Computation and Language
Beyond Sequences: Distilling Structured Decision Memory for LLM Recommendation
Leikun Liang, Guoshuai Wang, Xingsheng He et al. (+4 more)
Despite the adoption of large
language models (LLMs) in recommendation systems, prevailing approaches mostly model single-type behaviors (e.g., views or purchases). Even when incorporating multiple behaviors, existing methods flatten heterogeneous actions into homogeneous
token sequences, ignoring their distinct decision-making roles. This flattening fails to capture semantic hierarchies and contextual nuances in complex decision-making, such as trade-offs between price and quality. Consequently, performance degrades in critical “difficult-choice” scenarios involving highly similar items. To bridge this gap, we propose MARI (
Memory-Augmented Recommendation with Interpretability), which grounds predictions in explicit, structured decision evidence. MARI maintains a Decision
Memory Bank (DMB) that archives users' past rationales as Structured Decision
Memories (SDMs): concise records of goals, constraints, and trade-offs. These SDMs are generated offline via Post-Hoc Decision
Distillation from heterogeneous behaviors and user-generated content. By retrieving relevant SDMs to augment LLM
reasoning, MARI achieves interpretability and scalability without the prohibitive cost of processing long raw sequences. Extensive experiments show MARI significantly outperforms state-of-the-art baselines on standard next-item prediction and a newly introduced Difficult Choice Prediction task, incurring low latency overhead by decoupling
memory construction from online inference. Qualitative analyses reveal actionable, human-readable insights into user decision-making, marking a concrete step toward
reasoning-aware recommendation systems.
Read the full abstract ▾
All authors Leikun Liang, Guoshuai Wang, Xingsheng He, Yushan Han, Yunyi Xuan, Xiaoxiao Xu, Lin Qu
Computation and Language · Cross-listed from cs.CV
SAGE: Sink-Aware Guided Emphasis for Visual Grounding in Vision-Language Decoders
Jeonghyo Song and YoungJoon Yoo
Recent large
vision-
language models (VLMs) pair a visual encoder with a large
language model (LLM) and perform well on diverse image-text tasks, yet their reliability is often limited by decoder
attention pathologies that suppress visual evidence and exacerbate
hallucinations. In this paper, we revisit visual
attention sinks and uncover a structured, layer-dependent behavior: across
prompts, early and late decoder layers exhibit
prompt-invariant
attention collapse onto the same few image regions, which we term PIS (
Prompt-Invariant Sinks), whereas mid layers become
prompt-conditioned and drive
vision-
language alignment. This split suggests that treating sinks as a uniform effect is incomplete. Building on this insight, we propose SAGE (Sink-Aware Guided Emphasis), a lightweight intervention that steers decoder
attention away from PIS and toward query-dependent regions of interest (ROIs) using
token-aligned ROI masks derived from standard
vision backbones such as CLIP, ViT, and DINOv3. Evaluated on diverse
vision-encoder + decoder-only LLM VLM families, SAGE improves visual grounding, reduces
hallucinations, and yields consistent gains across public downstream
vision-
language benchmarks, including fine-grained visual discrimination settings where localized evidence is crucial, when instantiated with backbone-derived ROI masks.
Read the full abstract ▾
Computation and Language · Cross-listed from cs.AI
Who Verifies the Verifier? Co-Evolving Inspectable Graders with Self-Improving Agents
Xing Zhang, Guanghui Wang, Yanwei Cui et al. (+4 more)
We changed the
agent: did it actually get better? Every self-improving
agent loop answers this hundreds of times, and every answer comes from a verifier. On open-ended tasks none exists, so the loop is handed a hand-written rubric or a bare LLM judge grading output from a model like itself, inviting reward hacking and shared blind spots. We make the verifier the evolving object: an inspectable expression over small, mostly deterministic drawback detectors, synthesized from clustered failures, gated at birth, and selected for agreement with a ten-item anchored reference set plus consensus over unlabeled outputs, never for the
agent's score. On MBPP+ it gains +0.21 held-out agreement over the hand-authored seed composition, on every seed, and ends ahead of the bare LLM judge it contains. One finding should change how co-evolved verifiers are validated: removing the anchor guards collapses the verifier into a vacuous always-pass grader, yet that collapsed verifier trains skills just as well. Downstream task score cannot certify a self-evolved verifier. Score does answer sufficiency, and there an evolved verifier can substitute: Double Ratchet, pairing the verifier with a lifecycle-managed skill loop, retains 88-110% of the lift that ground truth or a rubric buys the same loop, across code generation, enterprise text-to-SQL, and reference-free report generation. When evolved skills gamed the report rubric, an outer judge caught it and one added detector repaired it; the judge itself was wrong until given the task contract.
Read the full abstract ▾
All authors Xing Zhang, Guanghui Wang, Yanwei Cui, Ziyuan Li, Wei Qiu, Bing Zhu, Peiyang He
Computation and Language · Cross-listed from cs.SD
Beyond Speech Captions: Speech-Rewarded Style Planning for Conversational Text-to-Speech
Shiao Zhu, Lianbo Liu, Sizhen Lyu et al. (+3 more)
Natural-
language style descriptions provide an interpretable interface between large
language models (LLMs) and controllable text-to-
speech (TTS). However, using descriptions as pseudo-labels compresses target acoustics into text, and descriptive fidelity need not imply effective control of a particular synthesizer. We empirically show that
speech-text
alignment only weakly predicts downstream acoustic similarity among candidate instructions for the same utterance. We therefore propose
Speech-Rewarded Style Planning (SRSP), which trains a text-based style planner through a frozen downstream TTS model. Given dialogue history and response text, the planner generates candidate instructions and is optimized with group-relative policy optimization (GRPO), using the teacher-forced likelihood of target
speech tokens as the reward. On an English subset of the ISCSLP 2026 CoT-TTS corpus, SRSP achieves higher
speech-style and emotion similarity to target
speech and lower mel-cepstral distortion than the Base LLM and target-audio-informed captioning baselines. LLM-based expressive
speech evaluation further shows gains over all baselines in contextual appropriateness and reference consistency.
