跳到正文
研踪PaperTrail

论文库

检索计算机领域论文，按研究方向、作者或关键词筛选。

部分采集异常
正在读取账号与研究记录…

全站更新概况

+0新增论文
+0其中 AI 相关
+2已有论文修订
+100新增 AI 导读

arXiv 最近完成扫描：2026-11-15 07:45 · 全站最近入库：2026-11-14 15:09各信源进度 →

统计口径与数据新鲜度

统计 2026-11-14 07:48 — 2026-11-15 07:48（北京时间） · 点击数字查看对应简报；AI 相关包含在新增论文内。

最近完成检查：2026-11-15 07:45最近发现新增版本：2026-11-14 15:09库内论文最新更新时间：2026-11-13 02:59

每分钟刷新已保存的数据。空检查不代表采集异常；新收录可能包含历史论文。AI 导读按新增版本计数。

23,348 已收录论文19,580 AI 相关6,729 有修订112 HF 社区发现

浏览论文库

研究日报 / 周报 →按信源浏览 →

时间范围跟随排序：版本更新、本站收录或首次发表。最新收录可能包含历史论文。部分信源异常，查看状态 →

找到 17 篇论文 · 按版本更新时间排序上轮扫描起点 2026-11-15 07:44（北京时间）
新论文 · v1arXiv:2611.15131

Duo: A Reciprocal World Model for Full-Duplex Interaction

Hengyuan Cao, Biao Gong, Guoshan Lu, Qiyuan Zhang, Lin Liu, Yiding Tian, Yilei Jiang, Yuxin Ma 等 20 位作者

原文摘要

In this report, we introduce the Reciprocal World Model (RWM) and present Duo v1, a full-duplex world model that continuously perceives audiovisual inputs while responding to users in real time. The RWM treats intelligence and physical reality as equally fundamental to what constitutes a world. Existing world models largely focus on simulating the latter, leaving reasoning and decision-making as external controls rather than intrinsic parts of the world itself. In RWM, we identify two equally fundamental capabilities: Intent, which provides the cognition underlying world interactions, and Realization, which generates coherent audio-visual behavior. To realize this formulation, Duo v1 introduces three key technical designs: (1) Leveraging autoregressive modeling for continuous interaction reasoning and participation decisions, rather than restricting it to instruction understanding for diffusion models. (2) Constructing chunk-specific conditions and synchronize tokens across Intent and Realization on a unified timeline. Module-specific attention masks enforce causal dependencies both within and across chunks, preventing access to future inputs. (3) Mitigating distribution shift in both modules through dual rollout distillation, training on autoregressively generated Intent tokens and synthesized audiovisual histories from Realization. Both modules continuously perceive real-time audio, video, and text, enabling full-duplex interaction with spontaneous conversations and interruptions. Beyond conversation, users can share real objects, reshape virtual scenes, guide assistants, or play audiovisual games without mouse or keyboard controls. The release of Duo v1 makes our first step, with the next version coming soon. We look forward to broader exploration of this emerging field.

cs.CV · cs.SD

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2512.15705

Batched Early-Exit LLM Inference Done Right with DREX

Xuting Liu, Daniel Alexander, Jiaheng Lu, Siva Kesava Reddy Kakarla, Behnaz Arzani, Vincent Liu

