Peer Revenue Past citations are no guarantee of future returns

Papers

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

Computer Science Computation and Language

cs.CL Watch Computation and Language

Thursday 12 February

Computation and Language

Thinking with Drafting: Optical Decompression via Logical Reconstruction

Jingxuan Wei, Honghao He, Caijun Jia et al. (+9 more)

Existing multimodal large language models have achieved high-fidelity visual perception and exploratory visual generation. However, a precision paradox persists in complex reasoning tasks: optical perception systems transcribe symbols without capturing logical topology, while pixel-based generative models produce visual artifacts lacking mathematical exactness. To bridge this gap, we propose that reasoning over visual inputs be reconceptualized as optical decompression-the process of reconstructing latent logical structures from compressed visual tokens. Guided by the axiom that Parsing is Reasoning, we introduce Thinking with Drafting (TwD), which utilizes a minimalist Domain-Specific Language (DSL) as a grounding intermediate representation. Unlike standard approaches that hallucinate answers directly, TwD forces the model to draft its mental model into executable code, rendering deterministic visual proofs for self-verification. To validate this, we present VisAlg, a visual algebra benchmark. Experiments demonstrate that TwD serve as a superior cognitive scaffold. Our work establishes a closed-loop system where visual generation acts not as a creative output but as a logical verifier, offering a generalizable path for visual reasoning.
Read the full abstract

All authors Jingxuan Wei, Honghao He, Caijun Jia, Siyuan Li, Zheng Sun, Yuhang Xu, Yuanyuan Lin, Linzhuang Sun, Yuchen Wu, Bihui Yu, Xiangxiang Zhang, Cheng Tan

Computation and Language · Cross-listed from cs.LG

DICE: Diffusion Large Language Models Excel at Generating CUDA Kernels

Haolei Bai, Lingcheng Kong, Xueyi Chen et al. (+3 more)

Diffusion large language models (dLLMs) have emerged as a compelling alternative to autoregressive (AR) LLMs, owing to their capacity for parallel token generation. This paradigm is particularly well-suited for code generation, where holistic structural planning and non-sequential refinement are critical. Despite this potential, tailoring dLLMs for CUDA kernel generation remains challenging, obstructed not only by the high specialization but also by the severe lack of high-quality training data. To address these challenges, we construct CuKe, an augmented supervised fine-tuning dataset optimized for high-performance CUDA kernels. On top of it, we propose a bi-phase curated reinforcement learning (BiC-RL) framework consisting of a CUDA kernel infilling stage and an end-to-end CUDA kernel generation stage. Leveraging this training framework, we introduce DICE, a series of diffusion large language models designed for CUDA kernel generation, spanning three parameter scales, 1.7B, 4B, and 8B. Extensive experiments on KernelBench demonstrate that DICE significantly outperforms both autoregressive and diffusion LLMs of comparable scale, establishing a new state-of-the-art for CUDA kernel generation.
Read the full abstract

All authors Haolei Bai, Lingcheng Kong, Xueyi Chen, Jianmian Wang, Zhiqiang Tao, Huan Wang

Computation and Language

Finding Sense in Nonsense with Generated Contexts: Perspectives from Humans and Language Models

Katrina Olsen and Sebastian Padó

Nonsensical and anomalous sentences have been instrumental in the development of computational models of semantic interpretation. A core challenge is to distinguish between what is merely anomalous (but can be interpreted given a supporting context) and what is truly nonsensical. However, it is unclear (a) how nonsensical, rather than merely anomalous, existing datasets are; and (b) how well LLMs can make this distinction. In this paper, we answer both questions by collecting sensicality judgments from human raters and LLMs on sentences from five semantically deviant datasets: both context-free and when providing a context. We find that raters consider most sentences at most anomalous, and only a few as properly nonsensical. We also show that LLMs are substantially skilled in generating plausible contexts for anomalous cases.
Read the full abstract

Computation and Language

PatientHub: A Unified Framework for Patient Simulation

Sahand Sabour, TszYam NG and Minlie Huang

As Large Language Models increasingly power role-playing applications, simulating patients has become a valuable tool for training counselors and scaling therapeutic assessment. However, prior work remains fragmented: existing approaches rely on incompatible, non-standardized profiles, prompts, and evaluation metrics, hindering reproducibility, fair comparison, and reuse. We introduce PatientHub, a unified and modular framework that standardizes the creation, simulation, and evaluation of LLM-based patients. Our framework provides 16 patient simulators, a graph-based orchestrator for multi-turn, multi-session interactions, and a configurable LLM-as-a-judge evaluator that supports multiple rubric types. Via our command-line interface, users can generate patient profiles, run simulations, and apply rubric-driven evaluation at the turn and session level. To demonstrate PatientHub's utility, we compare several supported simulators under a shared interaction protocol and showcase its extensibility by prototyping a new simulator variant with minimal additional code. By consolidating existing work into a single reproducible pipeline, our framework eliminates much of the infrastructure overhead that currently fragments this research and accelerates the development of new methods. Our code and data are publicly available via https://github.com/Sahandfer/PatientHub.
Read the full abstract

Computation and Language · Cross-listed from cs.AI

ThinkRouter: Efficient Reasoning via Routing Thinking between Latent and Discrete Spaces

Xin Xu, Tong Yu, Xiang Chen et al. (+3 more)

