Peer Revenue Fully funded

Papers

Fresh from arXiv, typeset properly. Highlighted words trade on the market.

Computer Science Computation and Language

cs.CL Watch Computation and Language

Saturday 27 December

Computation and Language · Cross-listed from cs.LG

AFA-LoRA: Enabling Non-Linear Adaptations in LoRA with Activation Function Annealing

Jiacheng Li, Jianchao Tan, Zhidong Yang et al. (+4 more)

Low-Rank Adaptation (LoRA) is a widely adopted parameter-efficient fine-tuning (PEFT) method. However, its linear adaptation process limits its expressive power. This means there is a gap between the expressive power of linear training and non-linear training. To bridge this gap, we propose AFA-LoRA, a novel training strategy that brings non-linear expressivity to LoRA while maintaining its seamless mergeability. Our key innovation is an annealed activation function that transitions from a non-linear to a linear transformation during training, allowing the adapter to initially adopt stronger representational capabilities before converging to a mergeable linear form. We implement our method on supervised fine-tuning, reinforcement learning, and speculative decoding. The results show that AFA-LoRA reduces the performance gap between LoRA and full-parameter training. This work enables a more powerful and practical paradigm of parameter-efficient adaptation.
Read the full abstract

All authors Jiacheng Li, Jianchao Tan, Zhidong Yang, Feiye Huo, Yerui Sun, Yuchen Xie, Xunliang Cai

Computation and Language

Accounting Reasoning in Large Language Models: Concepts, Evaluation, and Empirical Analysis

Jie Zhou, Xin Chen, Jie Zhang et al. (+1 more)

Large language models (LLMs) are increasingly reshaping learning paradigms, cognitive processes, and research methodologies across diverse domains. As their adoption expands, effectively integrating LLMs into professional fields and clarifying their role in domain-specific applications has become a key challenge for enterprise digital transformation and broader societal development. In the accounting domain, successful integration requires a systematic understanding of LLMs' domain-specific reasoning capabilities. In this study, we introduce the concept of accounting reasoning and propose a set of evaluation criteria grounded in an analysis of the training data characteristics of representative GLM-series models. These criteria establish a foundation for studying accounting-oriented reasoning paradigms and provide benchmarks for assessing and improving model performance. Building on this framework, we evaluate several representative LLMs, including GLM-6B, GLM-130B, GLM-4, and GPT-4, across a range of accounting reasoning tasks. Our experimental results show that prompt engineering strategies can yield varying degrees of performance improvement across models, with GPT-4 demonstrating the strongest overall accounting reasoning capability. Nevertheless, the results indicate that current LLMs remain insufficient for real-world accounting applications. In particular, further optimization is required for deployment in enterprise-level accounting scenarios to fully realize the potential value of LLMs in this domain.
Read the full abstract

All authors Jie Zhou, Xin Chen, Jie Zhang, Zhe Li

Computation and Language

HiFi-RAG: Hierarchical Content Filtering and Two-Pass Generation for Open-Domain RAG

Cattalyya Nuengsigkapian

Retrieval-Augmented Generation (RAG) in open-domain settings faces significant challenges regarding irrelevant information in retrieved documents and the alignment of generated answers with user intent. We present HiFi-RAG (Hierarchical Filtering RAG), the winning closed-source system in the Text-to-Text static evaluation of the MMU-RAGent NeurIPS 2025 Competition. Our approach moves beyond standard embedding-based retrieval via a multi-stage pipeline. We leverage the speed and cost-efficiency of Gemini 2.5 Flash (4-6x cheaper than Pro) for query formulation, hierarchical content filtering, and citation attribution, while reserving the reasoning capabilities of Gemini 2.5 Pro for final answer generation. On the MMU-RAGent validation set, our system outperformed the baseline, improving ROUGE-L to 0.274 (+19.6%) and DeBERTaScore to 0.677 (+6.2%). On Test2025, our custom dataset evaluating questions that require post-cutoff knowledge (post January 2025), HiFi-RAG outperforms the parametric baseline by 57.4% in ROUGE-L and 14.9% in DeBERTaScore.
Read the full abstract

Computation and Language · Cross-listed from cs.AI

Monadic Context Engineering

Yifan Zhang, Yang Yuan, Mengdi Wang et al. (+1 more)

The proliferation of Large Language Models (LLMs) has catalyzed a shift towards autonomous agents capable of complex reasoning and tool use. However, current agent architectures are frequently constructed using imperative, ad hoc patterns. This results in brittle systems plagued by difficulties in state management, error handling, and concurrency. This paper introduces Monadic Context Engineering (MCE), a novel architectural paradigm leveraging the algebraic structures of Functors, Applicative Functors, and Monads to provide a formal foundation for agent design. MCE treats agent workflows as computational contexts where cross-cutting concerns, such as state propagation, short-circuiting error handling, and asynchronous execution, are managed intrinsically by the algebraic properties of the abstraction. We demonstrate how Monads enable robust sequential composition, how Applicatives provide a principled structure for parallel execution, and crucially, how Monad Transformers allow for the systematic composition of these capabilities. This layered approach enables developers to construct complex, resilient, and efficient AI agents from simple, independently verifiable components. We further extend this framework to describe Meta-Agents, which leverage MCE for generative orchestration, dynamically creating and managing sub-agent workflows through metaprogramming.
Read the full abstract