Read the full abstract ▾
All authors Shiao Zhu, Lianbo Liu, Sizhen Lyu, Yuzhe Wang, Sheng Li, Takahiro Shinozaki
Computation and Language
SAIL: Scientific Agentic Intelligence via a Science-Aware Loop
SAIL Model Team: Boyuan Sun, Bryan Dai, Che Liu et al. (+8 more)
We introduce SAIL, an open model with 35B total and 3B active parameters for literature research, scientific coding, and multi-step research workflows. SAIL is developed through a science-aware improvement loop:
agents built on frontier AI models analyze its task failures and construct training tasks that address the underlying capability gaps. The diagnosis examines search and evidence selection in literature tasks, scientific assumptions and
reasoning in coding, and planning and revision in longer investigations. The
agents draw on paper collections and scientific code repositories to build problems, interaction trajectories, and executable tasks with the required environments and tools. We repeat this loop over multiple development cycles and train SAIL through supervised fine-tuning, specialist training, multi-teacher on-policy
distillation, and agentic reinforcement learning. SAIL achieves competitive performance across scientific research tasks with substantially fewer parameters than leading open-weight models.
Read the full abstract ▾
All authors SAIL Model Team: Boyuan Sun, Bryan Dai, Che Liu, Chi Liu, Derek Li, Hongming Piao, Mengzhuo Chen, Xidong Wang, Yan Shu, Yinda Chen, Ziyang Zeng
Computation and Language
Adversarial Cues in Decision Models Used as Judges: The Role of Request Presentation
Hongliang Liu
An answer judge instructed to grade the final commitment should reject an explicitly wrong final value even when an earlier value matches the reference. We show that adding one colon to a candidate can violate this requirement depending on the presentation of the structured judging request. Numeric references certify the error, and paired interventions distinguish the candidate edit from the integration's presentation choices. On 200 previously unused DROP and GSM8K source clusters, the edit increased Jev's false acceptance from 1.0% to 26.0% with three output labels and from 3.0% to 26.5% with the published four-label grading instruction under sorted JSON keys. Both candidate variants were rejected under insertion presentation. These interactions passed the prespecified statistical correction even though Jev met the control thresholds under both grading configurations and presentations. Most excess acceptances occurred among candidates assigned larger numerical errors. GPT-6 Sol produced no observed cue-condition false acceptances, with missing responses unresolved. The result shows that basic judging competence can coexist with sharply different vulnerability to a fixed candidate edit across logically equivalent request presentations. The reference-aware grammar and compound ordering change limit the finding's operational scope and leave its internal cause unmeasured.
Read the full abstract ▾
Computation and Language
BioBigBird: A Sparse Attention Model for Long-Range Dependency Processing in Biomedical Text
Roshan Balaji, Pavan Kumar S, Vasudev Gupta et al. (+3 more)
While domain-specific Large
Language Models (LLMs) have encoded vast biomedical knowledge, their limited context windows often hinder a deep understanding of nuanced relationships within and across texts. To address this limitation, we introduce BioBigBird, a bidirectional
language model pre-trained on extensive biomedical literature and clinical data, specifically designed to handle long-range dependencies. BioBigBird leverages a sparse
attention mechanism to process sequences up to 4096
tokens, and its training incorporates a multi-stage process to mitigate noise from the large-scale pre-training corpus. We further enhance its performance by employing a multi-task learning (MTL) framework that jointly optimizes for Named Entity Recognition and Relation Extraction. Comprehensive evaluations on the BLURB
benchmark reveal that our MTL-enhanced BioBigBird achieves highly competitive results against state-of-the-art models. Our work contributes an effective methodology for developing powerful, long-context
language models for specialized domains, demonstrating the value of extended sequence processing for complex text analysis. Our models are publicly available at https://huggingface.co/collections/bisectgroup/biobigbird.
Read the full abstract ▾
All authors Roshan Balaji, Pavan Kumar S, Vasudev Gupta, Sreejith N, Keerthana Sridhar, Nirav Bhatt
Computation and Language
Fact over Fiction: Detection of Pathological Hallucinations in Sinhala-to-English Neural Machine Translation
Navam Obeysekara and Nevidu Jayatilleke
Neural Machine Translation (NMT) models, while capable of producing highly fluent outputs, remain vulnerable to
hallucinations, which are translations that are natural yet semantically unrelated to the source. This vulnerability is acute in low-resource settings like Sinhala-to-English, where weak cross-lingual
alignment leads to
hallucinations. This paper introduces a framework for reference-free
hallucination detection in this
language pair. We present a 45,000-sample synthetic
dataset generated through a probabilistic chain of five linguistically motivated corruption strategies, with a semantic rescue mechanism that uses character-level similarity to distinguish
hallucinations from morphological variants. We fine-tune mDeBERTa-v3 for
token-level sequence labelling, reaching a
token-level F1 of 0.841 +/- 0.001 over three seeds on a source-disjoint test set, and study a three-signal ensemble integrating neural risk scores, sequence log-probabilities, and cross-lingual semantic
embeddings (LaBSE). A source-ablation control shows that the detector relies on the Sinhala source rather than on surface artefacts of the corruption process: shuffling or removing the source reduces sentence-level AUROC from 0.970 to chance. We
benchmark eight NMT systems spanning five model families and find that detector firings vary by an order of magnitude across architectures.