Recent work explores latent reasoning to improve reasoning efficiency by replacing explicit reasoning trajectories with continuous representations in a latent space, yet its effectiveness varies across settings. Analysis of model confidence dynamics under latent reasoning reveals that thinking trajectories ending in incorrect answers contain fewer low-confidence steps than those ending in correct answers. Meanwhile, we suggest that soft embeddings aggregated by multiple low-confidence thinking alternatives may introduce and propagate noise, leading to high confidence in unreliable reasoning trajectories. Motivated by these observations, ThinkRouter, an inference-time confidence-aware routing mechanism is proposed to avoid high confidence and noise for efficient reasoning. ThinkRouter routes thinking to the discrete token space when model confidence is low, and to the latent space otherwise. Extensive experiments on STEM reasoning and coding benchmarks across diverse large reasoning models demonstrate that ThinkRouter outperforms explicit CoT, random routing, and latent reasoning baselines in terms of accuracy, achieving an average improvement of 19.70 points in Pass@1, while reducing generation length by up to 15.55%. Further comprehensive analysis reveals that ThinkRouter can calibrate errors arising from explicit CoT and latent reasoning, and accelerates end-of-thinking token generation by globally lowering model confidence.
Read the full abstract

All authors Xin Xu, Tong Yu, Xiang Chen, Haoliang Wang, Julian McAuley, Saayan Mitra

Computation and Language · Cross-listed from cs.AI

PhyNiKCE: A Neurosymbolic Agentic Framework for Autonomous Computational Fluid Dynamics

E Fan, Lisong Shi, Zhengtong Li et al. (+1 more)

The deployment of autonomous agents for Computational Fluid Dynamics (CFD), is critically limited by the probabilistic nature of Large Language Models (LLMs), which struggle to enforce the strict conservation laws and numerical stability required for physics-based simulations. Reliance on purely semantic Retrieval Augmented Generation (RAG) often leads to "context poisoning," where agents generate linguistically plausible but physically invalid configurations due to a fundamental Semantic-Physical Disconnect. To bridge this gap, this work introduces PhyNiKCE (Physical and Numerical Knowledgeable Context Engineering), a neurosymbolic agentic framework for trustworthy engineering. Unlike standard black-box agents, PhyNiKCE decouples neural planning from symbolic validation. It employs a Symbolic Knowledge Engine that treats simulation setup as a Constraint Satisfaction Problem, rigidly enforcing physical constraints via a Deterministic RAG Engine with specialized retrieval strategies for solvers, turbulence models, and boundary conditions. Validated through rigorous OpenFOAM experiments on practical, non-tutorial CFD tasks using Gemini-2.5-Pro/Flash, PhyNiKCE demonstrates a 96% relative improvement over state-of-the-art baselines. Furthermore, by replacing trial-and-error with knowledge-driven initialization, the framework reduced autonomous self-correction loops by 59% while simultaneously lowering LLM token consumption by 17%. These results demonstrate that decoupling neural generation from symbolic constraint enforcement significantly enhances robustness and efficiency. While validated on CFD, this architecture offers a scalable, auditable paradigm for Trustworthy Artificial Intelligence in broader industrial automation.
Read the full abstract

All authors E Fan, Lisong Shi, Zhengtong Li, Chih-yung Wen

Computation and Language

Investigating Learner-Aware Design of LLM-Generated Educational Feedback

Momoka Furuhashi, Kouta Nakayama, Noboru Kawai et al. (+3 more)

Although large language models (LLMs) show promise for generating educational feedback, it remains unclear how feedback should be designed (e.g., tone and coverage) to support answer revision and learner evaluations across learner profiles. We define six feedback designs for multiple-choice biology questions, including a baseline design and five variants with additional feedback elements, and conduct an empirical study with 321 high school students. We evaluate feedback using immediate revision performance and six subjective evaluation criteria, and analyze differences in subjective evaluations across learner profiles based on personality traits. Our results show that presenting task-relevant information clearly is associated with better immediate revision performance and is favorably evaluated across learner profiles, while we observe descriptive differences in evaluation patterns, particularly for informational novelty and affective framing. These findings support further investigation of personalized LLM feedback design.
Read the full abstract

All authors Momoka Furuhashi, Kouta Nakayama, Noboru Kawai, Takashi Kodama, Saku Sugawara, Kyosuke Takami

Computation and Language

PACE: Prefix-Protected and Difficulty-Aware Compression for Efficient Reasoning

Ruixiang Feng, Yuntao Wen, Silin Zhou et al. (+14 more)

Language Reasoning Models (LRMs) achieve strong performance by scaling test-time computation but often suffer from “overthinking”, producing excessively long reasoning traces that increase latency and memory usage. Existing LRMs typically enforce conciseness with uniform length penalties, which over-compress crucial early deduction steps at the sequence level and indiscriminately penalize all queries at the group level. To solve these limitations, we propose \model, a dual-level framework for prefix-protected and difficulty-aware compression under hierarchical supervision. At the sequence level, prefix-protected optimization employs decaying mixed rollouts to maintain valid reasoning paths while promoting conciseness. At the group level, difficulty-aware penalty dynamically scales length constraints based on query complexity, maintaining exploration for harder questions while curbing redundancy on easier ones. Extensive experiments on DeepSeek-R1-Distill-Qwen (1.5B/7B) demonstrate that \model achieves a substantial reduction in token usage (up to 55.7%) while simultaneously improving accuracy (up to 4.1%) on math benchmarks, with generalization ability to code, science, and general domains.
Read the full abstract

All authors Ruixiang Feng, Yuntao Wen, Silin Zhou, Ke Shi, Yifan Wang, Ran Le, Zhenwei An, Zongchao Chen, Chen Yang, Guangyue Peng, Yiming Jia, Dongsheng Wang, Tao Zhang, Lisi Chen, Yang Song, Shen Gao, Shuo Shang