All authors Yifan Zhang, Yang Yuan, Mengdi Wang, Andrew Chi-Chih Yao

Computation and Language

Hallucination Detection and Evaluation of Large Language Model

Chenggong Zhang, Haopeng Wang and Hexi Meng

Hallucinations in Large Language Models (LLMs) pose a significant challenge, generating misleading or unverifiable content that undermines trust and reliability. Existing evaluation methods, such as KnowHalu, employ multi-stage verification but suffer from high computational costs. To address this, we integrate the Hughes Hallucination Evaluation Model (HHEM), a lightweight classification-based framework that operates independently of LLM-based judgments, significantly improving efficiency while maintaining high detection accuracy. We conduct a comparative analysis of hallucination detection methods across various LLMs, evaluating True Positive Rate (TPR), True Negative Rate (TNR), and Accuracy on question-answering (QA) and summarization tasks. Our results show that HHEM reduces evaluation time from 8 hours to 10 minutes, while HHEM with non-fabrication checking achieves the highest accuracy \(82.2\%\) and TPR \(78.9\%\). However, HHEM struggles with localized hallucinations in summarization tasks. To address this, we introduce segment-based retrieval, improving detection by verifying smaller text components. Additionally, our cumulative distribution function (CDF) analysis indicates that larger models (7B-9B parameters) generally exhibit fewer hallucinations, while intermediate-sized models show higher instability. These findings highlight the need for structured evaluation frameworks that balance computational efficiency with robust factual validation, enhancing the reliability of LLM-generated content.
Read the full abstract

Friday 26 December

Computation and Language

Measuring Stability Beyond Accuracy in Small Open-Source Medical Large Language Models for Pediatric Endocrinology

Vanessa D'Amario, Randy Daniel, Alessandro Zanetti et al. (+3 more)

Small open-source medical large language models (LLMs) offer promising opportunities for low-resource deployment and broader accessibility. However, their evaluation is often limited to accuracy on medical multiple choice question (MCQ) benchmarks, and lacks evaluation of consistency, robustness, or reasoning behavior. We use MCQ coupled to human evaluation and clinical review to assess six small open-source medical LLMs (HuatuoGPT-o1 (Chen 2024), Diabetica-7B, Diabetica-o1 (Wei 2024), Meditron3-8B (Sallinen2025), MedFound-7B (Liu 2025), and ClinicaGPT-base-zh (Wang 2023)) in pediatric endocrinology. In deterministic settings, we examine the effect of prompt variation on models' output and self-assessment bias. In stochastic settings, we evaluate output variability and investigate the relationship between consistency and correctness. HuatuoGPT-o1-8B achieved the highest performance. The results show that high consistency across the model response is not an indicator of correctness, although HuatuoGPT-o1-8B showed the highest consistency rate. When tasked with selecting correct reasoning, both HuatuoGPT-o1-8B and Diabetica-o1 exhibit self-assessment bias and dependency on the order of the candidate explanations. Expert review of incorrect reasoning rationales identified a mix of clinically acceptable responses and clinical oversight. We further show that system-level perturbations, such as differences in CUDA builds, can yield statistically significant shifts in model output despite stable accuracy. This work demonstrates that small, semantically negligible prompt perturbations lead to divergent outputs, raising concerns about reproducibility of LLM-based evaluations and highlights the output variability under different stochastic regimes, emphasizing the need of a broader diagnostic framework to understand potential pitfalls in real-world clinical decision support scenarios.
Read the full abstract

All authors Vanessa D'Amario, Randy Daniel, Alessandro Zanetti, Dhruv Edamadaka, Nitya Alaparthy, Joshua Tarkoff

Computation and Language

Safety boundary maintenance in consumer AI systems responding to pediatric health queries: a cross-platform benchmark evaluation under naturalistic and adversarially pressured conditions

Vahideh Zolfaghari, Leila Mashhadi, Mitra Ahadi et al. (+2 more)

Consumer artificial intelligence chatbots are now accessed by hundreds of millions of users seeking health information, yet systematic evaluation of their safety boundary maintenance under real-world caregiver pressure remains scarce. We evaluated PediatricSafetyBench-v2, a benchmark of 600 pediatrics health queries comprising 300 authentic caregiver queries sourced from the HealthCareMagic-100k-en physician consultation corpus and 300 matched adversarial variants incorporating six operationalized caregiver pressure patterns, across four consumer AI systems (GPT-4o-mini, Gemini-2.0-Flash, Claude-3.5-Haiku, and Llama-3.1-8B). Safety boundary maintenance was assessed using a validated five-component Safety Composite Score (maximum 15 points; safety-appropriate threshold of 10 or above), validated against independent human raters prior to full-corpus application (mean weighted kappa 0.76; Pearson r = 0.88). The overall safety-appropriate rate was 95.5%. Safety-oriented system prompt deployment improved safety-appropriate rates by 5.9 percentage points across all four models. Counter-intuitively, adversarial caregiver pressure was associated with higher rather than lower Safety Composite Score values for all four models across all ten topic categories and severity levels. False expertise claims were the most vulnerability-inducing pressure pattern, whereas emotional escalation was associated with the highest scores. Consumer AI systems maintain safety boundaries in the large majority of pediatrics health interactions. PediatricSafetyBench-v2 is publicly released for longitudinal safety monitoring.
Read the full abstract

All authors Vahideh Zolfaghari, Leila Mashhadi, Mitra Ahadi, Farzaneh Sedaghatkar, MohammadReza Kargozari

Computation and Language

Uncertainty-Aware Dynamic Knowledge Graphs for Reliable Question Answering

Yu Takahashi, Shun Takeuchi, Kexuan Xin et al. (+6 more)

Question answering (QA) systems are increasingly deployed across domains. However, their reliability is undermined when retrieved evidence is incomplete, noisy, or uncertain. Existing knowledge graph (KG) based QA frameworks typically represent facts as static and deterministic, failing to capture the evolving nature of information and the uncertainty inherent in reasoning. We present a demonstration of uncertainty-aware dynamic KGs, a framework that combines (i) dynamic construction of evolving KGs, (ii) confidence scoring and uncertainty-aware retrieval, and (iii) an interactive interface for reliable and interpretable QA. Our system highlights how uncertainty modeling can make QA more robust and transparent by enabling users to explore dynamic graphs, inspect confidence-annotated triples, and compare baseline versus confidence-aware answers. The target users of this demo are clinical data scientists and clinicians, and we instantiate the framework in healthcare: constructing personalized KGs from electronic health records, visualizing uncertainty across patient visits, and evaluating its impact on a mortality prediction task. This use case demonstrates the broader promise of uncertainty-aware dynamic KGs for enhancing QA reliability in high-stakes applications.
Read the full abstract