原文摘要

Early-exit (EE) LLMs let confident tokens skip a model's deep layers, but batched serving loses most of this gain. Current systems either force tokens to exit before their own ramp is confident, an error the ramp does not bound, or save almost nothing when decoding is weight-bound, because any continuing token makes a step read the deep weights. We present DREX, a batched EE serving system that never forces a token to exit. DREX introduces cross-iteration batching: tokens that continue past the exit layer wait there, and every few steps one deep pass over all waiting tokens reads the deep weights once for several steps. Its two policies differ in what such a token's request does: holding pauses it until the deep pass, and speculative holding lets it keep decoding from the exit layer's own guess, which the next deep pass verifies. Exited tokens are trusted and never verified. Using a calibrated cost model, a controller picks speculative holding where reading weights dominates a step and guesses are often right, and holding elsewhere. DREX keeps vLLM's asynchronous scheduling by planning each deep pass from exit counts of earlier steps. With synthetic exits where 90% of tokens exit at a quarter of the depth and 40% of guesses are correct, DREX achieves 2.22-2.24$\times$ decode speedup over non-EE across batch sizes on Llama-2-70B, while per-row skipping (vSkipper's design) gains at most 7.0%. On LayerSkip models, DREX adds no statistically resolved quality loss to the EE model's own. On CNN/DM at full memory, DREX with LayerSkip achieves 1.58-1.91$\times$ decode speedup while remaining within 1.4 ROUGE-2 points of the non-EE model. These speedups are 9-63% larger than EAGLE's across the evaluated model sizes, with each method measured against its own non-EE target.

cs.DC · cs.LG

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.11114

Compositional Conformal Risk Control for Multi-Stage LLM Cascades

Yuxia Sun, Jiahuang Wang, Haoting Liu, Zhiwei Huang, Tingrui Pei, Saiqin Long

原文摘要

Multi-stage large language model (LLM) cascades balance accuracy and inference cost, but reliable deployment also requires controlling error risk and identifying its sources across stages. We propose Compositional Conformal Risk Control (CompCRC), a risk-control framework for pre-specified multi-stage LLM cascades. We decompose the risk of accepting incorrect answers into additive stage-wise contributions over the full query population, linking local risk budgets to a system-level guarantee. CompCRC constructs and freezes candidate budget allocations and acceptance thresholds, selects one candidate on independent data, and certifies its stage-wise risk contributions on separate held-out data. Under conditional exchangeability, we establish a finite-sample guarantee: the marginal expected accepted-error risk of the complete deployment procedure does not exceed the global risk budget. At the reported comparison budget of $α=0.45$, all three stages of CompCRC pass their local checks on MMLU, GSM8K, and HellaSwag. Compared with Global-CRC, which passes system-level risk certification at the same budget, CompCRC reduces observed accepted-error risk by 1.33--2.00 percentage points, improves final accuracy by 0.67--2.00 percentage points, and reduces weighted inference cost by 30.37\% on HellaSwag. The selected cost-based budget allocation also improves accuracy per unit cost by 10.6\%--46.7\% over uniform allocation. Further experiments show that stage-wise statistics identify stages that fail local certification checks, providing diagnostic information alongside system-level risk control.

cs.LG

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.10846

Efficient SC-Flip Decoding for Polar Codes through Decision-Boundary Normalization

Gulinisha Yiming, Zhongjun Yang, Zhicheng Liu, Zhiming Ma, Xumin Ni, Yuting Liu

原文摘要

Accurate first erroneous position (FEP) identification is key to improving the error-correction performance of successive cancellation flip (SCF) decoding. However, FEP identification remains challenging with a limited number of re-decoding attempts. To address this issue, we propose decision-boundary normalization (DBN) to improve FEP identification and develop a novel DBN-SCF decoding algorithm. DBN normalizes the magnitude of each decision log-likelihood ratio (LLR) by the standard deviation predicted by the Gaussian approximation (GA) for the corresponding bit-channel. This allows a single normalized threshold to be shared across bit-channels within each code configuration, giving it a consistent statistical meaning and avoiding per-channel calibration. DBN-SCF decoding selects candidate flip positions from the information set in the natural successive cancellation (SC) decoding order using a threshold, without sorting. Moreover, DBN-SCF decoding achieves better error-correction performance with a limited number of re-decoding attempts. It also requires lower average decoding complexity than existing SCF decoding methods at a comparable level of error-correction performance.

cs.IT

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2609.34427

LLMs as Adaptive Meta-Solvers: Strategy-Diverse RL for Industrial-Scale Optimization

Shihao Zhang, Weiting Liu, Siyu Shao, Yitian Chen, Jianfeng Feng, Dongdong Ge, Yinyu Ye

