, Yushi Li, Xiaobo Jin
Abstract: As the scale and complexity of jailbreaking attacks on large language models (LLMs) continue to escalate, their efficiency and practical applicability are constrained, posing a profound challenge to LLM security. Jailbreaking techniques have advanced from manual prompt engineering to automated methodologies. Recent advances have automated jailbreaking approaches that harness LLMs to generate jailbreak instructions and adversarial examples, delivering encouraging results. Nevertheless, these methods universally include an LLM generation phase, which, due to the complexities of deploying and reasoning with LLMs, impedes effective implementation and broader adoption. To mitigate this issue, we introduce \textbf{Adversarial Prompt Distillation}, an innovative framework that integrates masked language modeling, reinforcement learning, and dynamic temperature control to distill LLM jailbreaking prowess into smaller language models (SLMs). This methodology enables efficient, robust jailbreak attacks while maintaining high success rates and accommodating a broader range of application contexts. Empirical evaluations affirm the approach's superiority in attack efficacy, resource optimization, and cross-model versatility. Our research underscores the practicality of transferring jailbreak capabilities to SLMs, reveals inherent vulnerabilities in LLMs, and provides novel insights to advance LLM security investigations. Our code is available at: https://github.com/lxgem/Efficient_and_Stealthy_Jailbreak_Attacks_via_Adversarial_Prompt.
URLs: https://github.com/lxgem/Efficient_and_Stealthy_Jailbreak_Attacks_via_Adversarial_Prompt.
Authors: Sahil Kale
Abstract: When artificial intelligence mistakes memorization for intelligence, it creates a dangerous mirage of reasoning. Existing studies treat memorization and self-knowledge deficits in LLMs as separate issues and do not recognize an intertwining link that degrades the trustworthiness of LLM responses. In our study, we utilize a novel framework to ascertain if LLMs genuinely learn reasoning patterns from training data or merely memorize them to assume competence across problems of similar complexity focused on STEM domains. Our analysis shows a noteworthy problem in generalization: LLMs draw confidence from memorized solutions to infer a higher self-knowledge about their reasoning ability, which manifests as an over 45% inconsistency in feasibility assessments when faced with self-validated, logically coherent task perturbations. This effect is most pronounced in science and medicine domains, which tend to have maximal standardized jargon and problems, further confirming our approach. Significant wavering within the self-knowledge of LLMs also shows flaws in current architectures and training patterns, highlighting the need for techniques that ensure a balanced, consistent stance on models' perceptions of their own knowledge for maximum AI explainability and trustworthiness. Our code and results are available publicly at https://github.com/Sahil-R-Kale/mirage_of_mastery
Authors: Renad Al-Monef, Hassan Alhuzali, Nora Alturayeif, Ashwag Alasmari
Abstract: As large language models (LLMs) become increasingly central to Arabic NLP applications, evaluating their understanding of regional dialects and cultural nuances is essential, particularly in linguistically diverse settings like Saudi Arabia. This paper introduces Absher, a comprehensive benchmark specifically designed to assess LLMs performance across major Saudi dialects. \texttt{Absher} comprises over 18,000 multiple-choice questions spanning six distinct categories: Meaning, True/False, Fill-in-the-Blank, Contextual Usage, Cultural Interpretation, and Location Recognition. These questions are derived from a curated dataset of dialectal words, phrases, and proverbs sourced from various regions of Saudi Arabia. We evaluate several state-of-the-art LLMs, including multilingual and Arabic-specific models. We also provide detailed insights into their capabilities and limitations. Our results reveal notable performance gaps, particularly in tasks requiring cultural inference or contextual understanding. Our findings highlight the urgent need for dialect-aware training and culturally aligned evaluation methodologies to improve LLMs performance in real-world Arabic applications.
Authors: Sungwoo Han, Jingun Kwon, Hidetaka Kamigaito, Manabu Okumura
Abstract: Recent studies have explored various approaches for treating candidate named entity spans as both source and target sequences in named entity recognition (NER) by leveraging large language models (LLMs). Although previous approaches have successfully generated candidate named entity spans with suitable labels, they rely solely on input context information when using LLMs, particularly, ChatGPT. However, NER inherently requires capturing detailed labeling requirements with input context information. To address this issue, we propose a novel method that leverages code-based prompting to improve the capabilities of LLMs in understanding and performing NER. By embedding code within prompts, we provide detailed BIO schema instructions for labeling, thereby exploiting the ability of LLMs to comprehend long-range scopes in programming languages. Experimental results demonstrate that the proposed code-based prompting method outperforms conventional text-based prompting on ten benchmarks across English, Arabic, Finnish, Danish, and German datasets, indicating the effectiveness of explicitly structuring NER instructions. We also verify that combining the proposed code-based prompting method with the chain-of-thought prompting further improves performance.
Authors: Gideon Vos, Maryam Ebrahimpour, Liza van Eijk, Zoltan Sarnyai, Mostafa Rahimi Azghadi
Abstract: Understanding how emotional expression in language relates to brain function is a challenge in computational neuroscience and affective computing. Traditional neuroimaging is costly and lab-bound, but abundant digital text offers new avenues for emotion-brain mapping. Prior work has largely examined neuroimaging-based emotion localization or computational text analysis separately, with little integration. We propose a computational framework that maps textual emotional content to anatomically defined brain regions without requiring neuroimaging. Using OpenAI's text-embedding-ada-002, we generate high-dimensional semantic representations, apply dimensionality reduction and clustering to identify emotional groups, and map them to 18 brain regions linked to emotional processing. Three experiments were conducted: i) analyzing conversational data from healthy vs. depressed subjects (DIAC-WOZ dataset) to compare mapping patterns, ii) applying the method to the GoEmotions dataset and iii) comparing human-written text with large language model (LLM) responses to assess differences in inferred brain activation. Emotional intensity was scored via lexical analysis. Results showed neuroanatomically plausible mappings with high spatial specificity. Depressed subjects exhibited greater limbic engagement tied to negative affect. Discrete emotions were successfully differentiated. LLM-generated text matched humans in basic emotion distribution but lacked nuanced activation in empathy and self-referential regions (medial prefrontal and posterior cingulate cortex). This cost-effective, scalable approach enables large-scale analysis of naturalistic language, distinguishes between clinical populations, and offers a brain-based benchmark for evaluating AI emotional expression.
Authors: Ziyi Yang, Weizhou Shen, Chenliang Li, Ruijun Chen, Fanqi Wan, Ming Yan, Xiaojun Quan, Fei Huang
Abstract: Progress in long-context reasoning for large language models (LLMs) has lagged behind other recent advances. This gap arises not only from the intrinsic difficulty of processing long texts, but also from the scarcity of reliable human annotations and programmatically verifiable reward signals. In this paper, we propose SPELL, a multi-role self-play reinforcement learning framework that enables scalable, label-free optimization for long-context reasoning. SPELL integrates three cyclical roles-questioner, responder, and verifier-within a single model to enable continual self-improvement. The questioner generates questions from raw documents paired with reference answers; the responder learns to solve these questions based on the documents; and the verifier evaluates semantic equivalence between the responder's output and the questioner's reference answer, producing reward signals to guide continual training. To stabilize training, we introduce an automated curriculum that gradually increases document length and a reward function that adapts question difficulty to the model's evolving capabilities. Extensive experiments on six long-context benchmarks show that SPELL consistently improves performance across diverse LLMs and outperforms equally sized models fine-tuned on large-scale annotated data. Notably, SPELL achieves an average 7.6-point gain in pass@8 on the strong reasoning model Qwen3-30B-A3B-Thinking, raising its performance ceiling and showing promise for scaling to even more capable models.