All authors Yu Takahashi, Shun Takeuchi, Kexuan Xin, Guillaume Pelat, Yoshiaki Ikai, Junya Saito, Jonathan Vitale, Shlomo Berkovsky, Amin Beheshti

Computation and Language

Bounded Hyperbolic Tangent: A Stable and Efficient Alternative to Pre-Layer Normalization in Large Language Models

Hoyoon Byun, Youngjun Choi, Taero Kim et al. (+2 more)

Pre-Layer Normalization (Pre-LN) is the de facto choice for large language models (LLMs) and is crucial for stable pretraining and effective transfer learning. However, Pre-LN incurs repeated statistical-computation overhead and remains vulnerable to the curse of depth, where hidden-state magnitudes and variances grow as the number of layers increases, destabilizing training. Efficiency-oriented normalization-free methods such as Dynamic Tanh (DyT) improve throughput but remain fragile at depth. To jointly address stability and efficiency, we propose Bounded Hyperbolic Tanh (BHyT), a drop-in replacement for Pre-LN. BHyT combines a tanh nonlinearity with explicit, data-driven input bounding to keep activations within a non-saturating range. It prevents depth-wise growth in activation magnitude and variance and provides a theoretical stability guarantee. For efficiency, BHyT computes exact statistics once per block and replaces a second normalization with a lightweight variance approximation. Empirically, BHyT demonstrates improved stability and efficiency during pretraining, achieving an average of 1.6% faster training and an average of 1.77% higher token generation throughput compared to RMSNorm, while maintaining strong pretraining-only and post-SFT performance across language understanding and reasoning benchmarks\footnote{Code is available at: https://github.com/MLAI-Yonsei/BHyT}.
Read the full abstract

All authors Hoyoon Byun, Youngjun Choi, Taero Kim, Sungrae Park, Kyungwoo Song

Computation and Language

StatLLaMA: Multi-Stage training for domain-optimized statistical large language models

Jing-Yi Zeng and Guan-Hua Huang

This study investigates how to efficiently build a domain-specialized large language model (LLM) for statistics using the lightweight LLaMA-3.2-3B family as the foundation model (FM). We systematically compare three multi-stage training pipelines–starting from a base FM with no instruction-following capability, a base FM augmented with post-hoc instruction tuning, and an instruction-tuned FM with strong general reasoning abilities–across continual pretraining, supervised fine-tuning (SFT), reinforcement learning from human feedback (RLHF) preference alignment, and downstream task fine-tuning (DTFT). Results show that pipelines beginning with a base FM fail to develop meaningful statistical reasoning, even after extensive instruction tuning, SFT, or RLHF alignment. In contrast, starting from LLaMA-3.2-3B-Instruct enables effective domain specialization. A comprehensive evaluation of SFT variants reveals clear trade-offs between domain expertise and general reasoning ability. We further demonstrate that direct preference optimization provides stable and effective RLHF preference alignment. Finally, we show that DTFT must be performed with extremely low intensity to avoid catastrophic forgetting in highly optimized models. The final model, StatLLaMA, achieves strong and balanced performance on benchmarks of mathematical reasoning, common-sense reasoning, and statistical expertise, offering a practical blueprint for developing resource-efficient statistical LLMs. The code is available at https://github.com/HuangDLab/StatLLaMA.
Read the full abstract

Computation and Language

NewsScope: Schema-Grounded Cross-Domain News Claim Extraction with Open Models

Nidhi Pandya

Automated news verification requires structured claim extraction, but existing approaches either lack schema compliance or generalize poorly across domains. This paper presents NewsScope, a cross-domain dataset, benchmark, and fine-tuned model for schema-grounded news claim extraction. The dataset contains 455 articles across politics, health, science/environment, and business, consisting of 395 in-domain articles and 60 out-of-source articles for generalization testing. LLaMA 3.1 8B was fine-tuned using LoRA on 315 training examples and evaluated on held-out in-domain (80 articles) and out-of-source (60 articles) test sets. Human evaluation on 400 claims shows NewsScope achieves 89.4% human-evaluated accuracy compared to GPT-4o-mini's 93.7% (p=0.07). NewsScope outperforms GPT-4o-mini on political claims (94.3% vs. 87.8%). A numeric grounding filter further improves accuracy to 91.6%, narrowing the gap to 2.1 percentage points. Inter-annotator agreement studies (160 claims) confirm labeling reliability (94.6% positive agreement on SUPPORTED judgments). The open-weight model enables offline deployment at approximately $15 on-demand compute (or $0 on free tiers). Code and benchmark are publicly released.
Read the full abstract

Computation and Language · Cross-listed from cs.SE

State-of-the-art Small Language Coder Model: Mify-Coder

Abhinav Parmar, Abhisek Panigrahi, Abhishek Kumar Dwivedi et al. (+93 more)

We present Mify-Coder, a 2.5B-parameter code model trained on 4.2T tokens using a compute-optimal strategy built on the Mify-2.5B foundation model. Mify-Coder achieves comparable accuracy and safety while significantly outperforming much larger baseline models on standard coding and function-calling benchmarks, demonstrating that compact models can match frontier-grade models in code generation and agent-driven workflows. Our training pipeline combines high-quality curated sources with synthetic data generated through agentically designed prompts, refined iteratively using enterprise-grade evaluation datasets. LLM-based quality filtering further enhances data density, enabling frugal yet effective training. Through disciplined exploration of CPT-SFT objectives, data mixtures, and sampling dynamics, we deliver frontier-grade code intelligence within a single continuous training trajectory. Empirical evidence shows that principled data and compute discipline allow smaller models to achieve competitive accuracy, efficiency, and safety compliance. Quantized variants of Mify-Coder enable deployment on standard desktop environments without requiring specialized hardware.
Read the full abstract