Computation and Language

Scene-Aware Memory Discrimination: Deciding Which Personal Knowledge Stays

Yijie Zhong, Mengying Guo, Zewei Wang et al. (+3 more)

Intelligent devices have become deeply integrated into everyday life, generating vast amounts of user interactions that form valuable personal knowledge. Efficient organization of this knowledge in user memory is essential for enabling personalized applications. However, current research on memory writing, management, and reading using large language models (LLMs) faces challenges in filtering irrelevant information and in dealing with rising computational costs. Inspired by the concept of selective attention in the human brain, we introduce a memory discrimination task. To address large-scale interactions and diverse memory standards in this task, we propose a Scene-Aware Memory Discrimination method (SAMD), which comprises two key components: the Gating Unit Module (GUM) and the Cluster Prompting Module (CPM). GUM enhances processing efficiency by filtering out non-memorable interactions and focusing on the salient content most relevant to application demands. CPM establishes adaptive memory standards, guiding LLMs to discern what information should be remembered or discarded. It also analyzes the relationship between user intents and memory contexts to build effective clustering prompts. Comprehensive direct and indirect evaluations demonstrate the effectiveness and generalization of our approach. We independently assess the performance of memory discrimination, showing that SAMD successfully recalls the majority of memorable data and remains robust in dynamic scenarios. Furthermore, when integrated into personalized applications, SAMD significantly enhances both the efficiency and quality of memory construction, leading to better organization of personal knowledge.
Read the full abstract

All authors Yijie Zhong, Mengying Guo, Zewei Wang, Zhongyang Li, Dandan Tu, Haofen Wang

Computation and Language · Cross-listed from cs.IR

Analytical Search

Yiteng Tu, Shuo Miao, Weihang Su et al. (+2 more)

Analytical information needs, such as trend analysis and causal impact assessment, are prevalent across various domains including law, finance, science, and much more. However, existing information retrieval paradigms, whether based on relevance-oriented document ranking or retrieval-augmented generation (RAG) with large language models (LLMs), often struggle to meet the end-to-end requirements of such tasks at the corpus scale. They either emphasize information finding rather than end-to-end problem solving, or simply treat everything as naive question answering, offering limited control over reasoning, evidence usage, and verifiability. As a result, they struggle to support analytical queries that have diverse utility concepts and high accountability requirements. In this paper, we propose analytical search as a distinct and emerging search paradigm designed to fulfill these analytical information needs. Analytical search reframes search as an evidence-governed, process-oriented analytical workflow that explicitly models analytical intent, retrieves evidence for fusion, and produces verifiable conclusions through structured, multi-step inference. We position analytical search in contrast to existing paradigms, and present a unified system framework that integrates query understanding, recall-oriented retrieval, reasoning-aware fusion, and adaptive verification. We also discuss potential research directions for the construction of analytical search engines. In this way, we highlight the conceptual significance and practical importance of analytical search and call on efforts toward the next generation of search engines that support analytical information needs.
Read the full abstract

All authors Yiteng Tu, Shuo Miao, Weihang Su, Yiqun Liu, Qingyao Ai

Computation and Language

PRIME: A Process-Outcome Alignment Benchmark for Verifiable Reasoning in Mathematics and Engineering

Xiangfeng Wang, Hangyu Guo, Yanlin Lai et al. (+11 more)

While model-based verifiers are essential for scaling Reinforcement Learning with Verifiable Rewards (RLVR), current outcome-centric verification paradigms primarily focus on the consistency between the final result and the ground truth, often neglecting potential errors in the derivation process. This leads to assigning positive rewards to correct answers produced from incorrect derivations. To bridge this gap, we introduce PRIME, a benchmark for evaluating verifiers on Process-Outcome Alignment verification in Mathematics and Engineering. Curated from a comprehensive collection of college-level STEM problems, PRIME comprises 2,530 high-difficulty samples through a consistency-based filtering pipeline. Through extensive evaluation, we find that current verifiers frequently fail to detect derivation flaws. Furthermore, we propose a process-aware RLVR training paradigm utilizing verifiers selected via PRIME. This approach substantially outperforms the outcome-only verification baseline, achieving absolute performance gains of 8.29%, 9.12%, and 7.31% on AIME24, AIME25, and Beyond-AIME, respectively, for the Qwen3-14B-Base model. Finally, we demonstrate a strong linear correlation ($R^2 > 0.92$) between verifier accuracy on PRIME and RLVR training effectiveness, validating PRIME as a reliable predictor for verifier selection.
Read the full abstract

All authors Xiangfeng Wang, Hangyu Guo, Yanlin Lai, Mitt Huang, Liang Zhao, Chengyuan Yao, Yinmin Zhang, Qi Han, Xiaoxiao Ren, Chun Yuan, Tong Xu, Zheng Ge, Xiangyu Zhang, Daxin Jiang

Computation and Language

SIGHT: Reinforcement Learning with Self-Evidence and Information-Gain Diverse Branching for Search Agent

Wenlin Zhong, Jinluan Yang, Yiquan Wu et al. (+3 more)