Read the full abstract ▾
Computation and Language · Cross-listed from cs.LG
From a Prompt to Repertoires: Evolving Functional REpertoires Enable LLM Continual Learning
Fengyuan Liu, Yue Wang, Hangxi Guo et al. (+4 more)
Continual learning remains challenging for large
language models, which must enable models to acquire new skills and knowledge without degrading existing capabilities. Existing approaches typically address this challenge by carefully designing how model parameters are updated. In contrast,
prompt optimization avoids costly parameter updates while achieving competitive or even superior performance to reinforcement learning methods such as GRPO on individual knowledge-intensive and
reasoning tasks. This raises a natural question: Can
prompt optimization, as an efficient adaptation approach, be directly applied to continual learning? Our analysis shows that, under sequential task adaptation, it suffers from catastrophic forgetting, while optimized
prompts accumulate rules that overfit to local task distributions. To address these limitations, we propose Evolving Functional REpertoires (EFRE), which replaces a single
prompt with a repertoire of functions that evolves as new tasks arrive: compatible updates refine existing functions, while conflicting updates trigger the emergence of new ones. On a three-task continual-learning stream, EFRE achieves a final average performance 7.50 percentage points higher than GRPO. Moreover, after adaptation to the Bio task, its performance on FinQA decreases by only 1.56 percentage points, compared with 25.10 percentage points for the base
prompt optimization method. We further instantiate EFRE in a minimal
agent system and observe consistent improvements across different backbone models. Overall, these results demonstrate EFRE's strong performance in continual learning for large
language models and highlight its substantial potential for continual learning in advanced
agent systems.
Read the full abstract ▾
All authors Fengyuan Liu, Yue Wang, Hangxi Guo, Fengyuan Liu, Chenxu Wu, Yanguang Liu, Mengnan Du
Computation and Language
SignRAG: Unified Retrieval-Augmented Gloss-Free Sign Language Translation
Zhi Rao, Yucheng Zhou, Qianran Sun et al. (+8 more)
Contemporary decoder-only large
language models (LLMs) have demonstrated strong capabilities across a wide range of domains. However, existing pretraining paradigms for gloss-free sign
language translation (SLT) are largely designed around conventional encoder-decoder pretrained
language models, which limits their direct applicability to decoder-only LLMs. To address this limitation, we propose SignRAG, a unified framework combining hierarchical pretraining, target-domain
retrieval augmentation, and
retrieval-aware reinforcement fine-tuning. Hierarchical pretraining first learns linguistically grounded sign representations and then jointly aligns the sign encoder with an LLM, mitigating cross-modal optimization imbalance. For downstream adaptation, SignRAG complements parameter-based fine-tuning with a target-domain
retrieval gallery that provides instance-specific translation cues. To ensure that retrieved contexts are used appropriately, we further introduce
Retrieval Utility-Guided Reinforcement Fine-Tuning (RUG-RFT), which combines translation-quality and
retrieval-utility rewards to encourage beneficial
retrieval use while suppressing harmful reliance. Experiments on multiple SLT
benchmarks establish new state-of-the-art performance. In particular, to the best of our knowledge, SignRAG is the first gloss-free approach to outperform gloss-supervised methods across all reported metrics on CSL-Daily. Our code has been released at \href{https://github.com/shahelaojieraozhi/SignRAG}{GitHub}, together with models of different sizes to support future academic research.
Read the full abstract ▾
All authors Zhi Rao, Yucheng Zhou, Qianran Sun, Yiqing Huang, Longcan Yuan, Jiayi Hou, Chengwen Yao, Lin Cheng, Donghui Sun, Xiaoxin Chen, Jun Wan
Computation and Language · Cross-listed from cs.AI
UniData: Universal Multimodal Instruction Generation Pipeline
Jiaqi Tang, Yi-Feng Wu, Yuting Zhang et al. (+12 more)
Multimodal Large
Language Models (MLLMs) are increasingly being applied in a wider range of real-world scenarios. However, due to the substantial labor cost, creating high-quality
multimodal instruction
datasets for MLLMs remains a significant challenge. Although some methods propose to generate instruction data, they often face limitations in modality support and struggle with generating multi-round instructions. To address these problems, we introduce UniData, a universal instruction generation pipeline, to transform simple user requirements into multi-round,
multimodal instructions. Specifically, UniData first expands user requirements into multiple diverse events. Using these events, UniData then integrates an any-to-any large model for
multimodal instruction generation. Finally, UniData enhances data quality by correcting irrelevant and redundant inference flow, leveraging correlations between instruction rounds. To train this pipeline, we also build UniDataset, a
dataset comprising 20,000 entries across nine modalities for improved
multimodal generation. Our experiments demonstrate that UniData achieves SOTA performance in data quality and can also enhance the understanding and generation capabilities of other
multimodal models.