Authors: Donghoon Jung, Jiwoo Choi, Songeun Chae, Seohyon Jung
Abstract: Evaluations of large language models (LLMs)' creativity have focused primarily on the quality of their outputs rather than the processes that shape them. This study takes a process-oriented approach, drawing on narratology to examine LLMs as computational authors. We introduce constraint-based decision-making as a lens for authorial creativity. Using controlled prompting to assign authorial personas, we analyze the creative preferences of the models. Our findings show that LLMs consistently emphasize Style over other elements, including Character, Event, and Setting. By also probing the reasoning the models provide for their choices, we show that distinctive profiles emerge across models and argue that our approach provides a novel systematic tool for analyzing AI's authorial creativity.
Authors: Asif Shahriar, Md Nafiu Rahman, Sadif Ahmed, Farig Sadeque, Md Rizwan Parvez
Abstract: In this work we present the first holistic survey of the agentic security landscape, structuring the field around three fundamental pillars: Applications, Threats, and Defenses. We provide a comprehensive taxonomy of over 160 papers, explaining how agents are used in downstream cybersecurity applications, inherent threats to agentic systems, and countermeasures designed to protect them. A detailed cross-cutting analysis shows emerging trends in agent architecture while revealing critical research gaps in model and modality coverage. A complete and continuously updated list of all surveyed papers is publicly available at https://github.com/kagnlp/Awesome-Agentic-Security.
Authors: Tuan T. Nguyen, John Le, Thai T. Vu, Willy Susilo, Heath Cooper
Abstract: Large language models (LLMs) achieve impressive performance across diverse tasks yet remain vulnerable to jailbreak attacks that bypass safety mechanisms. We present RAID (Refusal-Aware and Integrated Decoding), a framework that systematically probes these weaknesses by crafting adversarial suffixes that induce restricted content while preserving fluency. RAID relaxes discrete tokens into continuous embeddings and optimizes them with a joint objective that (i) encourages restricted responses, (ii) incorporates a refusal-aware regularizer to steer activations away from refusal directions in embedding space, and (iii) applies a coherence term to maintain semantic plausibility and non-redundancy. After optimization, a critic-guided decoding procedure maps embeddings back to tokens by balancing embedding affinity with language-model likelihood. This integration yields suffixes that are both effective in bypassing defenses and natural in form. Experiments on multiple open-source LLMs show that RAID achieves higher attack success rates with fewer queries and lower computational cost than recent white-box and black-box baselines. These findings highlight the importance of embedding-space regularization for understanding and mitigating LLM jailbreak vulnerabilities.
Authors: Huiyuan Xie, Chenyang Li, Huining Zhu, Chubin Zhang, Yuxiao Ye, Zhenghao Liu, Zhiyuan Liu
Abstract: Legal reasoning is a fundamental component of legal analysis and decision-making. Existing computational approaches to legal reasoning predominantly rely on generic reasoning frameworks such as syllogism, which do not comprehensively examine the nuanced process of legal reasoning. Moreover, current research has largely focused on criminal cases, with insufficient modeling for civil cases. In this work, we present a novel framework to explicitly model legal reasoning in the analysis of Chinese tort-related civil cases. We first operationalize the legal reasoning process in tort analysis into the three-module LexChain framework, with each module consisting of multiple finer-grained sub-steps. Informed by the LexChain framework, we introduce the task of tort legal reasoning and construct an evaluation benchmark to systematically assess the critical steps within analytical reasoning chains for tort analysis. Leveraging this benchmark, we evaluate existing large language models for their legal reasoning ability in civil tort contexts. Our results indicate that current models still fall short in accurately handling crucial elements of tort legal reasoning. Furthermore, we introduce several baseline approaches that explicitly incorporate LexChain-style reasoning through prompting or post-training. The proposed baselines achieve significant improvements in tort-related legal reasoning and generalize well to related legal analysis tasks, demonstrating the value of explicitly modeling legal reasoning chains to enhance the reasoning capabilities of language models.
Authors: Alan Saji, Raj Dabre, Anoop Kunchukuttan, Ratish Puduppully
Abstract: Large Reasoning Models (LRMs) achieve strong performance on mathematical, scientific, and other question-answering tasks, but their multilingual reasoning abilities remain underexplored. When presented with non-English questions, LRMs often default to reasoning in English, raising concerns about interpretability and the handling of linguistic and cultural nuances. We systematically compare an LRM's reasoning in English versus the language of the question. Our evaluation spans two tasks: MGSM and GPQA Diamond. Beyond measuring answer accuracy, we also analyze cognitive attributes in the reasoning traces. We find that English reasoning traces exhibit a substantially higher presence of these cognitive behaviors, and that reasoning in English generally yields higher final-answer accuracy, with the performance gap increasing as tasks become more complex. However, this English-centric strategy is susceptible to a key failure mode - getting "Lost in Translation," where translation steps lead to errors that would have been avoided by question's language reasoning.
Authors: Linyang He, Tianjun Zhong, Richard Antonello, Gavin Mischler, Micah Goldblum, Nima Mesgarani
Abstract: Understanding how the human brain progresses from processing simple linguistic inputs to performing high-level reasoning is a fundamental challenge in neuroscience. While modern large language models (LLMs) are increasingly used to model neural responses to language, their internal representations are highly "entangled," mixing information about lexicon, syntax, meaning, and reasoning. This entanglement biases conventional brain encoding analyses toward linguistically shallow features (e.g., lexicon and syntax), making it difficult to isolate the neural substrates of cognitively deeper processes. Here, we introduce a residual disentanglement method that computationally isolates these components. By first probing an LM to identify feature-specific layers, our method iteratively regresses out lower-level representations to produce four nearly orthogonal embeddings for lexicon, syntax, meaning, and, critically, reasoning. We used these disentangled embeddings to model intracranial (ECoG) brain recordings from neurosurgical patients listening to natural speech. We show that: 1) This isolated reasoning embedding exhibits unique predictive power, accounting for variance in neural activity not explained by other linguistic features and even extending to the recruitment of visual regions beyond classical language areas. 2) The neural signature for reasoning is temporally distinct, peaking later (~350-400ms) than signals related to lexicon, syntax, and meaning, consistent with its position atop a processing hierarchy. 3) Standard, non-disentangled LLM embeddings can be misleading, as their predictive success is primarily attributable to linguistically shallow features, masking the more subtle contributions of deeper cognitive processing.
Authors: Yudong Li, Peiru Yang, Feng Huang, Zhongliang Yang, Kecheng Wang, Haitian Li, Baocheng Chen, Xingyu An, Ziyu Liu, Youdan Yang, Kejiang Chen, Sifang Wan, Xu Wang, Yufei Sun, Liyan Wu, Ruiqi Zhou, Wenya Wen, Xingchi Gu, Tianxin Zhang, Yue Gao, Yongfeng Huang
Abstract: We introduce LiveSecBench, a continuously updated safety benchmark specifically for Chinese-language LLM application scenarios. LiveSecBench constructs a high-quality and unique dataset through a pipeline that combines automated generation with human verification. By periodically releasing new versions to expand the dataset and update evaluation metrics, LiveSecBench provides a robust and up-to-date standard for AI safety. In this report, we introduce our second release v251215, which evaluates across five dimensions (Public Safety, Fairness & Bias, Privacy, Truthfulness, and Mental Health Safety.) We evaluate 57 representative LLMs using an ELO rating system, offering a leaderboard of the current state of Chinese LLM safety. The result is available at https://livesecbench.intokentech.cn/.