All authors Abhinav Parmar, Abhisek Panigrahi, Abhishek Kumar Dwivedi, Abhishek Bhattacharya, Adarsh Ramachandra, Aditya Choudhary, Aditya Garg, Aditya Raj, Alankrit Bhatt, Alpesh Yadav, Anant Vishnu, Ananthu Pillai, Ankush Kumar, Aryan Patnaik, Aswatha Narayanan S, Avanish Raj Singh, Bhavya Shree Gadda, Brijesh Pankajbhai Kachhadiya, Buggala Jahnavi, Chidurala Nithin Krishna, Chintan Shah, Chunduru Akshaya, Debarshi Banerjee, Debrup Dey, Deepa R., Deepika B G, Faiz ur Rahman, Gagan Gayari, Gudhi Jagadeesh Kumar Naidu, Gursimar Singh, Harshal Tyagi, Harshini K, James Mani Vathalloor, Jayarama Nettar, Jayashree Gajjam, Joe Walter Sugil George, Kamalakara Sri Krishna Tadepalli, Kamalkumar Rathinasamy, Karan Chaurasia, Karthikeyan S, Kashish Arora, Kaushal Desai, Khushboo Buwade, Kiran Manjrekar, Malikireddy Venkata Sai Likhitha, Manjunath A, Mitali Mahavir Bedmutha, Mohammed Rafee Tarafdar, Nikhil Tiwari, Nikitha K Gigi, Pavan Ravikumar, Pendyala Swarnanjali, Piyush Anand, Prakash Chandrasekar, Prasanna Bhalchandra Gawade, Prasanth Sivan, Preeti Khurana, Priyanshi Babbar, Rajab Ali Mondal, Rajesh Kumar Vissapragada, Rajeshwari Ganesan, Rajeswari Koppisetti, Ramjee R., Ramkumar Thiruppathisamy, Rani G. S., S Reka, Samarth Gupta, Sandeep Reddy Kothakota, Sarathy K, Sathyanarayana Sampath Kumar, Saurabh Kumar, Shashank Khasare, Shenbaga Devi Venkatesh Kumar, Shiva Rama Krishna Parvatham, Shoeb Shaikh, Shrishanmathi A, Shubham Pathak, Sree Samhita Koppaka, Sreenivasa Raghavan K S, Sreeram Venkatasubramanian, Suprabha Desai Bojja, Swetha R, Syed Ahmed, Chinmai Harshitha Thota, Tushar Yadav, Veeravelly Kusumitha, V V S S Prasanth Patnaik, Vidya Sri Sesetti, Vijayakeerthi K, Vikram Raj Bakshi, Vinay K K, Vinoth Kumar Loganathan, Vipin Tiwari, Vivek Kumar Shrivastav, V Venkata Sri Datta Charan, Wasim Akhtar Khan

Computation and Language

LLM-Guided Exemplar Selection for Few-Shot Wearable-Sensor Human Activity Recognition

Elsen Ronando and Sozo Inoue

In this paper, we propose an LLM-Guided Exemplar Selection framework to address a key limitation in state-of-the-art Human Activity Recognition (HAR) methods: their reliance on large labeled datasets and purely geometric exemplar selection, which often fail to distinguish similar wearable sensor activities such as walking, walking upstairs, and walking downstairs. Our method incorporates semantic reasoning via an LLM-generated knowledge prior that captures feature importance, inter-class confusability, and exemplar budget multipliers, and uses it to guide exemplar scoring and selection. These priors are combined with margin-based validation cues, PageRank centrality, hubness penalization, and facility-location optimization to obtain a compact and informative set of exemplars. Evaluated on the UCI-HAR dataset under strict few-shot conditions, the framework achieves a macro F1-score of 88.78%, outperforming classical approaches such as random sampling, herding, and k-center. The results show that LLM-derived semantic priors, when integrated with structural and geometric cues, provide a stronger foundation for selecting representative sensor exemplars in few-shot wearable-sensor HAR.
Read the full abstract

Computation and Language

Towards Efficient Post-Training via Fourier-Driven Adapter Architectures

Donggyun Bae and Jongil Park

We propose a novel framework, termed Fourier-Activated Adapter (FAA), for parameter-efficient fine-tuning of large pre-trained language models. By incorporating random Fourier features into lightweight adapter modules, FAA decomposes intermediate representations into complementary low- and high-frequency components, enabling frequency-aware modulation of semantic information. This design allows the model to selectively emphasize informative frequency bands during adaptation while preserving the representational capacity of the frozen backbone. Extensive experiments on GLUE, E2E NLG, and instruction-tuning benchmarks demonstrate that FAA consistently achieves competitive or superior performance compared to existing parameter-efficient fine-tuning methods, while maintaining low computational and memory overhead. Ablation studies further verify the effectiveness of frequency-aware activation and adaptive weighting mechanisms, highlighting FAA as a robust and efficient approach for post-training large language models.
Read the full abstract

Computation and Language

The Syntax of qulk-clauses in Yemeni Ibbi Arabic: A Minimalist Approach

Zubaida Mohammed Albadani and Mohammed Q. Shormani

This study investigates the syntax of qulk-clauses in Yemeni Ibbi Arabic (YIA) within the Minimalist Program. The construction qulk-clause, a morphologically fused form meaning 'I said,' introduces embedded declarative interrogative, and imperative clauses, often eithout complementizer. The central proposal of this paper is that qulk-clauses are biclausal structures in which qulk functions a clause-embedding predicate sec;ecting a dull CP complement. By applying core minimalist operations, viz., Merge, Move, Agree, and Spell-out, the study provides a layered syntactic analysis of qulk-clauses, for illustrating how their derivation proceeds through standard computational steps and post-syntactic processes such as Morphological Merger. The proposal also accounts for dialect-specific features like bipartite negation, cliticization, and CP embedding. The findings offer theoretical contributions to generative syntax, specifically minimalism. The study concludes raising theoretical questions concerning extending the analysis to the addressee-clause kil-k 'you said'. It also provides insights into the possibility of the universality of minimalism.
Read the full abstract

Computation and Language · Cross-listed from cs.AI

Agent2World: Learning to Generate Symbolic World Models via Adaptive Multi-Agent Feedback