Read the full abstract ▾
All authors Jiaqi Tang, Yi-Feng Wu, Yuting Zhang, Hao Lu, Bowen Fu, Qing-Guo Chen, Xiaogang Xu, Yuwei Hu, Shiyin Lu, Wei Wei, Lei Zhang, Zhao Xu, Weihua Luo, Qifeng Chen, Ying-Cong Chen
Computation and Language
AdaptEvo: Learning under Imperfect Supervision through Confidence-Adaptive Training with Knowledge Evolution
AllSpark Team
Rule-governed contextual decision tasks require models to apply specified rules to case-specific context and evidence. Written rules can leave gaps in decision guidance and process evaluation, while reference judgments vary in their support from the rules and evidence. To address these challenges, we introduce AdaptEvo, a framework for learning under imperfect supervision that couples confidence-adaptive policy optimization with evolving decision knowledge and evaluation rubrics. Its Training module uses Confidence-Adaptive GRPO (CA-GRPO) to balance outcome and process rewards according to reference confidence. Its
Evolution module synthesizes reusable decision knowledge from recurring failures across training cases and refines process rubrics to detect overlooked errors. To support empirical evaluation, we construct an industrial
multimodal content moderation
dataset comprising a training set and In-Period and Out-of-Period test sets, with the latter collected under changed rules. Using Qwen3.6-35B-A3B, AdaptEvo achieves 61.9% exact-label accuracy and 72.2% binary decision accuracy on In-Period, exceeding GRPO by 7.5 and 3.7 percentage points, respectively. On Out-of-Period, the policy trained with CA-GRPO retains exact-label accuracy gains over the base model across evaluated checkpoints without injected decision knowledge, while GRPO declines with continued training. CA-GRPO also outperforms the tested fixed reward mixtures on both Out-of-Period metrics.
Read the full abstract ▾
Computation and Language · Cross-listed from cs.AI
RL-ARC: Calibrating Large Reasoning Models via Reasoning-guided Uncertainty
Gukhyeon Lee and SangKeun Lee
Language models (LMs) are commonly trained with Reinforcement Learning with Verifiable Rewards (RLVR) to enhance their
reasoning capabilities. However, since RLVR does not explicitly account for calibration during training, it can lead to severe calibration degradation, including overconfidence. Recent calibration-aware training methods for LMs, which incorporate objectives for uncertainty estimation into training, improve calibration but still exhibit overconfidence under distribution shift, while sacrificing
reasoning performance. To this end, we propose RL-ARC, a calibration-aware training framework that jointly leverages
reasoning confidence and answer confidence. Specifically, RL-ARC leverages
reasoning confidence as an auxiliary signal for calibrating answer confidence, applying it as
reasoning-guided regularization for correct cases and as an overconfidence penalty for incorrect cases. Comprehensive results across ID and OOD settings show that, beyond improving calibration, RL-ARC enables
reasoning models to adaptively estimate confidence based on the given question without substantially sacrificing
reasoning performance, thereby highlighting the importance of
reasoning confidence for training reliable
reasoning models.
Read the full abstract ▾
Computation and Language
Deception by Omission: Language Models Knowingly Hide Their Mistakes
Lucas Florin, Amelie Knecht, Ulysse Schaller et al. (+1 more)
Large
language models (LLMs) increasingly act as
agents with little human oversight, so potential mistakes they make can go unnoticed. Users then depend on the model to report what went wrong. An honest model discloses its mistakes, while a deceptive one conceals them. However, it is unclear how current LLMs behave in such situations. In this study, we prefill LLM trajectories with synthetic mistakes. The trajectories resemble real deployments in chat and agentic settings. Models fail to disclose their mistake in 36.4% of chat and 67.1% of agentic rollouts. In 2.4% and 5.3% of rollouts, respectively, they are aware of the mistake in their chain of thought but still deceptively conceal it. Rates vary by model: for instance, Gemini 3.5 Flash knowingly conceals mistakes in up to 19.9% of agentic rollouts. In 11.9% of chat and 51.8% of agentic rollouts, models show no awareness of mistakes, even though they reliably spot them when reviewing the same transcript as an outside observer. Our results show that, as
agents take on more tasks with less oversight, users cannot rely on them to self-report possible mistakes. Developers should instead use independent monitors that review
agent trajectories, or specifically train models to check their past actions and disclose what they find.
Read the full abstract ▾
All authors Lucas Florin, Amelie Knecht, Ulysse Schaller, Thilo Hagendorff
Computation and Language · Cross-listed from cs.FL
Type-Checking for Pattern-Based Tree Transformations
C. Aiswarya, Sahil Mhaskar and M. Praveen
We introduce and study pattern-based tree transformations. As an illustrating example, consider a source pattern $(x \cdot y) + (x \cdot z)$ and a target pattern $x \cdot (y + z)$ as a pair. This source pattern matches any expression $e$ of the form $(e_1 \cdot e_2) + (e_1 \cdot e_3)$ (by substituting $x$ with $e_1$, $y$ with $e_2$, and $z$ with $e_3$) and the pair transforms it into the expression $e_1 \cdot (e_2 + e_3)$ as dictated by the target pattern. Note that in this example, the set of expressions that match the source pattern is not a regular tree
language. We propose a model of tree transformations given by a finite representation of a (possibly infinite) set of such (source pattern, target pattern) pairs. The expressive power of this model comes at the cost of undecidability of checking equivalence. Nevertheless, we show that the type-checking problem is decidable for our model of pattern-based tree transformations. The type-checking problem asks whether applying a given transformation to trees having a given regular property (type) preserves the property. Our decision procedure is by a reduction to the emptiness problem of alternating tree automata.