原文摘要

Scaling LLM-based optimization from textbook-scale instances to real-world, industrial tasks remains a critical open challenge. Existing approaches are predominantly evaluated on small, self-contained textual problems and often commit to a solver-integrated paradigm, limiting their ability to handle the scale and structural diversity of practical optimization workloads. In this work, we propose a practical framework for training open-source LLMs to tackle real-world, industrial-scale optimization. We first show empirically that solver-integrated reasoning, exact combinatorial algorithm, and heuristic search exhibit complementary strengths across different problem structures and scales. Motivated by this, we introduce Strategy-Diverse Reinforcement Learning (SDRL), which trains LLMs as adaptive optimization meta-solvers. SDRL leverages this complementarity through a correctness-gated hierarchical diversity reward that promotes robust exploration across varying strategies and within each strategy, effectively preventing premature strategy collapse. We further introduce a mixed-format training scheme that jointly supports both self-contained textual problems and file-grounded instances. Across comprehensive evaluations, our framework outperforms existing fine-tuned methods and frontier models including DeepSeek-V4-Pro and GPT-5.5, both on average across benchmarks and on industrial-scale optimization tasks. We release our code and training data to facilitate future research https://github.com/tokentideAI/LLMMetaSolver.

cs.LG · cs.CL

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2606.29073

From Tool Connection to Execution Control: Benchmarking Security Invariants in MCP-Style Agent Runtimes

Ting Liu

原文摘要

Tool-connected agents need execution-time invariants that hold even when a model follows an adversarial instruction. We formulate eight execution-layer invariants (I1-I8) covering structured authority, canonical resource resolution, principal and delegation binding, handle lifecycle, explicit source-to-target transfer, request-correlated denial evidence, and single-use approval state. To test whether these protections require a centralized runtime or follow from the invariants themselves, we compare HCP, an in-memory runtime with managed handles and tasks, against an independent gateway/server ACL (B3) enforcing the same I1-I8 contract over identical grants and labels. Both block 18 modeled violations and complete 25 workflows, while 16 source-code ablations pinpoint the checks responsible. Outside our scripted corpus, we test three boundaries. First, on 24 held-out AgentDojo tasks with live models, DeepSeek reaches attacker goals in 4 direct sessions while both controllers deny every attacker call; however, strict reference-derived grants also deny 40 non-attacker calls, and first-divergence replay under an approve-all control reaches 4 new attacker goals. Second, across 132 conditions on unmodified ChainCaps and Playwright MCP servers, per-call grants block ungranted writes but cannot stop a secret passed as an ordinary argument to an authorized sink; session source budgets block that copy but refuse unrelated actions; and explicit handle pipes or sealed reads block handle-mediated exfiltration while preserving unrelated workflows. Third, on a matched MCP backend, HCP adds 0.13 ms per workflow outside MCP calls while retaining 7.7 MB of state over one thousand workflows versus 0.7 MB for B3. Identical decisions across HCP and B3 show that enforcing explicit I1-I8 invariants at controlled entrances secures tool execution across centralized and gateway designs.

cs.CR · cs.AI

查看论文
arXiv ↗PDF ↗
版本更新 · v2arXiv:2601.22447

What Does a Sparse Autoencoder Feature Do? A Weight-Based Account

Yiting Liu, Zhi-Hong Deng

原文摘要

Sparse autoencoders (SAEs) have emerged as a powerful technique for decomposing language model representations into interpretable features. Current interpretation pipelines infer feature semantics from activation patterns, implicitly assuming every feature has a semantic explanation while overlooking the computational roles features inherit from the SAE training objective. We introduce a weight-based interpretation framework that requires no activation data and grounds claims causally through targeted feature-ablation interventions. Across Gemma-2 and Llama-3.1, three depth-dependent signatures show that SAE features inherit the base model's geometry: semantic features form a U-shape under tied embeddings but bifurcate when untied, attention participation peaks mid-layer, and the two roles couple oppositely for input- vs. output-oriented features. This weight-based view supplies the causal half missing from activation-based interpretability.