Mengkang Hu, Bowei Xia, Yuran Wu et al. (+9 more)

Symbolic world models (e.g., PDDL domains or executable simulators) are central to model-based planning, but training LLMs to generate such world models is limited by the lack of large-scale verifiable supervision. Current approaches rely primarily on static validation methods that fail to catch behavior-level errors arising from interactive execution. In this paper, we propose Agent2World, a tool-augmented multi-agent framework that achieves strong inference-time world-model generation and also serves as a data engine for supervised fine-tuning, by grounding generation in multi-agent feedback. Agent2World follows a three-stage pipeline: (i) A Deep Researcher agent performs knowledge synthesis by web searching to address specification gaps; (ii) A Model Developer agent implements executable world models; And (iii) a specialized Testing Team conducts adaptive unit testing and simulation-based validation. Agent2World demonstrates superior inference-time performance across three benchmarks spanning both Planning Domain Definition Language (PDDL) and executable code representations, achieving consistent state-of-the-art results. Beyond inference, Testing Team serves as an interactive environment for the Model Developer, providing behavior-aware adaptive feedback that yields multi-turn training trajectories. The model fine-tuned on these trajectories substantially improves world-model generation, yielding an average relative gain of 30.95% over the same model before training. Project page: https://agent2world.github.io.
Read the full abstract

All authors Mengkang Hu, Bowei Xia, Yuran Wu, Ailing Yu, Yude Zou, Qiguang Chen, Shijian Wang, Jiarui Jin, Kexin Li, Wenxiang Jiao, Yuan Lu, Ping Luo

Computation and Language · Cross-listed from cs.AI

SciEvalKit: An Open-source Evaluation Toolkit for Scientific General Intelligence

Yiheng Wang, Yixin Chen, Shuo Li et al. (+33 more)

We introduce SciEvalKit, a unified benchmarking toolkit designed to evaluate AI models for science across a broad range of scientific disciplines and task capabilities. Unlike general-purpose evaluation platforms, SciEvalKit focuses on the core competencies of scientific intelligence, including Scientific Multimodal Perception, Scientific Multimodal Reasoning, Scientific Multimodal Understanding, Scientific Symbolic Reasoning, Scientific Code Generation, Science Hypothesis Generation and Scientific Knowledge Understanding. It supports six major scientific domains, spanning from physics and chemistry to astronomy and materials science. SciEvalKit builds a foundation of expert-grade scientific benchmarks, curated from real-world, domain-specific datasets, ensuring that tasks reflect authentic scientific challenges. The toolkit features a flexible, extensible evaluation pipeline that enables batch evaluation across models and datasets, supports custom model and dataset integration, and provides transparent, reproducible, and comparable results. By bridging capability-based evaluation and disciplinary diversity, SciEvalKit offers a standardized yet customizable infrastructure to benchmark the next generation of scientific foundation models and intelligent agents. The toolkit is open-sourced and actively maintained to foster community-driven development and progress in AI4Science.
Read the full abstract

All authors Yiheng Wang, Yixin Chen, Shuo Li, Yifan Zhou, Bo Liu, Hengjian Gao, Jiakang Yuan, Jia Bu, Wanghan Xu, Yuhao Zhou, Xiangyu Zhao, Zhiwang Zhou, Fengxiang Wang, Haodong Duan, Songyang Zhang, Jun Yao, Han Deng, Yizhou Wang, Jiabei Xiao, Jiaqi Liu, Encheng Su, Yujie Liu, Weida Wang, Junchi Yao, Shenghe Zheng, Haoran Sun, Runmin Ma, Xiangchao Yan, Bo Zhang, Dongzhan Zhou, Shufei Zhang, Peng Ye, Xiaosong Wang, Shixiang Tang, Wenlong Zhang, Lei Bai

Computation and Language

SmartSnap: Proactive Evidence Seeking for Self-Verifying Agents

Shaofei Cai, Yulei Qin, Haojia Lin et al. (+10 more)

Agentic reinforcement learning (RL) holds great promise for the development of autonomous agents under complex GUI tasks, but its scalability remains severely hampered by the verification of task completion. Existing task verification is treated as a passive, post-hoc process: a verifier (i.e., rule-based scoring script, reward or critic model, and LLM-as-a-Judge) analyzes the agent's entire interaction trajectory to determine if the agent succeeds. Such processing of verbose context that contains irrelevant, noisy history poses challenges to the verification protocols and therefore leads to prohibitive cost and low reliability. To overcome this bottleneck, we propose SmartSnap, a paradigm shift from this passive, post-hoc verification to proactive, in-situ self-verification by the agent itself. We introduce the Self-Verifying Agent, a new type of agent designed with dual missions: to not only complete a task but also to prove its accomplishment with curated snapshot evidences. Guided by our proposed 3C Principles (Completeness, Conciseness, and Creativity), the agent leverages its accessibility to the online environment to perform self-verification on a minimal, decisive set of snapshots. Such evidences are provided as the sole materials for a general LLM-as-a-Judge verifier to determine their validity and relevance. Experiments on mobile tasks across model families and scales demonstrate that our SmartSnap paradigm allows training LLM-driven agents in a scalable manner, bringing performance gains up to 26.08% and 16.66% respectively to 8B and 30B models. The synergizing between solution finding and evidence seeking facilitates the cultivation of efficient, self-verifying agents with competitive performance against DeepSeek V3.1 and Qwen3-235B-A22B. Code is available at: https://github.com/TencentYoutuResearch/SmartSnap
Read the full abstract

All authors Shaofei Cai, Yulei Qin, Haojia Lin, Zihan Xu, Gang Li, Yuchen Shi, Zongyi Li, Yong Mao, Siqi Cai, Xiaoyu Tan, Yitao Liang, Ke Li, Xing Sun

Computation and Language · Cross-listed from cs.LG

A2P-Vis: an Analyzer-to-Presenter Agentic Pipeline for Visual Insights Generation and Reporting