Read the full abstract ▾
Computation and Language
ReCal: Calibrating Structured Pruning for On-Policy Distillation Recovery
Houcheng Jiang, Mao Zheng, Mingyang Song et al. (+4 more)
Structured
pruning reduces the deployment cost of
reasoning language models, but the resulting capability degradation can hinder subsequent on-policy
distillation (OPD) recovery. Because OPD relies on student-generated trajectories,
pruning damage that persists after offline
distillation can limit its effectiveness. We propose RECAL, Recovery-Aware Calibration, a simple plug-and-play approach that improves OPD recovery by adjusting calibration before
pruning. RECAL uses forward KL between an unpruned teacher and a pruned probe to identify teacher-supported predictions disrupted by
pruning, then reweights calibration statistics to guide existing
pruning criteria toward preserving these predictions. Across multiple models and
pruning methods, RECAL consistently improves mathematical
reasoning after OPD, achieving gains of up to 16.7 percentage points on AIME, alongside improvements in most code-generation comparisons. Further analysis shows that RECAL reduces residual damage at heavily affected
tokens and establishes performance advantages that persist through recovery. These results demonstrate the value of recovery-aware calibration for improving on-policy
distillation recovery of pruned
reasoning models.
Read the full abstract ▾
All authors Houcheng Jiang, Mao Zheng, Mingyang Song, Qiyong Zhong, Jie Sun, Tianyu Zhang, Junfeng Fang
Computation and Language
MetaEncoder: Exploring the Limit of Bi-Encoders for Multimodal System One Decision Making with Natural Language Interface
Jianpeng Cheng, Guangyu Sun, Aashu Singh et al. (+11 more)
System One models output constrained decisions and probability distributions rather than free-form text generation. While prevailing paradigms rely on structured schema objects to encode state, intent, and candidate choices, we revisit a fully natural
language-based System One interface. In this framework, both the user request and each candidate option are expressed in natural
language, supported by
multimodal (image and
video) auxiliary inputs. We introduce MetaEncoder, which fine-tunes a pre-trained Muse-Glimmer 30B decoder into an instruction-following decision-making encoder. To scale effectively across both small closed-set (< 256) and massive open-set (millions) candidate spaces, MetaEncoder employs a bi-encoder architecture trained via unidirectional contrastive learning for request-candidate
alignment. We conduct extensive evaluations across 11
benchmark suites and 190 tasks spanning
multimodal decision-making, understanding (closed-set) and
retrieval (open-set), highlighting where MetaEncoder beats SOTA
multimodal encoders, as well as its current limits on
reasoning-intensive tasks.
Read the full abstract ▾
All authors Jianpeng Cheng, Guangyu Sun, Aashu Singh, Benyu Zhang, Haixing Dai, Hossein Mansour, Jiangfan Zhang, Shlok Kumar Mishra, Wei Sun, Xuanming Cui, Yanli Liu, Qi Guo, Max Xiangjun Fan, Jun Xiao
Computation and Language
From Retrieval to Reconstruction: Constructing Evolvable Cognitive Memory for Long-Term Dialogue
Zirui Liao, Zhengxian Wu, Zhuohong Chen et al. (+4 more)
Large
Language Models (LLMs) serving as long-term dialogue
agents require
memory systems that support reliable
reasoning over extended interactions. However, existing
Retrieval-Augmented Generation (RAG) frameworks typically treat
memory as passive storage, making it difficult to distinguish source-attributed beliefs from unattributed event/fact records and to connect evidence dispersed across sessions. We introduce CogMem, a cognitive
memory architecture based on the PEC$^2$F (Person-Event-Concept-Claim-Fact) graph schema. Dedicated Claim nodes preserve the source and target of subjective statements, while Fact and Event nodes represent semantic and episodic knowledge. Dialogue turns are incrementally converted into provenance-aware graph records, consolidated into higher-level facts, and reconciled into temporally scoped Claim views when the same source provides conflicting updates. For
retrieval, a rule-based controller driven by LLM intent parsing composes four deterministic graph operators—anchoring, traversal, intersection, and evidence grounding—to reconstruct query-relevant context. Experiments on LoCoMo and LongMemEval show strong performance, especially on multi-hop, temporal, and knowledge-update tasks. Ablations and a semantic-collapse probe support complementary contributions from epistemic separation, consolidation, and agentic
retrieval. Code: https://github.com/Silent-Rain02/CogMem.
Read the full abstract ▾
All authors Zirui Liao, Zhengxian Wu, Zhuohong Chen, Yunyao Yu, Xiaoyu Liu, Yifan Xu, Haoqian Wang
Computation and Language
BeliefScope: Diagnosing Evidence-Driven Revision and Pressure-Induced Shifts in Large Language Models
Shuai Guo and Yidong Cui
A
language model may revise the same proposition after receiving genuinely relevant evidence or after receiving directional user pressure that adds no relevant fact. The observable response shift alone therefore does not reveal which source drove the change. We introduce BeliefScope, a controlled black-box framework for separating these two sources of influence around a fixed target proposition. BeliefScope crosses Evidence and Pressure with factor-specific local controls and measures response changes through probability reports, categorical judgments, and action recommendations on channel-appropriate scales. To determine when these observable contrasts support reliable attribution, we evaluate the observation design under controlled synthetic conditions. Known-truth recovery and targeted ablations establish where Evidence- and Pressure-related effects can be separated, while semi-synthetic stress tests map how that recoverability changes as the observation process becomes noisier and more heterogeneous. Across a 36-family Qwen/Llama study, with targeted 12-family checks that also include Gemma3-12B, the resulting profiles show substantial evaluation-context dependence: broad model-level differences can change under matched controls, decoding, or response interfaces, while some narrower within-model patterns remain stable. Instruction interventions further show that reduced target-aligned Pressure following can reflect either stable resistance or movement in the opposite direction. BeliefScope summarizes these measurements as a conditional belief-response profile that keeps diagnostic effects tied to the evaluation conditions under which they are observed, together with explicit validity boundaries for that diagnosis.