Reinforcement Learning (RL) has empowered Large Language Models (LLMs) to master autonomous search for complex question answering. However, particularly within multi-turn search scenarios, this interaction introduces a critical challenge: search results often suffer from high redundancy and low signal-to-noise ratios. Consequently, agents easily fall into "Tunnel Vision," where the forced interpretation of early noisy retrievals leads to irreversible error accumulation. To address these challenges, we propose SIGHT, a framework that enhances search-based reasoning through Self-Evidence Support (SES) and Information-Gain Driven Diverse Branching. SIGHT distills search results into high-fidelity evidence via SES and calculates an Information Gain score to pinpoint pivotal states where observations maximally reduce uncertainty. This score guides Dynamic Prompting Interventions - including de-duplication, reflection, or adaptive branching - to spawn new branches with SES. Finally, by integrating SES and correctness rewards via Group Relative Policy Optimization, SIGHT internalizes robust exploration strategies without external verifiers. Experiments on single-hop and multi-hop QA benchmarks demonstrate that SIGHT significantly outperforms existing approaches, particularly in complex reasoning scenarios, using fewer search steps.
Read the full abstract

All authors Wenlin Zhong, Jinluan Yang, Yiquan Wu, Yi Liu, Jianhang Yao, Kun Kuang

Computation and Language

Pretraining A Large Language Model using Distributed GPUs: A Memory-Efficient Decentralized Paradigm

Jinrui Zhang, Chaodong Xiao, Aoqi Wu et al. (+2 more)

Pretraining large language models (LLMs) typically requires centralized clusters with thousands of high-memory GPUs (e.g., H100/A100). Recent decentralized training methods reduce communication overhead by employing federated optimization; however, they still need to train the entire model on each node, remaining constrained by GPU memory limitations. In this work, we propose SParse Expert Synchronization (SPES), a memory-efficient decentralized framework for pretraining mixture-of-experts (MoE) LLMs. SPES trains only a subset of experts per node, substantially lowering the memory footprint. Each node updates its local experts and periodically synchronizes with other nodes, eliminating full-parameter transmission while ensuring efficient knowledge sharing. To mitigate limited per-expert data utilization under sparse expert updates, we introduce an expert-merging warm-up strategy, where experts exchange knowledge early in training, to rapidly establish foundational capabilities. With SPES, we train a 2B-parameter MoE LLM using 16 standalone 48GB GPUs over internet connections, which achieves competitive performance with centrally trained LLMs under similar computational budgets. We further demonstrate scalability by training a 7B model from scratch and a 9B model upcycled from a dense checkpoint, both of which match prior centralized baselines. Our code is available at https://github.com/zjr2000/SPES.
Read the full abstract

All authors Jinrui Zhang, Chaodong Xiao, Aoqi Wu, Xindong Zhang, Lei Zhang

Computation and Language · Cross-listed from cs.CR

Stop Tracking Me! Proactive Defense Against Attribute Inference Attack in LLMs

Dong Yan, Jian Liang, Ran He et al. (+1 more)

Recent studies have shown that large language models (LLMs) can infer private user attributes (e.g., age, location, gender) from user-generated text shared online, enabling rapid and large-scale privacy breaches. Existing anonymization-based defenses are coarse-grained, lacking word-level precision in anonymizing privacy-leaking elements. Moreover, they are inherently limited as altering user text to hide sensitive cues still allows attribute inference to occur through models' reasoning capabilities. To address these limitations, we propose a unified defense framework that combines fine-grained anonymization (TRACE) with inference-preventing optimization (RPS). TRACE leverages attention mechanisms and inference chain generation to identify and anonymize privacy-leaking textual elements, while RPS employs a lightweight two-stage optimization strategy to induce model rejection behaviors, thereby preventing attribute inference. Evaluations across diverse LLMs show that TRACE-RPS reduces attribute inference accuracy from around 50% to below 5% on open-source models. In addition, our approach offers strong cross-model generalization, prompt-variation robustness, and utility-privacy tradeoffs. Our code is available at https://github.com/Jasper-Yan/TRACE-RPS.
Read the full abstract

All authors Dong Yan, Jian Liang, Ran He, Tieniu Tan

Computation and Language · Cross-listed from cs.LG

Adaptive Milestone Reward for GUI Agents

Congmin Zheng, Xiaoyun Mo, Xinbei Ma et al. (+10 more)

Reinforcement Learning (RL) has emerged as a mainstream paradigm for training Mobile GUI Agents, yet it struggles with the temporal credit assignment problem inherent in long-horizon tasks. A primary challenge lies in the trade-off between reward fidelity and density: outcome reward offers high fidelity but suffers from signal sparsity, while process reward provides dense supervision but remains prone to bias and reward hacking. To resolve this conflict, we propose the Adaptive Milestone Reward (ADMIRE) mechanism. ADMIRE constructs a verifiable, adaptive reward system by anchoring trajectory to milestones, which are dynamically distilled from successful explorations. Crucially, ADMIRE integrates an asymmetric credit assignment strategy that denoises successful trajectories and scaffolds failed trajectories. Extensive experiments demonstrate that ADMIRE consistently yields over 10% absolute improvement in success rate across different base models on AndroidWorld. Moreover, the method exhibits robust generalizability, achieving strong performance across diverse RL algorithms and heterogeneous environments such as web navigation and embodied tasks.
Read the full abstract

All authors Congmin Zheng, Xiaoyun Mo, Xinbei Ma, Qiqiang Lin, Yin Zhao, Jiachen Zhu, Xingyu Lou, Jun Wang, Zhaoxiang Wang, Weiwen Liu, Zhuosheng Zhang, Yong Yu, Weinan Zhang

Computation and Language

Multimodal Fact-Level Attribution for Verifiable Reasoning

David Wan, Han Wang, Ziyang Wang et al. (+3 more)