Authors: Aashray Reddy, Andrew Zagula, Nicholas Saban
Abstract: Large Language Models (LLMs) remain vulnerable to jailbreaking attacks where adversarial prompts elicit harmful outputs. Yet most evaluations focus on single-turn interactions while real-world attacks unfold through adaptive multi-turn conversations. We present AutoAdv, a training-free framework for automated multi-turn jailbreaking that achieves an attack success rate of up to 95% on Llama-3.1-8B within six turns, a 24% improvement over single-turn baselines. AutoAdv uniquely combines three adaptive mechanisms: a pattern manager that learns from successful attacks to enhance future prompts, a temperature manager that dynamically adjusts sampling parameters based on failure modes, and a two-phase rewriting strategy that disguises harmful requests and then iteratively refines them. Extensive evaluation across commercial and open-source models (Llama-3.1-8B, GPT-4o mini, Qwen3-235B, Mistral-7B) reveals persistent vulnerabilities in current safety mechanisms, with multi-turn attacks consistently outperforming single-turn approaches. These findings demonstrate that alignment strategies optimized for single-turn interactions fail to maintain robustness across extended conversations, highlighting an urgent need for multi-turn-aware defenses.
Authors: Mihael Arcan, David-Paul Niland
Abstract: Mental health disorders affect over one-fifth of adults globally, yet detecting such conditions from text remains challenging due to the subtle and varied nature of symptom expression. This study evaluates multiple approaches for mental health detection, comparing Large Language Models (LLMs) such as Llama and GPT with classical machine learning and transformer-based architectures including BERT, XLNet, and Distil-RoBERTa. Using the DAIC-WOZ dataset of clinical interviews, we fine-tuned models for anxiety, depression, and stress classification and applied synthetic data generation to mitigate class imbalance. Results show that Distil-RoBERTa achieved the highest F1 score (0.883) for GAD-2, while XLNet outperformed others on PHQ tasks (F1 up to 0.891). For stress detection, a zero-shot synthetic approach (SD+Zero-Shot-Basic) reached an F1 of 0.884 and ROC AUC of 0.886. Findings demonstrate the effectiveness of transformer-based models and highlight the value of synthetic data in improving recall and generalization. However, careful calibration is required to prevent precision loss. Overall, this work emphasizes the potential of combining advanced language models and data augmentation to enhance automated mental health assessment from text.
Authors: Yidan Sun, Mengying Zhu, Feiyue Chen, Yangyang Wu, Xiaolei Dan, Mengyuan Yang, Xiaolin Zheng, Shenglin Ben
Abstract: Large language models (LLMs) have demonstrated impressive performance in text generation tasks; however, their embedding spaces often suffer from the isotropy problem, resulting in poor discrimination of domain-specific terminology, particularly in legal and financial contexts. This weakness in terminology-level representation can severely hinder downstream tasks such as legal judgment prediction or financial risk analysis, where subtle semantic distinctions are critical. To address this problem, we propose TermGPT, a multi-level contrastive fine-tuning framework designed for terminology adaptation. We first construct a sentence graph to capture semantic and structural relations, and generate semantically consistent yet discriminative positive and negative samples based on contextual and topological cues. We then devise a multi-level contrastive learning approach at both the sentence and token levels, enhancing global contextual understanding and fine-grained terminology discrimination. To support robust evaluation, we construct the first financial terminology dataset derived from official regulatory documents. Experiments show that TermGPT outperforms existing baselines in term discrimination tasks within the finance and legal domains.
Authors: Gyubok Lee, Woosog Chay, Edward Choi
Abstract: Recent advances in Large Language Models (LLMs) have enabled the development of text-to-SQL models that allow clinicians to query structured data stored in Electronic Health Records (EHRs) using natural language. However, deploying these models for EHR question answering (QA) systems in safety-critical clinical environments remains challenging: incorrect SQL queries-whether caused by model errors or problematic user inputs-can undermine clinical decision-making and jeopardize patient care. While prior work has mainly focused on improving SQL generation accuracy or filtering questions before execution, there is a lack of a unified benchmark for evaluating independent post-hoc verification mechanisms (i.e., a component that inspects and validates the generated SQL before execution), which is crucial for safe deployment. To fill this gap, we introduce SCARE, a benchmark for evaluating methods that function as a post-hoc safety layer in EHR QA systems. SCARE evaluates the joint task of (1) classifying question answerability (i.e., determining whether a question is answerable, ambiguous, or unanswerable) and (2) verifying or correcting candidate SQL queries. The benchmark comprises 4,200 triples of questions, candidate SQL queries, and expected model outputs, grounded in the MIMIC-III, MIMIC-IV, and eICU databases. It covers a diverse set of questions and corresponding candidate SQL queries generated by seven different text-to-SQL models, ensuring a realistic and challenging evaluation. Using SCARE, we benchmark a range of approaches-from two-stage methods to agentic frameworks. Our experiments reveal a critical trade-off between question classification and SQL error correction, highlighting key challenges and outlining directions for future research.
Authors: Lin Yu, Xiaofei Han, Yifei Kang, Chiung-Yi Tseng, Danyang Zhang, Ziqian Bi, Zhimo Han
Abstract: Recent advances in large language models (LLMs) have enabled fluent dialogue systems, but most remain reactive and struggle in emotionally rich, goal-oriented settings such as marketing conversations. To address this limitation, we propose AffectMind, a multimodal affective dialogue agent that performs proactive reasoning and dynamic knowledge grounding to sustain emotionally aligned and persuasive interactions. AffectMind combines three components: a Proactive Knowledge Grounding Network (PKGN) that continuously updates factual and affective context from text, vision, and prosody; an Emotion--Intent Alignment Model (EIAM) that jointly models user emotion and purchase intent to adapt persuasion strategies; and a Reinforced Discourse Loop (RDL) that optimizes emotional coherence and engagement via reinforcement signals from user responses. Experiments on two newly curated marketing dialogue datasets, MM-ConvMarket and AffectPromo, show that AffectMind outperforms strong LLM-based baselines in emotional consistency (+26\%), persuasive success rate (+19\%), and long-term user engagement (+23\%), highlighting emotion-grounded proactivity as a key capability for commercial multimodal agents.
Authors: Ting-Yao Hu, Hema Swetha Koppula, Hadi Pouransari, Cem Koc, Oncel Tuzel, Raviteja Vemulapalli
Abstract: Large Language Models (LLMs) often suffer from hallucinations: output content that is not grounded in the input context, when performing long-form text generation tasks such as summarization. Prior works have shown that hallucinations can be reduced by iteratively critiquing and refining previously generated outputs using either the same model or a more powerful teacher model as the critique. However, these approaches either require additional test-time compute or assume access to more powerful teacher models, making them costly and less practical. In this work, we propose Self Critique and Refinement-based Preference Optimization (SCRPO), which is a self-supervised training framework that first constructs a preference dataset by leveraging the LLM's own critique and refinement capabilities, and then applies preference learning to improve the same LLM for faithful summarization. Experiments on three summarization benchmarks (XSUM CNNDM and SAMSum), demonstrate that our approach outperforms state-of-the-art self-supervised learning methods in terms of faithfulness metrics while either maintaining or improving other metrics that measure the overall quality of the summary. Moreover, compared to test-time refinement, our approach not only improves efficiency but also results in more faithful summaries.