Shuyu Gan, Renxiang Wang, James Mooney et al. (+1 more)

Automating end-to-end data science pipeline with AI agents still stalls on two gaps: generating insightful, diverse visual evidence and assembling it into a coherent, professional report. We present A2P-Vis, a two-part, multi-agent pipeline that turns raw datasets into a high-quality data-visualization report. The Data Analyzer orchestrates profiling, proposes diverse visualization directions, generates and executes plotting code, filters low-quality figures with a legibility checker, and elicits candidate insights that are automatically scored for depth, correctness, specificity, depth and actionability. The Presenter then orders topics, composes chart-grounded narratives from the top-ranked insights, writes justified transitions, and revises the document for clarity and consistency, yielding a coherent, publication-ready report. Together, these agents convert raw data into curated materials (charts + vetted insights) and into a readable narrative without manual glue work. We claim that by coupling a quality-assured Analyzer with a narrative Presenter, A2P-Vis operationalizes co-analysis end-to-end, improving the real-world usefulness of automated data analysis for practitioners. For the complete dataset report, please see: https://www.visagent.org/api/output/f2a3486d-2c3b-4825-98d4-5af25a819f56.
Read the full abstract

All authors Shuyu Gan, Renxiang Wang, James Mooney, Dongyeop Kang

Computation and Language

Introducing TrGLUE and SentiTurca: A Comprehensive Benchmark for Turkish General Language Understanding and Sentiment Analysis

Duygu Altinok

Evaluating the performance of various model architectures, such as transformers, large language models (LLMs), and other NLP systems, requires comprehensive benchmarks that measure performance across multiple dimensions. Among these, the evaluation of natural language understanding (NLU) is particularly critical as it serves as a fundamental criterion for assessing model capabilities. Thus, it is essential to establish benchmarks that enable thorough evaluation and analysis of NLU abilities from diverse perspectives. While the GLUE benchmark has set a standard for evaluating English NLU, similar benchmarks have been developed for other languages, such as CLUE for Chinese, FLUE for French, and JGLUE for Japanese. However, no comparable benchmark currently exists for the Turkish language. To address this gap, we introduce TrGLUE, a comprehensive benchmark encompassing a variety of NLU tasks for Turkish. In addition, we present SentiTurca, a specialized benchmark for sentiment analysis. To support researchers, we also provide fine-tuning and evaluation code for transformer-based models, facilitating the effective use of these benchmarks. TrGLUE comprises Turkish-native corpora curated to mirror the domains and task formulations of GLUE-style evaluations, with labels obtained through a semi-automated pipeline that combines strong LLM-based annotation, cross-model agreement checks, and subsequent human validation. This design prioritizes linguistic naturalness, minimizes direct translation artifacts, and yields a scalable, reproducible workflow. With TrGLUE, our goal is to establish a robust evaluation framework for Turkish NLU, empower researchers with valuable resources, and provide insights into generating high-quality semi-automated datasets.
Read the full abstract

Computation and Language · Cross-listed from cs.LG

Unifying Learning Dynamics and Generalization in Transformers Scaling Law

Chiwun Yang

The scaling law, a cornerstone of Large Language Model (LLM) development, predicts improvements in model performance with increasing computational resources. Yet, while empirically validated, its theoretical underpinnings remain poorly understood. This work formalizes the learning dynamics of transformer-based language models as an ordinary differential equation (ODE) system, then approximates this process to kernel behaviors. Departing from prior toy-model analyses, we rigorously analyze stochastic gradient descent (SGD) training for multi-layer transformers on sequence-to-sequence data with arbitrary data distribution, closely mirroring real-world conditions. Our analysis characterizes the convergence of generalization error to the irreducible risk as computational resources scale with data, especially during the optimization process. We establish matching upper and lower bounds on the excess risk, characterized by a distinct phase transition. In the initial optimization phase, the excess risk decays exponentially relative to the computational cost ${\sf C}$. However, once a specific resource allocation threshold is crossed, the system enters a statistical phase, where the generalization error follows a power-law decay of $\Theta(\mathsf{C}^{-1/7})$. These rates are certified by complementary lower bounds – statistical, via an information-theoretic two-point reduction, and optimization-side, via a first-order oracle argument – rendering the two-stage law tight up to constants, logarithmic factors, and a condition-number gap. Beyond this unified framework, our theory derives isolated scaling laws for model size, training time, and dataset size, elucidating how each variable independently governs the bounds of generalization.
Read the full abstract

Computation and Language

Context as a Tool: Context Management for Long-Horizon SWE-Agents

Shukai Liu, Jian Yang, Bo Jiang et al. (+4 more)

Agents based on large language models have recently shown strong potential on real-world software engineering (SWE) tasks that require long-horizon interaction with repository-scale codebases. However, most existing agents rely on append-only context maintenance or passively triggered compression heuristics, which often lead to context explosion, semantic drift, and degraded reasoning in long-running interactions. We propose CAT, a new context management paradigm that elevates context maintenance to a callable tool integrated into the decision-making process of agents. CAT formalizes a structured context workspace consisting of stable task semantics, condensed long-term memory, and high-fidelity short-term interactions, and enables agents to proactively compress historical trajectories into actionable summaries at appropriate milestones. To support context management for SWE-agents, we propose a trajectory-level supervision framework, CAT-GENERATOR, based on an offline data construction pipeline that injects context-management actions into complete interaction trajectories. Using this framework, we train a context-aware model, SWE-Compressor. Experiments on SWE-Bench-Verified demonstrate that SWE-Compressor reaches a 57.6% solved rate and significantly outperforms ReAct-based agents and static compression baselines, while maintaining stable and scalable long-horizon reasoning under a bounded context budget.
Read the full abstract

All authors Shukai Liu, Jian Yang, Bo Jiang, Yizhi Li, Jinyang Guo, Xianglong Liu, Bryan Dai

Computation and Language · Cross-listed from cs.CR

Toward Secure and Compliant AI: Organizational Standards and Protocols for NLP Model Lifecycle Management