Read the full abstract ▾
Computation and Language
When Do We Need On-Policy Distillation? Distilling on Offline Student Rollouts Is Often Better
Siyan Zhao, Yonggan Fu, Jindong Jiang et al. (+8 more)
On-policy
distillation (OPD) has become increasingly popular for transferring teacher capabilities to student models. In this work, we ask a critical research question: Is on-policy sampling always beneficial for distilling arbitrary teacher-student pairs? We show that a simple alternative, Semi-OPD, which distills from offline rollouts generated by the initial student, can often outperform OPD in both accuracy and training efficiency. Across 17 teacher-student pairs ranging from 1.5B to 235B parameters, Semi-OPD outperforms OPD in 14 cases, with up to +13.6% accuracy and 11.4x training speedup. We further find that the choice between OPD and Semi-OPD depends on the
alignment between the initial teacher and student, quantified by an output-
token overlap ratio: OPD is beneficial only when the two are highly aligned with high overlap ratios. Our deeper investigation suggests that effective
distillation requires on-policyness w.r.t. both the student and the teacher. For misaligned pairs, student rollouts can become increasingly off-policy w.r.t. the teacher as context length grows, weakening the
distillation signal. In contrast, Semi-OPD is often more stable, as it distills on shorter contexts while covering full trajectories and exposing the student to more teacher-preferred
tokens. Beyond proposing Semi-OPD as an efficient alternative, our work motivates the community to rethink when to use OPD and to study stronger OPD variants with meaningful teacher-student pairs.
Read the full abstract ▾
All authors Siyan Zhao, Yonggan Fu, Jindong Jiang, Shih-Yang Liu, Song Bian, Byung-Kwan Lee, Sharath Turuvekere Sreenivas, Wenliang Dai, Hanrong Ye, Aditya Grover, Pavlo Molchanov
Computation and Language
REMORY: Learning Residual Memory for Context Compaction
Hanchen Xia, Baoyou Chen, Yutang Ge et al. (+5 more)
Long-horizon
agents compact their history to continue within a finite context window, but a textual summary alone may not support every subsequent decision. We introduce REMORY, a neural
memory network that supplements the summary with a bounded sequence of soft
memory tokens. Given the history and summary, the network learns to generate
tokens that help a frozen LLM approximate the continuation it would produce with the full history. The
tokens are conditioned on the summary and appended after it, forming an analogue of a residual connection along the sequence dimension. On SummHay, REMORY improves source attribution at nearly unchanged insight coverage and approaches the full-context joint score using only 5.2% of the input positions. Across long-horizon
agent benchmarks, Qwen3.8-27B and GLM-5.3-Flash show consistent gains with residual
memory. Both models also exhibit substantially fewer repeated tool outputs and tool errors on BrowseComp and Terminal-Bench 2.1.
Read the full abstract ▾
All authors Hanchen Xia, Baoyou Chen, Yutang Ge, Naihao Deng, Senqiao Yang, Zilong Dong, Weihao Yuan, Siyu Zhu
Computation and Language
Phonological Interference in Multilingual Speech Models
Moran Yanuka, Raja Giryes and Morris Alper
Phoneme-level models transcribe or generate
speech as a sequence of phonemes, the smallest sound units that distinguish words. These models enable fine-grained pronunciation control and understanding, yet often fail on input that does not match any single training
language, such as
speech alternating between two
languages, known as code-switching, or low-resource
languages absent from training. We identify a systematic failure mode behind this, phonological interference: models assume the input is in a single
language and impose its phonology, overriding local phoneme-level decisions that conflict with the assumed
language. We measure interference by how often a model retains phonemes that one
language has but the other lacks. On code-switched input, two phone recognizers (
speech-to-phoneme models) and a phoneme-conditioned text-to-
speech model lose 32% to 79% of these phonemes, but lose far fewer of the phonemes both
languages share. On unseen
languages, we find that phone recognizers impose the phonology of the training
language they assign to the
speech, and the more confident the assignment, the more they lose phonemes the unseen
language has but the assigned
language lacks. We probe the models'
language estimate from their internal activations, and trace interference to a low dimensional subspace. On monolingual
speech, steering this subspace toward another
language makes the model lose the phonemes that only the original
language uses and produce phonemes that only the target
language has. We introduce windowed
language estimation (WLE), an inference time repair that replaces the model's
language estimate in this subspace with one computed from a short window around each position. On code-switched input, WLE removes 34% to 69% of the interference in all three models, and in the recognizers it leaves monolingual performance essentially unchanged.
Read the full abstract ▾
Computation and Language
Gated Memory: Admission-Controlled Memory Formation for Conversational AI
Preeti Saraswat, Divya Neelagiri and Ajay Manoj
Personalized conversational AI relies on long-term
memory systems that extract facts from user utterances and store them in persistent vector stores. Despite progress in
retrieval, deduplication, and lifecycle management, the formation stage, the moment a fact is first written to storage has received almost no principled
attention. We identify this as the binding constraint on
memory quality in production systems. Critical contextual signals, such as the distinction between a permanent user attribute and a transient situation, exist only in the original utterance and are irreversibly lost the moment extraction produces a subject-relation-object triple. No downstream process can recover them. We propose Gated
Memory, a lightweight, modular formation framework that interposes two decision checkpoints between conversation and storage: an admission gate that evaluates every candidate fact against the full utterance context before extraction runs, and a conditional enrichment stage that grounds admitted facts through an entity scope taxonomy with
privacy constraints. The gate evaluates only the current exchange while using prior turns as read-only reference context, and produces a structured formation record. Admitted content is decomposed into atomic facts, each categorized, tagged with provenance (directly stated versus inferred), scoped to its condition of applicability, and grounded in resolved time and place, subject to a constraint that no entity absent from the context may be asserted. On the LoCoMo-10
benchmark with atypical emotional density in utterance data, Gated
Memory achieves an overall +2.6% relative improvement in LLM-judge accuracy over a strong baseline with identical
retrieval and generation, establishing formation quality as a measurable constraint on
memory performance.