cs.LG

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.02815

Predictive Safety Guardrails Do Not Always Make LLM Agents Safer

Weixiang Zhao, Zhihao Liu, Guancheng Li, Yanyan Zhao, Bing Qin, Ting Liu

原文摘要

Large language model agents increasingly take consequential actions in digital and physical environments, where safety risks may emerge only after multiple interaction steps. This has motivated world-model-driven predictive safety guardrails, which anticipate future consequences before execution. However, existing evaluations largely focus on end-to-end safety outcomes, leaving unclear whether safety gains arise from reliable prediction and faithful use of predicted risks. In this work, we systematically diagnose the prediction-to-action pipeline along three dimensions: how world-model predictions are integrated into decision making, whether the predictions themselves are reliable, and whether their content is faithfully used. Across seven policy models and two safety environments, we find that stronger prediction involvement does not yield monotonic safety gains and can reduce task completion while substantially increasing inference cost. We further find that world-model predictions remain insufficiently reliable, especially over longer horizons, where safety prediction degrades substantially relative to near-term forecasting. Finally, we find that safety gains can persist even under corrupted predictions, while locally verified predictions still leave subsequent safety violations, showing that predictive content is not faithfully incorporated into the decision process. Together, these findings expose a fundamental prediction-to-action gap in current predictive guardrails: producing more or seemingly accurate foresight is insufficient unless that foresight is both reliable and faithfully translated into safer decisions.

cs.AI

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2611.02438

As Expected? Efficient LLM Planning via Expectation-Observation Consistency Verification

Qingfu Zhu, Huanyang Zheng, Ting Liu, Wanxiang Che

原文摘要

LLM planning often relies on reactive executors (e.g., ReAct) that generate a single action at each step to remain grounded in the latest observations, incurring substantial inference cost. Batched executors reduce this cost by predicting multiple actions in a single model invocation, yet they are prone to compounding errors, as intermediate observations are ignored. We propose an expectation-observation consistency verification (EOCV) mechanism for batched executors, preserving grounding while improving efficiency. The mechanism generates expectations of action outcomes and compares them with the resulting observations. Consistency permits continued execution, whereas mismatches trigger early termination and replanning from the current observation. Experiments on text-based and multimodal benchmarks across multiple backbones show broadly comparable average task performance between EOCV and reactive executors, with a 29-35% reduction in average weighted token usage. Moreover, the mechanism is model-agnostic and remains effective when integrated into state-of-the-art planners.

cs.CL

查看论文
arXiv ↗PDF ↗
新论文 · v1arXiv:2610.40453HF 社区推荐

PhysInsert：基于几何先验与双向物体-场景交互的物理合理物体插入

PhysInsert: Physically Plausible Object Insertion with Geometric Priors and Bidirectional Object-Scene Interactions

Yue Gong, Le Zhang, Rui Chen, Song Chang, Xiaochao Qu, Luoqi Liu, Ting Liu, Yebin Liu

AI 速读 基于摘要

提出 PhysInsert 框架，通过几何引导适配与双向物体-场景空间监督，实现参考图像驱动的物理合理物体插入。

cs.CV

arXiv ↗PDF ↗
查看原文摘要

We present PhysInsert, a framework for physically plausible reference-based object insertion that jointly adapts objects and their surrounding scenes. Existing methods often rely on masked inputs, generation-based pairs, or training pairs with limited viewpoint variation, leaving pose and illumination adaptation and bidirectional object-scene interactions insufficiently constrained. We construct the Insert World dataset with strictly aligned Blender-rendered counterfactual pairs and object references across viewpoints and lighting conditions. Building on these data, Geometry-Guided Adaptation couples a VGGT geometry stream with image and text streams under target-scene feature supervision to guide spatially consistent insertion. Complementary 2D Object-Scene Spatial Supervision jointly supervises object occupancy and environmental changes while grounding spatial instructions in object regions through image-text attention. We introduce PhysInsert-Bench with dedicated cross-environment and cross-view cases requiring illumination and pose adaptation beyond appearance copying. Experiments on PhysInsert-Bench and DreamBooth demonstrate improved image fidelity; a user study further shows more accurate placement and improved lighting integration, perspective, and pose plausibility.