Authors: Frederic Blum
Abstract: The statistical over-representation of phonological features in the basic vocabulary of languages is often interpreted as reflecting potentially universal sound symbolic patterns. However, most of those results have not been tested explicitly for reproducibility and might be prone to biases in the study samples or models. Many studies on the topic do not adequately control for genealogical and areal dependencies between sampled languages, casting doubts on the robustness of the results. In this study, we test the robustness of a recent study on sound symbolism of basic vocabulary concepts which analyzed 245 languages.The new sample includes data on 2864 languages from Lexibank. We modify the original model by adding statistical controls for spatial and phylogenetic dependencies between languages. The new results show that most of the previously observed patterns are not robust, and in fact many patterns disappear completely when adding the genealogical and areal controls. A small number of patterns, however, emerges as highly stable even with the new sample. Through the new analysis, we are able to assess the distribution of sound symbolism on a larger scale than previously. The study further highlights the need for testing all universal claims on language for robustness on various levels.
Authors: Sherman Wong, Zhenting Qi, Zhaodong Wang, Nathan Hu, Samuel Lin, Jun Ge, Erwin Gao, Wenlin Chen, Yilun Du, Minlan Yu, Ying Zhang
Abstract: Real-world software engineering tasks require coding agents that can operate over massive repositories, sustain long-horizon sessions, and reliably coordinate complex toolchains at test time. Existing research-grade coding agents offer transparency but struggle when scaled to heavier, production-level workloads, while production-grade systems achieve strong practical performance but provide limited extensibility, interpretability, and controllability. We introduce the Confucius Code Agent (CCA), a software engineering agent that can operate at large-scale codebases. CCA is built on top of the Confucius SDK, an agent development platform structured around three complementary perspectives: Agent Experience (AX), User Experience (UX), and Developer Experience (DX). The SDK integrates a unified orchestrator with hierarchical working memory for long-context reasoning, a persistent note-taking system for cross-session continual learning, and a modular extension system for reliable tool use. In addition, we introduce a meta-agent that automates the synthesis, evaluation, and refinement of agent configurations through a build-test-improve loop, enabling rapid adaptation to new tasks, environments, and tool stacks. Instantiated with these mechanisms, CCA demonstrates strong performance on real-world software engineering tasks. On SWE-Bench-Pro, CCA reaches a Resolve@1 of 54.3%, exceeding prior research baselines and comparing favorably to commercial results, under identical repositories, model backends, and tool access.
Authors: Hong Su
Abstract: Large Language Models (LLMs) excel at extracting common patterns from large-scale corpora, yet they struggle with rare, low-resource, or previously unseen scenarios-such as niche hardware deployment issues or irregular IoT device behaviors-because such cases are sparsely represented in training data. Moreover, LLMs rely primarily on implicit parametric memory, which limits their ability to explicitly acquire, recall, and refine methods, causing them to behave predominantly as intuition-driven predictors rather than deliberate, method-oriented learners. Inspired by how humans learn from rare experiences, this paper proposes a human-inspired learning framework that integrates two complementary mechanisms. The first, Obvious Record, explicitly stores cause--result (or question--solution) relationships as symbolic memory, enabling persistent learning even from single or infrequent encounters. The second, Maximum-Entropy Method Discovery, prioritizes and preserves methods with high semantic dissimilarity, allowing the system to capture diverse and underrepresented strategies that are typically overlooked by next-token prediction. Verification on a benchmark of 60 semantically diverse question--solution pairs demonstrates that the proposed entropy-guided approach achieves stronger coverage of unseen questions and significantly greater internal diversity than a random baseline, confirming its effectiveness in discovering more generalizable and human-inspired methods.
Authors: Aheli Poddar (Institute of Engineering & Management, Kolkata), Saptarshi Sahoo (Indian Statistical Institute, Chennai), Sujata Ghosh (Indian Statistical Institute, Chennai)
Abstract: We study syllogistic reasoning in LLMs from the logical and natural language perspectives. In process, we explore fundamental reasoning capabilities of the LLMs and the direction this research is moving forward. To aid in our studies, we use 14 large language models and investigate their syllogistic reasoning capabilities in terms of symbolic inferences as well as natural language understanding. Even though this reasoning mechanism is not a uniform emergent property across LLMs, the perfect symbolic performances in certain models make us wonder whether LLMs are becoming more and more formal reasoning mechanisms, rather than making explicit the nuances of human reasoning.
Authors: Nguyen Tien Dong, Minh-Anh Nguyen, Thanh Dat Hoang, Nguyen Tuan Ngoc, Dao Xuan Quang Minh, Phan Phi Hai, Nguyen Thi Ngoc Anh, Dang Van Tu, Binh Vu
Abstract: The rapid advancement of large language models (LLMs) has enabled new possibilities for applying artificial intelligence within the legal domain. Nonetheless, the complexity, hierarchical organization, and frequent revisions of Vietnamese legislation pose considerable challenges for evaluating how well these models interpret and utilize legal knowledge. To address this gap, Vietnamese Legal Benchmark (VLegal-Bench) is introduced, the first comprehensive benchmark designed to systematically assess LLMs on Vietnamese legal tasks. Informed by Bloom's cognitive taxonomy, VLegal-Bench encompasses multiple levels of legal understanding through tasks designed to reflect practical usage scenarios. The benchmark comprises 10,450 samples generated through a rigorous annotation pipeline, where legal experts label and cross-validate each instance using our annotation system to ensure every sample is grounded in authoritative legal documents and mirrors real-world legal assistant workflows, including general legal questions and answers, retrieval-augmented generation, multi-step reasoning, and scenario-based problem solving tailored to Vietnamese law. By providing a standardized, transparent, and cognitively informed evaluation framework, VLegal-Bench establishes a solid foundation for assessing LLM performance in Vietnamese legal contexts and supports the development of more reliable, interpretable, and ethically aligned AI-assisted legal systems.
Authors: Chenkai Xu, Yijie Jin, Jiajun Li, Yi Tu, Guoping Long, Dandan Tu, Mingcong Song, Hongjie Si, Tianqi Hou, Junchi Yan, Zhijie Deng
Abstract: Diffusion Large Language Models (dLLMs) have demonstrated significant potential for high-speed inference. However, current confidence-driven decoding strategies are constrained by limited parallelism, typically achieving only 1--3 tokens per forward pass (TPF). In this work, we identify that the degree of parallelism during dLLM inference is highly sensitive to the Token Filling Order (TFO). Then, we introduce Lookahead PArallel Decoding LoPA, a training-free, plug-and-play algorithm, to identify a superior TFO and hence accelerate inference. LoPA concurrently explores distinct candidate TFOs via parallel branches, and selects the one with the highest potential for future parallelism based on branch confidence. We apply LoPA to the state-of-the-art D2F model and observe a substantial enhancement in decoding efficiency. Notably, LoPA increases the TPF of D2F-Dream to 10.1 on the GSM8K while maintaining performance superior to the Dream baseline. Furthermore, to facilitate this unprecedented degree of parallelism, we develop a specialized multi-device inference system featuring Branch Parallelism (BP), which achieves a single-sample throughput of 1073.9 tokens per second under multi-GPU deployment. The code is available at https://github.com/zhijie-group/LoPA.