Multimodal large language models (MLLMs) are increasingly used for real-world tasks involving multi-step reasoning and long-form generation, where reliability requires grounding model outputs in heterogeneous input sources and verifying individual factual claims. However, existing multimodal grounding benchmarks and evaluation methods focus on simplified, observation-based scenarios or limited modalities and fail to assess attribution in complex multimodal reasoning. We introduce MuRGAt (Multimodal Reasoning with Grounded Attribution), a benchmark for evaluating fact-level multimodal attribution in settings that require reasoning beyond direct observation. Given inputs spanning video, audio, and other modalities, MuRGAt requires models to generate answers with explicit reasoning and precise citations, where each citation specifies both modality and temporal segments. To enable reliable assessment, we introduce an automatic evaluation framework that strongly correlates with human judgments. Benchmarking with human and automated scores reveals that even strong MLLMs frequently hallucinate citations despite correct reasoning. Moreover, we observe a key trade-off: increasing reasoning depth or enforcing structured grounding often degrades accuracy, highlighting a significant gap between internal reasoning and verifiable attribution.
Read the full abstract

All authors David Wan, Han Wang, Ziyang Wang, Elias Stengel-Eskin, Hyunji Lee, Mohit Bansal

Computation and Language · Cross-listed from cs.CR

Jailbreaking Leaves a Trace: Understanding and Detecting Jailbreak Attacks from Internal Representations of Large Language Models

Sri Durga Sai Sowmya Kadali and Evangelos E. Papalexakis

Jailbreaking large language models (LLMs) has emerged as a critical security challenge with the widespread deployment of conversational AI systems. Adversarial users exploit these models through carefully crafted prompts to elicit restricted or unsafe outputs, a phenomenon commonly referred to as Jailbreaking. Despite numerous proposed defense mechanisms, attackers continue to develop adaptive prompting strategies, and existing models remain vulnerable. This motivates approaches that examine the internal behavior of LLMs rather than relying solely on prompt-level defenses. In this work, we study jailbreaking from both security and interpretability perspectives by analyzing how internal representations differ between jailbreak and benign prompts. We conduct a systematic layer-wise analysis across multiple open-source models, including GPT-J, LLaMA, Mistral, and the state-space model Mamba, and identify consistent latent-space patterns associated with harmful inputs. We then propose a tensor-based latent representation framework that captures structure in hidden activations and enables lightweight jailbreak detection without model fine-tuning or auxiliary LLM-based detectors. We further demonstrate that the latent signals can be used to actively disrupt jailbreak execution at inference time. On an abliterated LLaMA-3.1-8B model, selectively bypassing high-susceptibility layers blocks 78% of jailbreak attempts while preserving benign behavior on 94% of benign prompts. This intervention operates entirely at inference time and introduces minimal overhead, providing a scalable foundation for achieving stronger coverage by incorporating additional attack distributions or more refined susceptibility thresholds. Our results provide evidence that jailbreak behavior is rooted in identifiable internal structures and suggest a complementary, architecture-agnostic direction for improving LLM security.
Read the full abstract

Computation and Language

When Audio-LLMs Don't Listen: A Cross-Linguistic Study of Modality Arbitration

Jayadev Billa

When audio and text conflict, speech-enabled language models follow text far more often than they do when arbitrating between two conflicting text sources, even under explicit instructions to trust the audio. We introduce ALME (Audio-LLM Modality Evaluation), a dataset of 57,602 controlled audio-text conflict stimuli across eight languages, together with Text Dominance Ratio (TDR), which measures how often a model follows conflicting text when instructed to follow audio. Gemini 2.0 Flash and GPT-4o show TDR 10–26$\times$ higher than a baseline that replaces audio with its transcript under otherwise identical conditions (Gemini 2.0 Flash: 16.6% vs. 1.6%; GPT-4o: 23.2% vs. 0.9%). These results suggest that text dominance reflects not only information content, but also an asymmetry in arbitration accessibility, i.e., how easily the model can use competing representations at decision time. Framing the transcript as deliberately corrupted reduces TDR by 80%, whereas forcing explicit transcription increases it by 14%. A fine-tuning ablation further suggests that arbitration behavior depends more on LLM reasoning than on the audio input path alone. Across four audio-LLMs, we observe the same qualitative pattern with substantial cross-model and cross-linguistic variation.
Read the full abstract

Computation and Language

ADRD-Bench: A Preliminary LLM Benchmark for Alzheimer's Disease and Related Dementias

Guangxin Zhao, Jiahao Zheng, Malaz Boustani et al. (+4 more)

Large language models (LLMs) have shown great potential for healthcare applications. However, existing evaluation benchmarks provide minimal coverage of Alzheimer's Disease and Related Dementias (ADRD). To address this gap, we introduce ADRD-Bench, a preliminary ADRD-specific LLM benchmark. ADRD-Bench has two components: 1) ADRD Unified QA, a synthesis of 1,438 questions consolidated from seven established medical benchmarks, providing a unified assessment of clinical knowledge; and 2) ADRD Caregiving QA, a novel set of 149 questions derived from a nationally adopted, large clinical trials supported brain health management program, mitigating the lack of practical caregiving context in existing benchmarks. We evaluated 36 state-of-the-art LLMs on the proposed ADRD-Bench. Results showed that the accuracy of open-weight general models, open-weight medical models, and frontier closed-source general models ranged from 0.63 to 0.93 (mean: 0.77; std: 0.09), 0.47 to 0.93 (mean: 0.81; std: 0.14), and 0.83 to 0.93 (mean: 0.90; std: 0.03), respectively. While top-tier models achieved high accuracies (>0.9), case studies revealed inconsistent reasoning quality and stability, highlighting a critical need for domain-specific improvement to enhance LLMs' knowledge and reasoning grounded in daily caregiving data. The entire dataset is available at https://github.com/IIRL-ND/ADRD-Bench.
Read the full abstract