新论文 · v1arXiv:2610.35540

Beyond Modeling Order: Dual-Path Spatiotemporal Representation Learning for Cross-Subject EEG Emotion Recognition

Zehui Xiao, Yi Zheng, Guixing Fu, Wenkai Lu, Lingfei He, Tao Tan, Yue Liu, Silun Wang 等 10 位作者

原文摘要

Cross-subject generalization remains a major challenge for EEG-based emotion recognition due to pronounced inter-subject variability in neural activity and emotion-related spatiotemporal patterns. Existing methods mainly alleviate inter subject differences through distribution alignment or contrastive learning, while paying limited attention to representation bias associated with fixed spatiotemporal modeling orders. We hypothesize that high-level emotion representations of the same EEG signal should remain consistent across different spatiotemporal modeling paths. Based on this hypothesis, we propose a Dual-Path Spatiotemporal Contrastive Learning (DP-STCL) framework for cross-subject EEG emotion recognition. DP-STCL employs two complementary encoders with independent parameters, namely a Temporal-to-Spatial (TS) path and a Spatial to-Temporal (ST) path, to model EEG spatiotemporal information under different integration orders. During representation learning, the same emotional stimulus is used to establish cross subject associations, while three complementary relations jointly constrain representation variations across subjects and modeling paths. This promotes stable and discriminative stimulus-related emotion representations. To further handle limited labeled target subject data, we introduce a LoRA-based few-shot adaptation strategy with low parameter overhead. Extensive cross-subject experiments on SEED, SEED-IV, SEED-V, and FACED show that DP-STCL consistently outperforms the compared methods, demonstrating its effectiveness in improving cross-subject EEG emotion recognition.

cs.LG

查看论文
arXiv ↗PDF ↗
首次观测 · 记录 #1ACL Anthology · ACL 2026:acl:2026.acl-long.1019ACL Anthology · ACL 2026

MAESTRO: Meta-learning Adaptive Estimation of Scalarization Trade-offs for Reward Optimization

Yang Zhao, Hepeng Wang Harbin Institute of Technology, Xiao Ding, Yangou Ouyang, Bibo Cai, Kai Xiong Harbin Institute of Technology, Jinglong Gao Research Center for Social Computing and Information Retrieval, Zhouhao Sun 等 11 位作者

原文摘要

Group-Relative Policy Optimization (GRPO) has emerged as an efficient paradigm for aligning Large Language Models (LLMs), yet its efficacy is primarily confined to domains with verifiable ground truths. Extending GRPO to open-domain settings remains a critical challenge, as unconstrained generation entails multi-faceted and often conflicting objectives—such as creativity versus factuality—where rigid, static reward scalarization is inherently suboptimal. To address this, we propose MAESTRO (Meta-learning Adaptive Estimation of Scalarization Trade-offs for Reward Optimization), which introduces a meta-cognitive orchestration layer that treats reward scalarization as a dynamic latent policy, leveraging the model’s terminal hidden states as a semantic bottleneck to perceive task-specific priorities. We formulate this as a contextual bandit problem within a bi-level optimization framework, where a lightweight Conductor network co-evolves with the policy by utilizing group-relative advantages as a meta-reward signal. Across seven benchmarks, MAESTRO consistently outperforms single-reward and static multi-objective baselines, while preserving the efficiency advantages of GRPO, and in some settings even reducing redundant generation.

cs.CL

Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表

首次观测 · 记录 #1ACL Anthology · ACL 2026:acl:2026.acl-long.1097ACL Anthology · ACL 2026