Authors: Darshil Chauhan, Adityasinh Solanki, Vansh Patel, Kanav Kapoor, Ritvik Jain, Aditya Bansal, Pratik Narang, Dhruv Kumar
Abstract: Automatic Speech Recognition (ASR) holds immense potential to assist in clinical documentation and patient report generation, particularly in resource-constrained regions. However, deployment is currently hindered by a technical deadlock: a severe "Reality Gap" between laboratory performance and noisy, real-world clinical audio, coupled with strict privacy and resource constraints. We quantify this gap, showing that a robust multilingual model (IndicWav2Vec) degrades to a 40.94% WER on rural clinical data from India, rendering it unusable. To address this, we explore a zero-data-exfiltration framework enabling localized, continual adaptation via Low-Rank Adaptation (LoRA). We conduct a rigorous investigative study of continual learning strategies, characterizing the trade-offs between data-driven and parameter-driven stability. Our results demonstrate that multi-domain Experience Replay (ER) yields the primary performance gains, achieving a 17.1% relative improvement in target WER and reducing catastrophic forgetting by 55% compared to naive adaptation. Furthermore, we observed that standard Elastic Weight Consolidation (EWC) faced numerical stability challenges when applied to LoRA in noisy environments. Our experiments show that a stabilized, linearized formulation effectively controls gradient magnitudes and enables stable convergence. Finally, we verify via a domain-specific spot check that acoustic adaptation is a fundamental prerequisite for usability which cannot be bypassed by language models alone.
Authors: Qizhe Yang, Boxuan Liang, Hao Chen, Guoqiang Li
Abstract: Zero-knowledge proof (ZKP) systems have surged attention and held a fundamental role in contemporary cryptography. Zero-knowledge succinct non-interactive argument of knowledge (zk-SNARK) protocols dominate the ZKP usage, implemented through arithmetic circuit programming paradigm. However, underconstrained or overconstrained circuits may lead to bugs. The former refers to circuits that lack the necessary constraints, resulting in unexpected solutions and causing the verifier to accept a bogus witness, and the latter refers to circuits that are constrained excessively, resulting in lacking necessary solutions and causing the verifier to accept no witness. This paper introduces a novel approach for pinpointing two distinct types of bugs in ZKP circuits. The method involves encoding the arithmetic circuit constraints to polynomial equation systems and solving them over finite fields by the computer algebra system. The classification of verification results is refined, greatly enhancing the expressive power of the system. A tool, AC4, is proposed to represent the implementation of the method. Experiments show that AC4 demonstrates a increase in the solved rate, showing a 29% improvement over Picus and CIVER, and a slight improvement over halo2-analyzer, a checker for halo2 circuits. Within a solvable range, the checking time has also exhibited noticeable improvement, demonstrating a magnitude increase compared to previous efforts.
Authors: Arian Askari, Chuan Meng, Mohammad Aliannejadi, Zhaochun Ren, Evangelos Kanoulas, Suzan Verberne
Abstract: Existing generative retrieval (GR) methods rely on training-based indexing, which fine-tunes a model to memorise associations between queries and the document identifiers (docids) of relevant documents. Training-based indexing suffers from high training costs, under-utilisation of pre-trained knowledge in large language models (LLMs), and limited adaptability to dynamic document corpora. To address the issues, we propose a few-shot indexing-based GR framework (Few-Shot GR). It has a few-shot indexing process without any training, where we prompt an LLM to generate docids for all documents in a corpus, ultimately creating a docid bank for the entire corpus. During retrieval, we feed a query to the same LLM and constrain it to generate a docid within the docid bank created during indexing, and then map the generated docid back to its corresponding document. Moreover, we devise few-shot indexing with one-to-many mapping to further enhance Few-Shot GR. Experiments show that Few-Shot GR achieves superior performance to state-of-the-art GR methods requiring heavy training.
Authors: Marc Canby, Adam Davies, Chirag Rastogi, Julia Hockenmaier
Abstract: Causal probing aims to analyze foundation models by examining how intervening on their representation of various latent properties impacts their outputs. Recent works have cast doubt on the theoretical basis of several leading causal probing methods, but it has been unclear how to systematically evaluate the effectiveness of these methods in practice. To address this, we define two key causal probing desiderata: completeness (how thoroughly the representation of the target property has been transformed) and selectivity (how little non-targeted properties have been impacted). We find that there is an inherent tradeoff between the two, which we define as reliability, their harmonic mean. We introduce an empirical analysis framework to measure and evaluate these quantities, allowing us to make the first direct comparisons between different families of leading causal probing methods (e.g., linear vs. nonlinear, or concept removal vs. counterfactual interventions). We find that: (1) all methods show a clear tradeoff between completeness and selectivity; (2) more complete and reliable methods have a greater impact on LLM behavior; and (3) nonlinear interventions are almost always more reliable than linear interventions. Our project webpage is available at: https://ahdavies6.github.io/causal_probing_reliability/
URLs: https://ahdavies6.github.io/causal_probing_reliability/
Authors: Shijue Huang, Hongru Wang, Wanjun Zhong, Zhaochen Su, Jiazhan Feng, Bowen Cao, Yi R. Fung
Abstract: Modern large reasoning models demonstrate impressive problem-solving capabilities by employing sophisticated reasoning strategies. However, they often struggle to balance efficiency and effectiveness, frequently generating unnecessarily lengthy reasoning chains for simple problems. In this work, we propose AdaCtrl, a novel framework to support both difficulty-aware adaptive reasoning budget allocation and explicit user control over reasoning depth. AdaCtrl dynamically adjusts its reasoning length based on self-assessed problem difficulty, while also allowing users to manually control the budget to prioritize either efficiency or effectiveness. This is achieved through a two-stage training pipeline: an initial cold-start fine-tuning phase to instill the ability to self-aware difficulty and adjust reasoning budget, followed by a difficulty-aware reinforcement learning (RL) stage that refines the model's adaptive reasoning strategies and calibrates its difficulty assessments based on its evolving capabilities during online training. To enable intuitive user interaction, we design explicit length-triggered tags that function as a natural interface for budget control. Empirical results show that AdaCtrl adapts reasoning length based on estimated difficulty, compared to the standard training baseline that also incorporates fine-tuning and RL, it yields performance improvements and simultaneously reduces response length by 10.06% and 12.14% on the more challenging AIME2024 and AIME2025 datasets, which require elaborate reasoning, and by 62.05% and 91.04% on the MATH500 and GSM8K datasets, where more concise responses are sufficient. Furthermore, AdaCtrl enables precise user control over the reasoning budget, allowing for tailored responses to meet specific needs.
Authors: Dana Arad, Aaron Mueller, Yonatan Belinkov
Abstract: Sparse Autoencoders (SAEs) have been proposed as an unsupervised approach to learn a decomposition of a model's latent space. This enables useful applications such as steering - influencing the output of a model towards a desired concept - without requiring labeled data. Current methods identify SAE features to steer by analyzing the input tokens that activate them. However, recent work has highlighted that activations alone do not fully describe the effect of a feature on the model's output. In this work, we draw a distinction between two types of features: input features, which mainly capture patterns in the model's input, and output features, which have a human-understandable effect on the model's output. We propose input and output scores to characterize and locate these types of features, and show that high values for both scores rarely co-occur in the same features. These findings have practical implications: after filtering out features with low output scores, we obtain 2-3x improvements when steering with SAEs, making them competitive with supervised methods.
Authors: Vadim Kurochkin, Yaroslav Aksenov, Daniil Laptev, Daniil Gavrilov, Nikita Balagansky
Abstract: Sparse Autoencoders (SAEs) have demonstrated significant promise in interpreting the hidden states of language models by decomposing them into interpretable latent directions. However, training and interpreting SAEs at scale remains challenging, especially when large dictionary sizes are used. While decoders can leverage sparse-aware kernels for efficiency, encoders still require computationally intensive linear operations with large output dimensions. To address this, we propose KronSAE, a novel architecture that factorizes the latent representation via Kronecker product decomposition, drastically reducing memory and computational overhead. Furthermore, we introduce mAND, a differentiable activation function approximating the binary AND operation, which improves interpretability and performance in our factorized framework.