Sunil Arora and John Hastings

Natural Language Processing (NLP) systems are increasingly used in sensitive domains such as healthcare, finance, and government, where they handle large volumes of personal and regulated data. However, these systems introduce distinct risks related to security, privacy, and regulatory compliance that are not fully addressed by existing AI governance frameworks. This paper introduces the Secure and Compliant NLP Lifecycle Management Framework (SC-NLP-LMF), a comprehensive six-phase model designed to ensure the secure operation of NLP systems from development to retirement. The framework, developed through a systematic PRISMA-based review of 45 peer-reviewed and regulatory sources, aligns with leading standards, including NIST AI RMF, ISO/IEC 42001:2023, the EU AI Act, and MITRE ATLAS. It integrates established methods for bias detection, privacy protection (differential privacy, federated learning), secure deployment, explainability, and secure model decommissioning. A healthcare case study illustrates how SC-NLP-LMF detects emerging terminology drift (e.g., COVID-related language) and guides compliant model updates. The framework offers organizations a practical, lifecycle-wide structure for developing, deploying, and maintaining secure and accountable NLP systems in high-risk environments.
Read the full abstract

Computation and Language

Self-attention vector output similarities reveal how machines pay attention

Tal Halevi, Yarden Tzach, Ronit D. Gross et al. (+2 more)

The self-attention mechanism has significantly advanced the field of natural language processing, facilitating the development of advanced language-learning machines. Although its utility is widely acknowledged, the precise mechanisms of self-attention underlying its advanced learning and the quantitative characterization of this learning process remains an open research question. This study introduces a new approach for quantifying information processing within the self-attention mechanism. The analysis conducted on the BERT-12 architecture reveals that, in the final layers, the attention map focuses on sentence separator tokens, suggesting a practical approach to text segmentation based on semantic features. Based on the vector space emerging from the self-attention heads, a context similarity matrix, measuring the scalar product between two token vectors was derived, revealing distinct similarities between different token vector pairs within each head and layer. The findings demonstrated that different attention heads within an attention block focused on different linguistic characteristics, such as identifying token repetitions in a given text or recognizing a token of common appearance in the text and its surrounding context. This specialization is also reflected in the distribution of distances between token vectors with high similarity as the architecture progresses. The initial attention layers exhibit substantially long-range similarities; however, as the layers progress, a more short-range similarity develops, culminating in a preference for attention heads to create strong similarities within the same sentence. Finally, the behavior of individual heads was analyzed by examining the uniqueness of their most common tokens in their high similarity elements. Each head tends to focus on a unique token from the text and builds similarity pairs centered around it.
Read the full abstract

All authors Tal Halevi, Yarden Tzach, Ronit D. Gross, Shalom Rosner, Ido Kanter

Computation and Language

Broken Words, Broken Performance: Effect of Tokenization on Performance of LLMs

Sachin Pawar, Manoj Apte, Kshitij Jadhav et al. (+2 more)

Tokenization is the first step in training any Large Language Model (LLM), where the text is split into a sequence of tokens as per the model's fixed vocabulary. This tokenization in LLMs is different from the traditional tokenization in NLP where the text is split into a sequence of "natural" words. In LLMs, a natural word may also be broken into multiple tokens due to limited vocabulary size of the LLMs (e.g., Mistral's tokenizer splits "martial" into "mart" and "ial"). In this paper, we hypothesize that such breaking of natural words negatively impacts LLM performance on various NLP tasks. To quantify this effect, we propose a set of penalty functions that compute a tokenization penalty for a given text for a specific LLM, indicating how "bad" the tokenization is. We establish statistical significance of our hypothesis on multiple NLP tasks for a set of different LLMs.
Read the full abstract

All authors Sachin Pawar, Manoj Apte, Kshitij Jadhav, Girish Keshav Palshikar, Nitin Ramrakhiyani

Computation and Language

SWE-RM: Execution-free Feedback For Software Engineering Agents

KaShun Shum, Binyuan Hui, Jiawei Chen et al. (+6 more)

Execution-based feedback like unit testing is widely used in the development of coding agents through test-time scaling (TTS) and reinforcement learning (RL). This paradigm requires scalable and reliable collection of unit test cases to provide accurate feedback, and the resulting feedback is often sparse and cannot effectively distinguish between trajectories that are both successful or both unsuccessful. In contrast, execution-free feedback from reward models can provide more fine-grained signals without depending on unit test cases. Despite this potential, execution-free feedback for realistic software engineering (SWE) agents remains underexplored. Aiming to develop versatile reward models that are effective across TTS and RL, however, we observe that two verifiers with nearly identical TTS performance can nevertheless yield very different results in RL. Intuitively, TTS primarily reflects the model's ability to select the best trajectory, but this ability does not necessarily generalize to RL. To address this limitation, we identify two additional aspects that are crucial for RL training: classification accuracy and calibration. We then conduct comprehensive controlled experiments to investigate how to train a robust reward model that performs well across these metrics. In particular, we analyze the impact of various factors such as training data scale, policy mixtures, and data source composition. Guided by these investigations, we introduce SWE-RM, an accurate and robust reward model adopting a mixture-of-experts architecture with 30B total parameters and 3B activated during inference. SWE-RM substantially improves SWE agents on both TTS and RL performance. For example, it increases the accuracy of Qwen3-Coder-Flash from 51.6% to 62.0%, and Qwen3-Coder-Max from 67.0% to 74.6% on SWE-Bench Verified using TTS, achieving new state-of-the-art performance among open-source models.
Read the full abstract

All authors KaShun Shum, Binyuan Hui, Jiawei Chen, Lei Zhang, X. W., Jiaxi Yang, Yuzhen Huang, Junyang Lin, Junxian He

Computation and Language

Accelerate Speculative Decoding with Sparse Computation in Verification

Jikai Wang, Jianchao Tan, Yuxuan Hu et al. (+6 more)