Read the full abstract ▾
Computation and Language · Cross-listed from cs.LG
Why On-Policy Distillation Sometimes Fails: Vanishing Learning Signals
Lei Zhao, Qichao Zhao, Bowen Zuo et al. (+1 more)
On-policy
distillation (OPD) enables effective capability transfer between
language models, yet the mechanisms underlying its failures are not fully understood. Across code generation and mathematical
reasoning, OPD with larger-scale teachers exhibits early loss plateaus, with an average final loss reduction of 25.1% after 200 updates, compared with 96.2% for self-RL teachers, obtained by further reinforcement learning (RL) training of the initial student. To understand this difference, we analyze OPD as an idealized continuous-time dynamical system in the small-learning-rate limit. Our training-log diagnostics associate these plateaus with an early decline in a gradient-based learning-signal proxy while substantial loss remains; these measurements do not establish why the underlying gradient weakens. We further prove a local recovery guarantee for teachers sufficiently close to the initial student in a shared parameterization under regularity conditions, offering a conditional explanation for the success of self-RL teachers in our experiments. Across runs with and without loss plateaus, we observe small relative parameter changes (0.025-0.098%) and high similarity between the student's representations before and after OPD (linear CKA $>0.98$ across layers). These observations suggest that limited representation adaptation may contribute to learning-signal collapse, a hypothesis that remains to be tested. Code is available at https://github.com/leizhao7/opd-learning-signals.
Read the full abstract ▾
All authors Lei Zhao, Qichao Zhao, Bowen Zuo, Qishi Zhan
Computation and Language · Cross-listed from cs.LG
Read What Matters: Query-Adaptive Quantization for KV Caches
Siddharth Bhandari, Lucas Gretta, Krishna Balasubramanian et al. (+1 more)
KV-cache entries are stored before their future queries are known, but each decoding query needs precision in different places. We study this mismatch using separate budgets for retained bits and bits fetched per query. ReadKV stores each key and value in a progressive code whose prefixes support different reconstruction precisions. For each query, it allocates key-channel prefixes using the query, computes
attention from the reconstructed keys, and then allocates value-
token prefixes using that
attention. Stored entries remain unchanged. Each stage optimizes a calibrated distortion objective under a fixed budget; we prove exact allocation under diminishing refinement gains and relate these objectives to
attention-output error. We also exhibit a finite-dimensional
attention family where query-dependent access strictly outperforms every query-independent reader at the same read budget, even with unrestricted competing encoders and decoders. Across six base models, reading four bits on average from an eight-bit cache increases C4 perplexity by at most 0.66%, using about one quarter of the logical reads and half the retained capacity of a 16-bit cache. It is consistently more accurate than storing and fully reading four bits at the same payload-read budget. Retaining more bits than each query fetches is aimed at long-context decoding, where the cache bytes moved per step, rather than the weights, dominate cost. Long-context question answering and
retrieval on two instruction-tuned models provide additional quality evidence. On the tested 8K-
token, batch-one, single-layer workload on an NVIDIA A10G, a restricted eight-bit ReadKV reader with a two-bit mean payload-read budget has 39% lower latency than the tested TurboQuant codec.
Read the full abstract ▾
All authors Siddharth Bhandari, Lucas Gretta, Krishna Balasubramanian, Shiva Kasiviswanathan
Computation and Language · Cross-listed from cs.CV
MiniVer-V: Identifying Minimal Sufficient Evidence for Short Video Verification
Leran Chen, Lingnan Kong and Zile Cai
A core challenge in short-
video fact-checking is identifying which evidence is sufficient to support a verification conclusion. Existing approaches either give the verifier all available evidence, introducing noise, or select evidence by topical relevance, which conflates relatedness with sufficiency. We identify evidential sufficiency as the selection criterion: whether a subset of evidence is adequate to support a confident verdict without redundancy. We introduce MiniVer-V, a
benchmark of 195 short
videos with three-way verdict annotations (supported, refuted, insufficient) and 5,510
multimodal evidence units spanning visual keyframes,
speech transcripts, and web-retrieved external sources. We propose a two-layer verification framework that separates claim-
video consistency, assessed from internal evidence, from factual verdict determination, which additionally requires external corroboration. On top of it, a sufficiency-driven greedy search assembles evidence until a sufficiency threshold is met and outputs insufficient when the candidate pool is exhausted, rather than forcing a verdict. With Claude Sonnet 4, the method reaches a Macro-F1 of 0.510 using 4.5 evidence units on average (16% of the full evidence set), statistically indistinguishable from the full-evidence baseline (0.518 with 27.7 units), while significantly improving recognition of insufficient cases over the same search without abstention. The efficiency result replicates with GPT-5.5 and holds only partially with an open-weight Qwen2.5-72B verifier. Ablations show that external evidence is indispensable for factual determination, while internal
video evidence grounds the verdict in claim-
video consistency. These findings suggest that evidence-efficient verification is achievable, and that explicit abstention is needed when evidence is genuinely inadequate.