Authors: Kieu Dang, Phung Lai, NhatHai Phan, Yelong Shen, Ruoming Jin, Abdallah Khreishah, My T. Thai
Abstract: Large Language Models (LLMs) have transformed natural language processing, demonstrating impressive capabilities across diverse tasks. However, deploying these models introduces critical risks related to intellectual property violations and potential misuse, particularly as adversaries can imitate these models to steal services or generate misleading outputs. We specifically focus on model stealing attacks, as they are highly relevant to proprietary LLMs and pose a serious threat to their security, revenue, and ethical deployment. While various watermarking techniques have emerged to mitigate these risks, it remains unclear how far the community and industry have progressed in developing and deploying watermarks in LLMs. To bridge this gap, we aim to develop a comprehensive systematization for watermarks in LLMs by 1) presenting a detailed taxonomy for watermarks in LLMs, 2) proposing a novel intellectual property classifier to explore the effectiveness and impacts of watermarks on LLMs under both attack and attack-free environments, 3) analyzing the limitations of existing watermarks in LLMs, and 4) discussing practical challenges and potential future directions for watermarks in LLMs. Through extensive experiments, we show that despite promising research outcomes and significant attention from leading companies and community to deploy watermarks, these techniques have yet to reach their full potential in real-world applications due to their unfavorable impacts on model utility of LLMs and downstream tasks. Our findings provide an insightful understanding of watermarks in LLMs, highlighting the need for practical watermarks solutions tailored to LLM deployment.
Authors: Hongyuan Dong, Dingkang Yang, Xiao Liang, Chao Feng, Jiao Ran
Abstract: Learning rate is widely regarded as crucial for effective foundation model pretraining. Recent research explores and demonstrates the transferability of learning rate configurations across varying model and dataset sizes, etc. Nevertheless, these approaches are constrained to specific training scenarios and typically necessitate extensive hyperparameter tuning on proxy models. In this work, we propose \textbf{AdaLRS}, a plug-in-and-play adaptive learning rate search algorithm that conducts online optimal learning rate search via optimizing loss descent velocities. We provide theoretical and experimental analyzes to show that foundation model pretraining loss and its descent velocity are both convex and share the same optimal learning rate. Relying solely on training loss dynamics, AdaLRS involves few extra computations to guide the search process, and its convergence is guaranteed via theoretical analysis. Experiments on both LLM and VLM pretraining show that AdaLRS adjusts suboptimal learning rates to the neighborhood of optimum with marked efficiency and effectiveness, with model performance improved accordingly. We also show the robust generalizability of AdaLRS across varying training scenarios, such as different model sizes, training paradigms, base learning rate scheduler choices, and hyperparameter settings.
Authors: Bingning Huang, Tu Nguyen, Matthieu Zimmer
Abstract: Recent advances in reasoning with large language models (LLMs) have shown the effectiveness of Monte Carlo Tree Search (MCTS) for generating high quality intermediate trajectories, particularly in math and symbolic domains. Inspired by this, we explore how MCTS derived trajectories, traditionally used for training value or reward models, can be repurposed to improve policy optimization in verifier guided reinforcement learning (RL). Specifically, we focus on Group Relative Policy Optimization (GRPO), a recent algorithm that enables consistent policy learning from group relative judgments. We reframe GRPO into a staged training paradigm, leveraging a teacher's MCTS rollouts to construct a tree structured curriculum of prefixes. This introduces the novel challenge of computing advantages for training samples that originate from different prefixes, each with a distinct expected return. To address this, we propose Staged Advantage Estimation (SAE), a framework for computing low variance, prefix aware advantages by projecting rewards onto a constraint set that respects the tree's hierarchy. Our empirical results on mathematical reasoning tasks show that SAE improves final accuracy over standard GRPO. This outcome is grounded in our theoretical analysis, which confirms that SAE reduces gradient variance, a principled path to improved sample efficiency. We demonstrate this through practical SAE implementations, comparing efficient heuristics against a formal quadratic program.
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: Kaijian Zou, Aaron Xiong, Yunxiang Zhang, Frederick Zhang, Yueqi Ren, Jirong Yang, Ayoung Lee, Shitanshu Bhushan, Lu Wang
Abstract: Competitive programming problems increasingly serve as valuable benchmarks to evaluate the coding capabilities of large language models (LLMs) due to their complexity and ease of verification. Yet, current coding benchmarks face limitations such as lack of exceptionally challenging problems, insufficient test case coverage, reliance on online platform APIs that limit accessibility. To address these issues, we introduce LiveOIBench, a comprehensive benchmark featuring 403 expert-curated Olympiad-level competitive programming problems, each with an average of 60 expert-designed test cases. The problems are sourced directly from 72 official contests of 14 Informatics Olympiads in different regions conducted between 2023 and 2025. LiveOIBench distinguishes itself through four key features: (1) meticulously curated high-quality tasks with detailed subtask rubrics and extensive private test cases; (2) direct integration of elite contestant performance data to enable informative comparison against top-performing humans; (3) planned continuous, contamination-free updates from newly released Olympiad problems; and (4) a self-contained evaluation system facilitating offline and easy-to-reproduce assessments. Benchmarking 34 popular general-purpose and reasoning LLMs, we find that GPT-5 achieves a notable 81.76th percentile, a strong result that nonetheless falls short of top human contestants, who usually place above 90th. In contrast, among open-weight reasoning models, GPT-OSS-120B achieves only a 60th percentile, underscoring significant capability disparities from frontier closed models. Detailed analyses indicate that robust reasoning models prioritize precise problem analysis over excessive exploration, suggesting future models should emphasize structured analysis and minimize unnecessary exploration. All data, code, and leaderboard results are publicly available on our website.
Authors: Qingchuan Yang, Simon Mahns, Sida Li, Anri Gu, Jibang Wu, Haifeng Xu
Abstract: Forecasting is not only a fundamental intellectual pursuit but also is of significant importance to societal systems such as finance and economics. With the rapid advances of large language models (LLMs) trained on Internet-scale data, it raises the promise of employing LLMs to forecast real-world future events, an emerging paradigm we call "LLM-as-a-Prophet". This paper systematically investigates such predictive intelligence of LLMs. To this end, we build Prophet Arena, a general evaluation benchmark that continuously collects live forecasting tasks and decomposes each task into distinct pipeline stages, in order to support our controlled and large-scale experimentation. Our comprehensive evaluation reveals that many LLMs already exhibit impressive forecasting capabilities, reflected in, e.g., their small calibration errors, consistent prediction confidence and promising market returns. However, we also uncover key bottlenecks towards achieving superior predictive intelligence via LLM-as-a-Prophet, such as LLMs' inaccurate event recalls, misunderstanding of data sources and slower information aggregation compared to markets when resolution nears.
Authors: Quoc Anh Nguyen, Bernard Cheng, Kelvin Soh
Abstract: Automatic Lyrics Transcription (ALT) for Vietnamese music presents unique challenges due to its tonal complexity and dialectal variations, but remains largely unexplored due to the lack of a dedicated dataset. Therefore, we curated the first large-scale Vietnamese ALT dataset (VietLyrics), comprising 647 hours of songs with line-level aligned lyrics and metadata to address these issues. Our evaluation of current ASRbased approaches reveal significant limitations, including frequent transcription errors and hallucinations in non-vocal segments. To improve performance, we fine-tuned Whisper models on the VietLyrics dataset, achieving superior results compared to existing multilingual ALT systems, including LyricWhiz. We publicly release VietLyrics and our models, aiming to advance Vietnamese music computing research while demonstrating the potential of this approach for ALT in low-resource language and music.