Speculative decoding accelerates autoregressive language model inference by verifying multiple draft tokens in parallel. However, the verification stage often becomes the dominant computational bottleneck, especially for long-context inputs and mixture-of-experts (MoE) models. Existing sparsification methods are designed primarily for standard token-by-token autoregressive decoding to remove substantial computational redundancy in LLMs. This work systematically adopts different sparse methods on the verification stage of the speculative decoding and identifies structured redundancy across multiple dimensions. Based on these observations, we propose a sparse verification framework that jointly sparsifies attention, FFN, and MoE components during the verification stage to reduce the dominant computation cost. The framework further incorporates an inter-draft token and inter-layer retrieval reuse strategy to further reduce redundant computation without introducing additional training. Extensive experiments across summarization, question answering, and mathematical reasoning datasets demonstrate that the proposed methods achieve favorable efficiency-accuracy trade-offs, while maintaining stable acceptance length.
Read the full abstract

All authors Jikai Wang, Jianchao Tan, Yuxuan Hu, Jiayu Qin, Yerui Sun, Yuchen Xie, Xunliang Cai, Juntao Li, Min Zhang

Computation and Language

Explainable Statute Prediction via Attention-based Model and LLM Prompting

Sachin Pawar, Girish Keshav Palshikar, Anindita Sinha Banerjee et al. (+2 more)

In this paper, we explore the problem of automatic statute prediction where for a given case description, a subset of relevant statutes are to be predicted. Here, the term "statute" refers to a section, a sub-section, or an article of any specific Act. Addressing this problem would be useful in several applications such as AI-assistant for lawyers and legal question answering system. For better user acceptance of such Legal AI systems, we believe the predictions should also be accompanied by human understandable explanations. We propose two techniques for addressing this problem of statute prediction with explanations – (i) AoS (Attention-over-Sentences) which uses attention over sentences in a case description to predict statutes relevant for it and (ii) LLMPrompt which prompts an LLM to predict as well as explain relevance of a certain statute. AoS uses smaller language models, specifically sentence transformers and is trained in a supervised manner whereas LLMPrompt uses larger language models in a zero-shot manner and explores both standard as well as Chain-of-Thought (CoT) prompting techniques. Both these models produce explanations for their predictions in human understandable forms. We compare statute prediction performance of both the proposed techniques with each other as well as with a set of competent baselines, across two popular datasets. Also, we evaluate the quality of the generated explanations through an automated counter-factual manner as well as through human evaluation.
Read the full abstract

All authors Sachin Pawar, Girish Keshav Palshikar, Anindita Sinha Banerjee, Nitin Ramrakhiyani, Basit Ali

Computation and Language

CricBench: A Multilingual Benchmark for Evaluating LLMs in Cricket Analytics

Parth Agarwal, Navya Kommuri, Trizal Garg et al. (+7 more)

Cricket is the second most popular sport worldwide, with billions of fans seeking advanced statistical insights unavailable through standard web searches. Although LLMs have advanced significantly in Text-to-SQL tasks, their capability to handle domain-specific nuances and multilingual requirements in sports analytics remains under-explored. We present CricBench, a benchmark suite evaluating the intrinsic SQL generation abilities of LLMs on cricket data across four formats: Test, ODI, T20I, and IPL. We curate a Gold-Standard dataset of 2,654 evaluation instances across four languages (English, Hindi, Punjabi, and Telugu). We evaluate seven models, GPT-5 Mini, Claude Sonnet 4, DeepSeek R1 and V3, Qwen 235B, Llama 3.1, and Gemma 2, using schema-only prompting. No single model dominates across all formats: GPT-5 Mini leads on Test cricket (12.4% DMA), Qwen 235B leads on IPL (28.7%) and T20I (17.5%), and all models score 0% on hard ODI queries. All models show a stark disconnect between syntactic validity (>98% execution accuracy) and semantic correctness (<29% DMA), with a domain gap of 37-55 percentage points versus BIRD. To our knowledge, CricBench is the first Text-to-SQL benchmark for cricket analytics.
Read the full abstract

All authors Parth Agarwal, Navya Kommuri, Trizal Garg, Prisha Singhal, Dhruv Shah, Vaibhav Devraj, Yash Sinha, Jagat Sesh Challa, Murari Mandal, Dhruv Kumar

Computation and Language

Bridging the Copyright Gap: Do Large Vision-Language Models Recognize and Respect Copyrighted Content?

Naen Xu, Jinghuai Zhang, Changjiang Li et al. (+7 more)

Large vision-language models (LVLMs) have achieved remarkable advancements in multimodal reasoning tasks. However, their widespread accessibility raises critical concerns about potential copyright infringement. Will LVLMs accurately recognize and comply with copyright regulations when encountering copyrighted content (i.e., user input, retrieved documents) in the context? Failure to comply with copyright regulations may lead to serious legal and ethical consequences, particularly when LVLMs generate responses based on copyrighted materials (e.g., retrieved book experts, news reports). In this paper, we present a comprehensive evaluation of various LVLMs, examining how they handle copyrighted content – such as book excerpts, news articles, music lyrics, and code documentation when they are presented as visual inputs. To systematically measure copyright compliance, we introduce a large-scale benchmark dataset comprising 50,000 multimodal query-content pairs designed to evaluate how effectively LVLMs handle queries that could lead to copyright infringement. Given that real-world copyrighted content may or may not include a copyright notice, the dataset includes query-content pairs in two distinct scenarios: with and without a copyright notice. For the former, we extensively cover four types of copyright notices to account for different cases. Our evaluation reveals that even state-of-the-art closed-source LVLMs exhibit significant deficiencies in recognizing and respecting the copyrighted content, even when presented with the copyright notice. To solve this limitation, we introduce a novel tool-augmented defense framework for copyright compliance, which reduces infringement risks in all scenarios. Our findings underscore the importance of developing copyright-aware LVLMs to ensure the responsible and lawful use of copyrighted content.
Read the full abstract

All authors Naen Xu, Jinghuai Zhang, Changjiang Li, Hengyu An, Chunyi Zhou, Jun Wang, Boyu Xu, Yuyuan Li, Tianyu Du, Shouling Ji