Consolidation or Adaptation? PRISM: Disentangling SFT and RL Data via Gradient Concentration

Yang Zhao, Yangou Ouyang, Xiao Ding, Hepeng Wang Harbin Institute of Technology, Bibo Cai, Kai Xiong Harbin Institute of Technology, Jinglong Gao Research Center for Social Computing and Information Retrieval, Zhouhao Sun 等 11 位作者

原文摘要

While Hybrid Supervised Fine-Tuning (SFT) followed by Reinforcement Learning (RL) has become the standard paradigm for training LLM agents, effective mechanisms for data allocation between these stages remain largely underexplored. Current data arbitration strategies often rely on surface-level heuristics that fail to diagnose intrinsic learning needs. Since SFT targets pattern consolidation through imitation while RL drives structural adaptation via exploration, misaligning data with these functional roles causes severe optimization interference. We propose PRISM, a dynamics-aware framework grounded in Schema Theory that arbitrates data based on its degree of cognitive conflict with the model’s existing knowledge. By analyzing the spatial geometric structure of gradients, PRISM identifies data triggering high spatial concentration as high-conflict signals that require RL for structural restructuring. In contrast, data yielding diffuse updates is routed to SFT for efficient consolidation. Extensive experiments on WebShop and ALFWorld demonstrate that PRISM achieves a Pareto improvement, outperforming state-of-the-art hybrid methods while reducing computational costs by up to 3.22 \times. Our findings suggest that disentangling data based on internal optimization regimes is crucial for scalable and robust agent alignment.

cs.CL

Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表

首次观测 · 记录 #1ACL Anthology · ACL 2026:acl:2026.acl-long.115ACL Anthology · ACL 2026

Toward Secure Tuning: Mitigating Security Risks from Instruction Fine-Tuning

Yanrui Du, Fenglei Fan City University of Hong Kong, Sendong Zhao, Jiawei Cao, Ming Ma, Danyang Zhao Harbin Institute of Technology, Shuren Qi City University of Hong Kong, Ting Liu Harbin Institute of Technology 等 9 位作者

原文摘要

Instruction Fine-Tuning (IFT) has emerged as a critical technique for customizing Large Language Models (LLMs) to meet diverse downstream applications. However, recent studies have revealed that IFT can compromise the built-in security mechanisms of LLMs, thereby posing significant security risks. Although defense methods targeting various training stages have been proposed, they either face challenges in practical deployment or exhibit instability and limited performance gains. In our study, we propose a novel SWAT method that introduces a key idea: shifting more of the learning burden onto security-robust parameters. To this end, our study investigates how module-level parameters affect LLMs’ internal security feature space, aiming to uncover robustness patterns in parameters. Guided by this analysis, we identify a robust module set (Mods_Rob) that exhibits minimal effects on LLMs’ security feature space. Leveraging this insight, SWAT proceeds in two phases: (1) a warm-up phase that preferentially trains Mods_Rob to learn low-level features with minimal security risk, followed by (2) standard tuning to achieve optimal task performance. Across diverse knowledge-intensive datasets, scenarios, and LLMs, SWAT substantially reduces security risks without sacrificing task performance gains.

cs.CL

Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表

首次观测 · 记录 #1ACL Anthology · ACL 2026:acl:2026.acl-long.1204ACL Anthology · ACL 2026

TinyJudge: Unverifiable Constraint Alignment via Lightweight Specialist Ensembles

Yirong Zeng, Yufei Liu Huawei Technologies Ltd. and Peking University, Xiao Ding, Yutai Hou Huawei Technologies Ltd., Yuxian Wang, Wu Ning Huawei Technologies Ltd., Haonan Song Huawei Technologies Ltd., Dandan Tu 等 12 位作者