Authors: Jiali Cheng, Anjishnu Kumar, Roshan Lal, Rishi Rajasekaran, Hani Ramezani, Omar Zia Khan, Oleg Rokhlenko, Sunny Chiu-Webster, Gang Hua, Hadi Amiri
Abstract: Large Language Model (LLM) web agents often struggle with long-horizon web navigation and web task completion in new websites, producing inefficient action sequences unless fine-tuned on environment-specific data. We show that experience-driven memory, combined with look-ahead action simulation, is sufficient for LLM agents to adapt to unseen web environments by remembering past failures and predicting the consequences of future actions. We introduce WebATLAS (Actor-Critic Task-completion with Look-ahead Action Simulation), a memory-augmented LLM web agent that learns a lightweight internal model of the environment from interaction experience and performs hypothetical action rollouts before acting in the real world. WebATLAS builds a persistent cognitive map via curiosity-driven exploration, stores interaction outcomes as experience-based memory, and evaluates candidate actions in cognitive space using a planner--simulator--critic loop. This enables the agent to reuse past experience, avoid previously unsuccessful behaviors, and generate more efficient plans. We evaluate WebATLAS on the WebArena-Lite benchmark for autonomous web navigation and demonstrate a success rate of 63%, outperforming the previous state-of-the-art at 53.9%. Unlike previous systems, our modular architecture requires no website-specific LLM fine-tuning. Ablation studies confirm that experience-driven memory, look-ahead action simulation, and hierarchical replanning play complementary roles in enabling robust, training-free web agents.
Authors: Abhishek Panigrahi, Bingbin Liu, Sadhika Malladi, Sham Kakade, Surbhi Goel
Abstract: Knowledge distillation is an efficient strategy to use data generated by large "teacher" language models to train smaller capable "student" models, but selecting the optimal teacher for a specific student-task combination requires expensive trial-and-error. We propose a lightweight score called GRACE to quantify how effective a teacher will be for post-training a student model. GRACE measures distributional properties of the student's gradients without access to a verifier, teacher logits, teacher internals, or test data. From an information-theoretic perspective, GRACE connects to leave-one-out stability of gradient-based algorithms, which controls the generalization performance of the distilled students. On GSM8K and MATH, GRACE correlates strongly (up to 86% Spearman correlation) with the performance of the distilled LLaMA and OLMo students. In particular, training a student using the GRACE-selected teacher can improve the performance by up to 7.4% over naively using the best-performing teacher. Further, GRACE can provide guidance on crucial design choices in distillation, including (1) the best temperature to use when generating from the teacher, (2) the best teacher to use given a size constraint, and (3) the best teacher to use within a specific model family. Altogether, our findings demonstrate that GRACE can efficiently and effectively identify a strongly compatible teacher for a given student and provide fine-grained guidance on how to perform distillation.
Authors: Aarush Sinha, Pavan Kumar S, Roshan Balaji, Nirav Pravinbhai Bhatt
Abstract: Hard negatives are essential for training effective retrieval models. Hard-negative mining typically relies on ranking documents using cross-encoders or static embedding models based on similarity metrics such as cosine distance. Hard negative mining becomes challenging for biomedical and scientific domains due to the difficulty in distinguishing between source and hard negative documents. However, referenced documents naturally share contextual relevance with the source document but are not duplicates, making them well-suited as hard negatives. In this work, we propose BiCA: Biomedical Dense Retrieval with Citation-Aware Hard Negatives, an approach for hard-negative mining by utilizing citation links in 20,000 PubMed articles for improving a domain-specific small dense retriever. We fine-tune the GTE_small and GTE_Base models using these citation-informed negatives and observe consistent improvements in zero-shot dense retrieval using nDCG@10 for both in-domain and out-of-domain tasks on BEIR and outperform baselines on long-tailed topics in LoTTE using Success@5. Our findings highlight the potential of leveraging document link structure to generate highly informative negatives, enabling state-of-the-art performance with minimal fine-tuning and demonstrating a path towards highly data-efficient domain adaptation.
Authors: Patrick Amadeus Irawan, Ikhlasul Akmal Hanif, Muhammad Dehan Al Kautsar, Genta Indra Winata, Fajri Koto, Alham Fikri Aji
Abstract: Although the cultural dimension has been one of the key aspects in evaluating Vision-Language Models (VLMs), their ability to remain stable across diverse cultural inputs remains largely untested, despite being crucial to support diversity and multicultural societies. Existing evaluations often rely on benchmarks featuring only a singular cultural concept per image, overlooking scenarios where multiple, potentially unrelated cultural cues coexist. To address this gap, we introduce ConfusedTourist, a novel cultural adversarial robustness suite designed to assess VLMs' stability against perturbed geographical cues. Our experiments reveal a critical vulnerability, where accuracy drops heavily under simple image-stacking perturbations and even worsens with its image-generation-based variant. Interpretability analyses further show that these failures stem from systematic attention shifts toward distracting cues, diverting the model from its intended focus. These findings highlight a critical challenge: visual cultural concept mixing can substantially impair even state-of-the-art VLMs, underscoring the urgent need for more culturally robust multimodal understanding.
Authors: Rohin Manvi, Joey Hong, Tim Seyde, Maxime Labonne, Mathias Lechner, Sergey Levine
Abstract: Large language models excel at reasoning but lack key aspects of introspection, including anticipating their own success and the computation required to achieve it. Humans use real-time introspection to decide how much effort to invest, when to make multiple attempts, when to stop, and when to signal success or failure. Without this, LLMs struggle to make intelligent meta-cognition decisions. Test-time scaling methods like Best-of-N drive up cost and latency by using a fixed budget of samples regardless of the marginal benefit of each one at any point in generation, and the absence of confidence signals can mislead people, prevent appropriate escalation to better tools, and undermine trustworthiness. Learned verifiers or reward models can provide confidence estimates, but do not enable adaptive inference and add substantial cost by requiring extra models or forward passes. We present ZIP-RC, an adaptive inference method that equips models with zero-overhead inference-time predictions of reward and cost. At every token, ZIP-RC reuses reserved or unused logits in the same forward pass as next-token prediction to output a joint distribution over final reward and remaining length -- no extra models, architecture change, or inference overhead. This full joint distribution is used to compute a sampling utility which is the linear combination of the expected maximum reward, total compute, and latency of set of samples if generated to completion. During inference, we maximize this utility with meta-actions that determine which prefix of tokens to continue or initiate sampling from. On mixed-difficulty mathematical benchmarks, ZIP-RC improves accuracy by up to 12% over majority voting at equal or lower average cost, and traces smooth Pareto frontiers between quality, compute, and latency. By providing real-time reward-cost introspection, ZIP-RC enables adaptive, efficient reasoning.
Authors: Donghang Duan, Xu Zheng, Yuefeng He, Chong Mu, Leyi Cai, Lizong Zhang
Abstract: Current LLM-based text anonymization frameworks usually rely on remote API services from powerful LLMs, which creates an inherent privacy paradox: users must disclose data to untrusted third parties for guaranteed privacy preservation. Moreover, directly migrating current solutions to local small-scale models (LSMs) offers a suboptimal solution with severe utility collapse. Our work argues that this failure stems not merely from the capability deficits of LSMs, but significantly from the inherent irrationality of the greedy adversarial strategies employed by current state-of-the-art (SOTA) methods. To address this, we propose Rational Localized Adversarial Anonymization (RLAA), a fully localized and training-free framework featuring an Attacker-Arbitrator-Anonymizer architecture. We model the anonymization process as a trade-off between Marginal Privacy Gain (MPG) and Marginal Utility Cost (MUC), and demonstrate that greedy strategies tend to drift into an irrational state. Instead, RLAA introduces an arbitrator that acts as a rationality gatekeeper, validating the attacker's inference to filter out feedback providing negligible privacy benefits. This mechanism promotes a rational early-stopping criterion, and structurally prevents utility collapse. Extensive experiments on different benchmarks demonstrate that RLAA achieves a superior privacy-utility trade-off compared to strong baselines.