Read the full abstract ▾
Computation and Language · Cross-listed from cs.RO
SafeInferCom: Safe Inference-Time Compute via Verifier-Guided Mid-Generation Intervention for Robotic Task Planning
Weizhe Xu, Jialiang Fan, Mengyu Liu et al. (+1 more)
Large
Reasoning Language Models (LRLMs) enable multi-step
reasoning for robotic task planning, but continued
reasoning can overwrite valid intermediate plans or leave constraint violations unresolved, reducing planning reliability and wasting inference-time computation. We develop an inference-time monitor that exposes and verifies intermediate plans without disrupting the original decoding trajectory. Building on this monitor, we propose SafeInferCom, a formal verifier-guided framework that preserves valid intermediate plans and directs error correction during generation. Experiments across multiple LRLMs and planning domains reveal
reasoning-response inconsistency and limited self-correction under one-shot inference. SafeInferCom improves planning success and accelerates error correction relative to one-shot inference. When combined with iterative refinement, it further improves success while reducing
token usage compared with refinement alone. We additionally evaluate SafeInferCom in VirtualHome and provide a real-world robotic-arm demonstration.
Read the full abstract ▾
All authors Weizhe Xu, Jialiang Fan, Mengyu Liu, Fanxin Kong
Computation and Language · Cross-listed from cs.LG
The Lattice of Transition Laws
T. Y. Tsui, Jiatao Gu and Lingjie Liu
Diffusion and autoregression (AR) have long been seen as different categories of generative models, with
diffusion specialising in continuous fields and AR specialising in discrete
tokens. Recent work seeks to combine the advantages of the two models, and each hybrid fixes its decoding schedule by design. In this paper, we ask whether the performance of decoding schedules of one model can be predicted before decoding at a fixed number of steps. We describe
diffusion, AR, and models in between as paths on one corruption
lattice, and define the cost of a schedule as the dependence its parallel steps discard. The cost shows that the fewest steps of a zero-cost schedule are set by the geometry of the data, in the same way for
tokens and for continuous fields. In particular, for data that are Markov on a graph and dependent along its paths, the fewest steps equal the graph's treedepth, which is logarithmic in the length of a sequence and linear in the side length of a grid. With fewer steps than the treedepth, every schedule pays a positive cost, whose ranking we predict before decoding with a kernel of pairwise dependence estimated from pretrained weights. Across text generation, image generation, and
video generation, we verify most of the predictions about the rankings of different schedules under different metrics and
benchmarks. This work therefore provides a design principle for decoding for future AR models,
diffusion models, and anything in between. Our code is available at https://github.com/TSUITUENYUE/The-
Lattice-of-Transition-Laws.
Read the full abstract ▾
Computation and Language · Cross-listed from cs.LG
Bridging KV-Cache Quantization and Linear Attention: From Theory to Pretrained Weight Migration
Kaicheng Xiao, Liran Dong, Haotian Li et al. (+1 more)
KV-cache quantization and linear
attention are two representative approaches to tackling the storage and computational costs of
Transformers. KV-cache quantization compresses individual KV entries into discrete codes but retains all entries, whereas linear
attention recurrently aggregates multiple historical KV contributions into a fixed-size continuous state but can introduce interference. This contrast raises the question of whether per-KV compression and multi-KV aggregation can be bridged within a single mechanism for efficient
attention. We identify RAM-Net as such a bridge through soft assignments over a discrete address space. These assignments determine recurrent updates to the continuous slot state associated with each address. Under a restricted RAM-Net construction, we prove that soft address assignments extend hard quantized matching to a separable read-write overlap that locally approximates full-
attention similarity and supports recurrent aggregation. These connections further enable
Transformer-to-RAM-Net weight migration through a new path based on a soft-quantized intermediate construction. Across nine pretrained
Transformer models from 0.3B to 7B parameters, RAM-Net recovers an average of 87.1% of the teachers' accuracy gains over random guessing across six commonsense and knowledge tasks using only a 500M-
token budget per model.
Read the full abstract ▾
All authors Kaicheng Xiao, Liran Dong, Haotian Li, Guoliang Xing
Computation and Language · Cross-listed from cs.SD
Selective Listening: Mechanism-Guided Control of Audio Influence in Large Audio-Language Models
Yulin Sun, Kele Xu and Yong Dou
Large audio-
language models (LALMs) exploit
multimodal evidence, yet task-irrelevant audio can alter text-
reasoning decisions when listening is unnecessary. Aggregate Accuracy can hide this paired drift because audio-induced repairs and damages may cancel. Paired drift analysis and targeted interventions identify architecture-specific, intervention-sensitive late audio pathways as actionable control points. We introduce ICAP-Gate, which applies mechanism-guided, task-conditioned control to each model's pathway. Across four LALMs, two
reasoning benchmarks, and environmental-sound and natural-
speech interference, ICAP-Gate has lower point estimates for Influence Rate and Answer Flip than ungated inference in all 16 full-split model–condition evaluations. Fixed suppression degrades automatic
speech recognition (ASR) across all four models, whereas ICAP-Gate matches ungated ASR performance by preserving the pathway for explicit audio-demand instructions. ICAP-Gate has lower paired-drift point estimates than mitigation prompting in all four evaluated settings and provides competitive stabilization relative to eight-sample Self-Consistency while using one generation per query; in controlled ARC measurements, Self-Consistency incurs $7.0$–$9.2\times$ ungated latency. These results establish selective modality influence control as a design principle for robust
multimodal reasoning.
Read the full abstract ▾