All authors Guangxin Zhao, Jiahao Zheng, Malaz Boustani, Jarek Nabrzyski, Yiyu Shi, Meng Jiang, Zhi Zheng

Wednesday 11 February

Computation and Language

Prompt-Dependent Ranking of Large Language Models with Uncertainty Quantification

Angel Rodrigo Avelar Menendez, Yufeng Liu and Xiaowu Dai

Rankings derived from pairwise comparisons are central to many economic and computational systems. In the context of large language models (LLMs), rankings are typically constructed from human preference data and presented as leaderboards that guide deployment decisions. However, existing approaches rely on point estimates, implicitly treating rankings as fixed objects despite substantial estimation noise and context-dependent performance variation. Acting on such rankings can lead to misallocation and welfare loss when apparent differences are not statistically meaningful. We study prompt-dependent ranking inference under pairwise human preferences and develop a framework for decision-safe rankings with statistically valid uncertainty guarantees. We model preferences using a contextual Bradley-Terry-Luce model in which the latent utility of each model depends on the input prompt. Rather than targeting point estimates of utilities, we directly conduct inference on induced rankings, constructing confidence sets based on simultaneous confidence intervals for pairwise utility differences. This approach yields statistically valid marginal and simultaneous confidence sets for prompt-specific ranks. Our framework connects recent advances in rank inference to contextual preference learning and provides tools for robust ranking-based decision-making. Empirically, using large-scale human preference data from LLM evaluations, we show that rankings vary substantially across prompt characteristics and that many apparent rank differences are not statistically distinguishable. We further demonstrate how uncertainty-aware rankings identify dominance only when supported by the data and otherwise return partial orders.
Read the full abstract

Computation and Language

Compressed Sensing for Capability Localization in Large Language Models

Anna Bair, Yixuan Even Xu, Mingjie Sun et al. (+1 more)

Large language models (LLMs) exhibit a wide range of capabilities, including mathematical reasoning, code generation, and linguistic behaviors. We show that Transformer architectures contain small subsets of attention heads that are necessary for certain capabilities. Zeroing out as few as five task-specific heads can degrade performance by up to $60\%$ on standard benchmarks measuring the capability of interest, while largely preserving performance on unrelated tasks. We introduce a compressed sensing-based method that exploits the sparsity of these heads to identify them via strategic knockouts and a small number of model evaluations. We validate these findings across Llama and Qwen models ranging from 1B to 14B parameters and a diverse set of capabilities including mathematical abilities and code generation, revealing a modular organization in which specialized capabilities are dependent on sparse, functionally distinct components. Overall, our results suggest that capability localization is a general organizational principle of Transformer language models, with implications for interpretability, model editing, and AI safety. Code is released at https://github.com/locuslab/llm-components.
Read the full abstract

All authors Anna Bair, Yixuan Even Xu, Mingjie Sun, J. Zico Kolter

Computation and Language

The CompMath-MCQ Dataset: Are LLMs Ready for Higher-Level Math?

Bianca Raimondi, Francesco Pivi, Davide Evangelista et al. (+1 more)

The evaluation of Large Language Models (LLMs) on mathematical reasoning has largely focused on elementary problems, competition-style questions, or formal theorem proving, leaving graduate-level and computational mathematics relatively underexplored. We introduce CompMath-MCQ, a new benchmark dataset for assessing LLMs on advanced mathematical reasoning in a multiple-choice setting. The dataset consists of 1{,}500 originally authored questions by professors of graduate-level courses, covering topics including Linear Algebra, Numerical Optimization, Vector Calculus, Probability, and Python-based scientific computing. Three option choices are provided for each question, with exactly one of them being correct. To ensure the absence of data leakage, all questions are newly created and not sourced from existing materials. The validity of questions is verified through a procedure based on cross-LLM disagreement, followed by manual expert review. By adopting a multiple-choice format, our dataset enables objective, reproducible, and bias-free evaluation through lm_eval library. Baseline results with state-of-the-art LLMs indicate that advanced computational mathematical reasoning remains a significant challenge. We release CompMath-MCQ at the following link: https://github.com/biancaraimondi/CompMath-MCQ.git
Read the full abstract

All authors Bianca Raimondi, Francesco Pivi, Davide Evangelista, Maurizio Gabbrielli

Computation and Language

Training-free Dropout Sampling for Semantic Token Acceptance in Speculative Decoding

Jeongtae Lee, Minjung Jo, Hyunjoon Jeong et al. (+5 more)

Speculative decoding accelerates large language model inference by proposing tokens with a lightweight draft model and selectively accepting them using a target model. This work introduces DropMatch, a novel approach that matches draft tokens to the predictive distribution of the target model via Monte Carlo dropout applied exclusively to the LM head, enabling sampling-based acceptance decisions. By generating multiple decoding paths, our method forms an empirical token distribution against which draft tokens are evaluated for consistency. This acceptance mechanism enables the model to adaptively control the size of decoding paths under an appropriate dropout probability, preventing substantial distortion of the target model predictive distribution. The proposed method operates in a training-free, data-free, and calibration-free manner, requires no architectural modification to pretrained models, and can be orthogonally integrated with a wide range of existing speculative decoding and inference acceleration techniques. Experiments across multiple benchmarks demonstrate that our approach increases acceptance length while maintaining competitive task performance, yielding inference speedups ranging from 1.09x to 1.33x over the standard baseline, and up to an additional 1.09x speedup when applied on top of EAGLE3.
Read the full abstract