Authors: Nikhil Verma, Joonas Linnosmaa, Leonardo Espinosa-Leal, Napat Vajragupta
Abstract: The paper presents the formulation, implementation, and evaluation of the ArcGD optimiser. The evaluation is conducted initially on a non-convex benchmark function and subsequently on a real-world ML dataset. The initial comparative study using the Adam optimiser is conducted on a stochastic variant of the highly non-convex and notoriously challenging Rosenbrock function, renowned for its narrow, curved valley, across dimensions ranging from 2D to 1000D and an extreme case of 50,000D. Two configurations were evaluated to eliminate learning-rate bias: (i) both using ArcGD's effective learning rate and (ii) both using Adam's default learning rate. ArcGD consistently outperformed Adam under the first setting and, although slower under the second, achieved superior final solutions in most cases. In the second evaluation, ArcGD is evaluated against state-of-the-art optimizers (Adam, AdamW, Lion, SGD) on the CIFAR-10 image classification dataset across 8 diverse MLP architectures ranging from 1 to 5 hidden layers. ArcGD achieved the highest average test accuracy (50.7%) at 20,000 iterations, outperforming AdamW (46.6%), Adam (46.8%), SGD (49.6%), and Lion (43.4%), winning or tying on 6 of 8 architectures. Notably, while Adam and AdamW showed strong early convergence at 5,000 iterations, but regressed with extended training, whereas ArcGD continued improving, demonstrating generalization and resistance to overfitting without requiring early stopping tuning. Strong performance on geometric stress tests and standard deep-learning benchmarks indicates broad applicability, highlighting the need for further exploration. Moreover, it is also shown that a limiting variant of ArcGD can be interpreted as a sign-based momentum-like update, highlighting conceptual connections between the inherent mechanisms of ArcGD and the Lion optimiser.
Authors: Jiin Park, Hyuna Jeon, Yoonseo Lee, Jisu Hong, Misuk Kim
Abstract: Ontology-based knowledge graph (KG) construction is a core technology that enables multidimensional understanding and advanced reasoning over domain knowledge. Industrial standards, in particular, contain extensive technical information and complex rules presented in highly structured formats that combine tables, scopes of application, constraints, exceptions, and numerical calculations, making KG construction especially challenging. In this study, we propose a method that organizes such documents into a hierarchical semantic structure, decomposes sentences and tables into atomic propositions derived from conditional and numerical rules, and integrates them into an ontology-knowledge graph through LLM-based triple extraction. Our approach captures both the hierarchical and logical structures of documents, effectively representing domain-specific semantics that conventional methods fail to reflect. To verify its effectiveness, we constructed rule, table, and multi-hop QA datasets, as well as a toxic clause detection dataset, from industrial standards, and implemented an ontology-aware KG-RAG framework for comparative evaluation. Experimental results show that our method achieves significant performance improvements across all QA types compared to existing KG-RAG approaches. This study demonstrates that reliable and scalable knowledge representation is feasible even for industrial documents with intertwined conditions, constraints, and scopes, contributing to future domain-specific RAG development and intelligent document management.
Authors: Lucas Monteiro Paes, Nivedha Sivakumar, Yinong Oliver Wang, Masha Fedzechkina Donaldson, Barry-John Theobald, Luca Zappella, Nicholas Apostoloff
Abstract: Generative models are often deployed to make decisions on behalf of users, such as vision-language models (VLMs) identifying which person in a room is a doctor to help visually impaired individuals. Yet, VLM decisions are influenced by the perceived demographic attributes of people in the input, which can lead to biased outcomes like failing to identify women as doctors. Moreover, when reducing bias leads to performance loss, users may have varying needs for balancing bias mitigation with overall model capabilities, highlighting the demand for methods that enable controllable bias reduction during inference. Activation steering is a popular approach for inference-time controllability that has shown potential in inducing safer behavior in large language models (LLMs). However, we observe that current steering methods struggle to correct biases, where equiprobable outcomes across demographic groups are required. To address this, we propose Direct Steering Optimization (DSO) which uses reinforcement learning to find linear transformations for steering activations, tailored to mitigate bias while maintaining control over model performance. We demonstrate that DSO achieves state-of-the-art trade-off between fairness and capabilities on both VLMs and LLMs, while offering practitioners inference-time control over the trade-off. Overall, our work highlights the benefit of designing steering strategies that are directly optimized to control model behavior, providing more effective bias intervention than methods that rely on pre-defined heuristics for controllability.
Authors: Pengcheng Jiang, Jiacheng Lin, Zhiyi Shi, Zifeng Wang, Luxi He, Yichen Wu, Ming Zhong, Peiyang Song, Qizheng Zhang, Heng Wang, Xueqiang Xu, Hanwen Xu, Pengrui Han, Dylan Zhang, Jiashuo Sun, Chaoqi Yang, Kun Qian, Tian Wang, Changran Hu, Manling Li, Quanzheng Li, Hao Peng, Sheng Wang, Jingbo Shang, Chao Zhang, Jiaxuan You, Liyuan Liu, Pan Lu, Yu Zhang, Heng Ji, Yejin Choi, Dawn Song, Jimeng Sun, Jiawei Han
Abstract: Cutting-edge agentic AI systems are built on foundation models that can be adapted to plan, reason, and interact with external tools to perform increasingly complex and specialized tasks. As these systems grow in capability and scope, adaptation becomes a central mechanism for improving performance, reliability, and generalization. In this paper, we unify the rapidly expanding research landscape into a systematic framework that spans both agent adaptations and tool adaptations. We further decompose these into tool-execution-signaled and agent-output-signaled forms of agent adaptation, as well as agent-agnostic and agent-supervised forms of tool adaptation. We demonstrate that this framework helps clarify the design space of adaptation strategies in agentic AI, makes their trade-offs explicit, and provides practical guidance for selecting or switching among strategies during system design. We then review the representative approaches in each category, analyze their strengths and limitations, and highlight key open challenges and future opportunities. Overall, this paper aims to offer a conceptual foundation and practical roadmap for researchers and practitioners seeking to build more capable, efficient, and reliable agentic AI systems.
Authors: Peter Chen, Xiaopeng Li, Ziniu Li, Wotao Yin, Xi Chen, Tianyi Lin
Abstract: This paper examines the exploration-exploitation trade-off in reinforcement learning with verifiable rewards (RLVR), a framework for improving the reasoning of Large Language Models (LLMs). Recent studies suggest that RLVR can elicit strong mathematical reasoning in LLMs through two seemingly paradoxical mechanisms: spurious rewards, which suppress exploitation by rewarding outcomes unrelated to the ground truth, and entropy minimization, which suppresses exploration by pushing the model toward more confident and deterministic outputs, highlighting a puzzling dynamic: both discouraging exploitation and discouraging exploration improve reasoning performance, yet the underlying principles that reconcile these effects remain poorly understood. We focus on two fundamental questions: (i) how policy entropy relates to performance, and (ii) whether spurious rewards yield gains, potentially through the interplay of clipping bias and model contamination. Our results show that clipping bias under spurious rewards reduces policy entropy, leading to more confident and deterministic outputs, while entropy minimization alone is insufficient for improvement. We further propose a reward-misalignment model explaining why spurious rewards can enhance performance beyond contaminated settings. Our findings clarify the mechanisms behind spurious-reward benefits and provide principles for more effective RLVR training.