原文摘要

Instruction Following (IF) is a core capability of LLMs, requiring strict adherence to diverse constraints, ranging from verifiable ones (e.g., output length) to unverifiable ones (e.g., tone). Reinforcement learning with verifiable rewards has emerged as a paradigm for IF tasks, leveraging LLM-as-a-judge to assess unverifiable constraints. However, we empirically find that this approach remains a significant bottleneck, suffering from severe reward hacking and higher computational overhead. In this work, we first analyze the generalization capabilities of unverifiable constraints and discover that specific constraints exhibit distinct, high-generalization patterns. Motivated by this, we propose TinyJudge, a framework that employs an ensemble of specialized tiny language models (e.g., 0.6B) to provide rewards for soft constraints. By distilling expertise from frontier models into these tiny models, it achieves high-precision, lightweight evaluation. Extensive evaluations across five benchmarks demonstrate that TinyJudge outperforms the baselines by ~10% in average performance and 12% in reward precision. Crucially, it also achieves a 3× speedup in total training time. Our work provides a scalable and robust path for aligning LLMs with unverifiable human instructions.

cs.CL

Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表

首次观测 · 记录 #1ACL Anthology · ACL 2026:acl:2026.acl-long.1250ACL Anthology · ACL 2026

Question Tells You Where the Answer Is: Intention-aware Long-Context KV Cache Compression

Liang Zhao, Xiaocheng Feng, Weihong Zhong Harbin Institute of Technology, Lei Huang, Kun Zhu, Baoxin Wang IFLYTEK CO.LTD., Dayong Wu, Guoping Hu IFLYTEK CO.LTD. 等 10 位作者

原文摘要

The increasing context window greatly extends the capabilities of large language models, but on the other hand, it incurs an unaffordable memory overhead and computational latency due to the increasing Key-Value (KV) cache size. Recent KV cache compression methods manage to reduce the cache size by dropping irrelevant KVs. However, these methods often fail to identify crucial KVs for generation while excluding others accurately, resulting in severe information loss. To address this gap, we propose IntentKV, an intention-aware KV cache eviction method that identifies and retains crucial KVs according to the attention distribution of intention, which semantically reflects the user’s goal and determines which part of the context is relevant. The consistency between the semantics and attention distribution is further substantiated through meticulously designed experiments. On this basis, IntentKV first distinguishes intention tokens from the vanilla context tokens based on their attention distribution distances. Then, the block-wise cumulative attention is calculated via aggregating the intention token attention. Finally, blocks that acquire high cumulative attention are picked and stored in KV cache. We evaluate our method across diverse long-context tasks and models. Results demonstrate that IntentKV can effectively maintain the model performance while reducing the KV cache size from 128K to 2K, leading to a 6.3x increase in decoding speed and 7.8x enhancement in memory efficiency compared to the default setting.

cs.CL

Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表

首次观测 · 记录 #1ACL Anthology · ACL 2026:acl:2026.acl-long.1338ACL Anthology · ACL 2026

Exploring and Distilling Multi-Dimensional Clues for Interpretable Social Bot Detection

Yi Han, Haiqi Lu, Lizi Liao Singapore Management University, Shuhan Zhou, Yuanxing Liu Harbin Institute of Technology, Weinan Zhang Harbin Institute of Technology, Ting Liu Harbin Institute of Technology

原文摘要

Social bot accounts have long been disseminating disinformation and engaging in malicious activities on social media platforms. Detecting these social bots has become a critical and urgent task, essential for maintaining a healthy online ecosystem. Existing social bot detection research usually provides detection results directly without corresponding supportive explanations, making it difficult to assess the extent to which such predictions are trustworthy. This is a key concern for online moderation. In this work, we explore the detection interpretation and summarize a four-dimensional clue framework from individual and social perspectives. We propose CDRBot, which primarily employs outcome-reward reinforcement learning to train inspectors to generate faithful, grounded, and readable clues from the User Information, Semantic Features, Interactive Situation, and Behavioral Pattern. These clues are then integrated to make final predictions. Experimental results demonstrate that our approach outperforms other baselines in detection performance. The generated clues are faithful, grounded, and readable, and can significantly enhance the performance of large language models in social bot detection.

cs.CL

Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) · 2026 · 正式发表