All authors Jeongtae Lee, Minjung Jo, Hyunjoon Jeong, Gunho Park, Sunghyeon Woo, Joonghoon Kim, Se Jung Kwon, Dongsoo Lee

Computation and Language

Fragile Thoughts: How Large Language Models Handle Chain-of-Thought Perturbations

Ashwath Vaithinathan Aravindan and Mayank Kejriwal

Chain-of-Thought (CoT) prompting has emerged as a foundational technique for eliciting reasoning from Large Language Models (LLMs), yet the robustness of this approach to corruptions in intermediate reasoning steps remains poorly understood. This paper presents a comprehensive empirical evaluation of LLM robustness to a structured taxonomy of 5 CoT perturbation types: MathError, UnitConversion, Sycophancy, SkippedSteps, and ExtraSteps. We evaluate 13 models spanning three orders of magnitude in parameter count, testing their ability to complete mathematical reasoning tasks despite perturbations injected in the reasoning chain. Our key findings reveal heterogeneous vulnerability patterns: MathError perturbations produce the most severe degradation in small models (50-60% accuracy loss) but show strong scaling benefits; UnitConversion remains challenging across all scales (>5% loss even for midsized models); ExtraSteps incur minimal accuracy degradation (0-6%) even for the smallest of models; Sycophancy and SkippedSteps produce modest effects ($\sim$10% loss for small models) and slightly improve with scale. Scaling relationships show that model size serve as a protective factor against many perturbations but not always. These findings have direct implications for deploying LLMs in multi-stage reasoning pipelines and underscore the necessity of task-specific robustness assessments and mitigation strategies. The code and results are available at https://github.com/Mystic-Slice/CoTPerturbation
Read the full abstract

Computation and Language · Cross-listed from cs.LG

Routing Absorption in Sparse Attention: Why Random Gates Are Hard to Beat

Keston Aquino-Michaels

Learned gates can approximate sparse attention patterns on frozen transformers, yet provide limited benefit over random gates when trained jointly with the model. We investigate this difference in a controlled 31M-parameter transformer and attribute it to routing absorption: model representations co-adapt to the imposed mask, reducing the incremental benefit of learned routing. Four experiments characterize the phenomenon. Differentiable soft gating yields perplexities of 48.73 plus or minus 0.60 with learned gates and 49.83 plus or minus 0.04 with frozen random gates over three seeds. Hard top-k masking provides no gradient path to the gate scores in the tested implementation. Gates distilled onto co-adapted and dense-trained Q/K/V both achieve high F1 against oracle masks, but hard-mask deployment yields perplexities of 601.6 and 48.6, respectively. Stochastic mask training also leaves a substantial deployment penalty: dense evaluation yields 78.2 perplexity, compared with 37.3 for the dense baseline. Experiments on Qwen3-1.7B show that increasing the number of trainable attention layers reduces the gap between learned and random gates. We relate these results to co-adaptation in Mixture-of-Experts and propose parameter asymmetry between the gate and the model as a contributing mechanism. For the tested per-query, token-level gates, freezing the model provides stable routing targets and enables effective post-hoc sparsification. The results motivate random-routing controls and separate evaluation of routing quality and model adaptation in sparse attention methods.
Read the full abstract

Computation and Language · Cross-listed from cs.CY

Designing Explainable AI for Healthcare Reviews: Guidance on Adoption and Trust

Eman Alamoudi and Ellis Solaiman

Patients increasingly rely on online reviews when choosing healthcare providers, yet the sheer volume of these reviews can hinder effective decision-making. This paper summarises a mixed-methods study aimed at evaluating a proposed explainable AI system that analyses patient reviews and provides transparent explanations for its outputs. The survey (N=60) indicated broad optimism regarding usefulness (82% agreed it saves time; 78% that it highlights essentials), alongside strong demand for explainability (84% considered it important to understand why a review is classified; 82% said explanations would increase trust). Around 45% preferred combined text-and-visual explanations. Thematic analysis of open-ended survey responses revealed core requirements such as accuracy, clarity and simplicity, responsiveness, data credibility, and unbiased processing. In addition, interviews with AI experts provided deeper qualitative insights, highlighting technical considerations and potential challenges for different explanation methods. Drawing on TAM and trust in automation, the findings suggest that high perceived usefulness and transparent explanations promote adoption, whereas complexity and inaccuracy hinder it. This paper contributes actionable design guidance for layered, audience-aware explanations in healthcare review systems.
Read the full abstract

Computation and Language

The Million-Label NER: Breaking Scale Barriers with GLiNER bi-encoder

Ihor Stepanov, Mykhailo Shtopko, Dmytro Vodianytskyi et al. (+1 more)

This paper introduces GLiNER-bi-Encoder, a novel architecture for Named Entity Recognition (NER) that harmonizes zero-shot flexibility with industrial-scale efficiency. While the original GLiNER framework offers strong generalization, its joint-encoding approach suffers from quadratic complexity as the number of entity labels increases. Our proposed bi-encoder design decouples the process into a dedicated label encoder and a context encoder, effectively removing the context-window bottleneck. This architecture enables the simultaneous recognition of thousands, and potentially millions, of entity types with minimal overhead. Experimental results demonstrate state-of-the-art zero-shot performance, achieving 61.5 percent Micro-F1 on the CrossNER benchmark. Crucially, by leveraging pre-computed label embeddings, GLiNER-bi-Encoder achieves up to a 130 times throughput improvement at 1024 labels compared to its uni-encoder predecessors. Furthermore, we introduce GLiNKER, a modular framework that leverages this architecture for high-performance entity linking across massive knowledge bases such as Wikidata.
Read the full abstract

All authors Ihor Stepanov, Mykhailo Shtopko, Dmytro Vodianytskyi, Oleksandr Lukashov

Computation and Language

MultiCube-RAG for Multi-hop Question Answering

Jimeng Shi, Wei Hu, Runchu Tian et al. (+8 more)

Multi-hop question answering (QA) necessitates multi-step reasoning and retrieval across interconnected subjects, attributes, and relations. Existing retrieval-augmented generation (RAG) methods struggle to capture these structural semantics accurately, resulting in suboptimal performance. Graph-based RAGs structure such information in graphs, but the resulting graphs are often noisy and computationally expensive. Moreover, most methods rely on single-step retrieval, neglecting the need for multi-hop reasoning processes. Recent training-based approaches attempt to incentivize the large language models (LLMs) for iterative reasoning and retrieval, but their training processes are prone to unstable convergence and high computational overhead. To address these limitations, we devise an ontology-based cube structure with multiple and orthogonal dimensions to model structural subjects, attributes, and relations. Built on the cube structure, we propose MultiCube-RAG, a training-free method consisting of multiple cubes for multi-step reasoning and retrieval. Each cube specializes in modeling a class of subjects, so that MultiCube-RAG flexibly selects the most suitable cubes to acquire the relevant knowledge precisely. To enhance the query-based reasoning and retrieval, our method decomposes a complex multi-hop query into a set of simple subqueries along cube dimensions and conquers each of them sequentially. Experiments on four multi-hop QA datasets show that MultiCube-RAG improves response accuracy by 8.9% over the average performance of various baselines. Notably, we also demonstrate that our method performs with greater efficiency and inherent explainability.
Read the full abstract

All authors Jimeng Shi, Wei Hu, Runchu Tian, Bowen Jin, Wonbin Kweon, SeongKu Kang, Yunfan Kang, Dingqi Ye, Sizhe Zhou, Shaowen Wang, Jiawei Han

Computation and Language

Mitigating Gradient Inversion Risks in Language Models via Token Obfuscation

Xinguo Feng, Zhongkui Ma, Zihan Wang et al. (+2 more)

Training and fine-tuning large-scale language models largely benefit from collaborative learning, but the approach has been proven vulnerable to gradient inversion attacks (GIAs), which allow adversaries to reconstruct private training data from shared gradients. Existing defenses mainly employ gradient perturbation techniques, e.g., noise injection or gradient pruning, to disrupt GIAs' direct mapping from gradient space to token space. However, these methods often fall short due to the retention of semantics similarity across gradient, embedding, and token spaces. In this work, we propose a novel defense mechanism named GHOST (gradient shield with obfuscated tokens), a token-level obfuscation mechanism that neutralizes GIAs by decoupling the inherent connections across gradient, embedding, and token spaces. GHOST is built upon an important insight: due to the large scale of the token space, there exist semantically distinct yet embedding-proximate tokens that can serve as the shadow substitutes of the original tokens, which enables a semantic disconnection in the token space while preserving the connection in the embedding and gradient spaces. GHOST comprises a searching step, which identifies semantically distinct candidate tokens using a multi-criteria searching process, and a selection step, which selects optimal shadow tokens to ensure minimal disruption to features critical for training by preserving alignment with the internal outputs produced by original tokens. Evaluation across diverse model architectures (from BERT to Llama) and datasets demonstrates the remarkable effectiveness of GHOST in protecting privacy (as low as 1% in recovery rate) and preserving utility (up to 0.92 in classification F1 and 5.45 in perplexity), in both classification and generation tasks against state-of-the-art GIAs and adaptive attack scenarios.
Read the full abstract

All authors Xinguo Feng, Zhongkui Ma, Zihan Wang, Alsharif Abuadbba, Guangdong Bai

Computation and Language

Every Little Helps: Building Knowledge Graph Foundation Model with Fine-grained Transferable Multi-modal Tokens

Yichi Zhang, Zhuo Chen, Lingbing Guo et al. (+2 more)

Multi-modal knowledge graph reasoning (MMKGR) aims to predict the missing links by exploiting both graph structure information and multi-modal entity contents. Most existing works are designed for a transductive setting, which learns dataset-specific embeddings and struggles to generalize to new KGs. Recent knowledge graph foundation models (KGFMs) improve cross-KG transfer, but they mainly exploit structural patterns and ignore rich multi-modal signals. We address these gaps by proposing a token-based foundation model (TOFU) for MMKGR, which exhibits strong generalization across different MMKGs. TOFU discretizes structural, visual, and textual information into modality-specific tokens. TOFU then employs a hierarchical fusion architecture with mixture-of-message mechanisms, aiming to process these tokens and obtain transferable features for MMKGR. Experimental results on 17 transductive, inductive, and fully-inductive MMKGs show that TOFU generally outperforms strong KGFM and MMKGR baselines, delivering strong performance on unseen MMKGs.
Read the full abstract

All authors Yichi Zhang, Zhuo Chen, Lingbing Guo, Wen Zhang, Huajun Chen