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

+197新增论文
+151其中 AI 相关
+16已有论文修订
+91新增 AI 导读

arXiv 最近完成扫描:2026-11-22 06:41 · 全站最近入库:2026-11-21 12:19各信源进度 →

统计口径与数据新鲜度

统计 2026-11-21 07:16 — 2026-11-22 07:16(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:2026-11-22 06:41最近发现新增版本:2026-11-21 12:19库内论文最新更新时间:2026-11-20 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。

29,158 已收录论文24,181 AI 相关8,992 有修订112 HF 社区发现

浏览论文库

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 21,490 篇论文 · 按版本更新时间排序上轮扫描起点 2026-11-22 06:39(北京时间)
版本更新 · v4arXiv:2602.02343

Why Steering Works: Toward a Unified View of Language Model Parameter Dynamics

Ziwen Xu, Chenyan Wu, Hengyu Sun, Haiwen Hong, Mengru Wang, Yunzhi Yao, Longtao Huang, Hui Xue 等 12 位作者

原文摘要

Methods for controlling large language models (LLMs), including local weight fine-tuning, LoRA-based adaptation, and activation-based interventions, are often studied in isolation, obscuring their connections and making comparison difficult. In this work, we present a unified view that frames these interventions as dynamic weight updates induced by a control signal, placing them within a single conceptual framework. Building on this view, we propose a unified preference-utility analysis that separates control effects into preference, defined as the tendency toward a target concept, and utility, defined as coherent and task-valid generation, and measures both on a shared log-odds scale using polarity-paired contrastive examples. Across methods, we observe a consistent trade-off between preference and utility: stronger control increases preference while predictably reducing utility. We further explain this behavior through an activation manifold perspective, in which control shifts representations along target-concept directions to enhance preference, while utility declines primarily when interventions push representations off the model's valid-generation manifold. Finally, we introduce a new steering approach SPLIT guided by this analysis that improves preference while better preserving utility. Code is available at https://github.com/zjunlp/EasyEdit/blob/main/examples/SPLIT.md.

cs.CL · cs.AI · cs.CV · cs.IR · cs.LG

版本更新 · v3arXiv:2603.04836

Beyond Text: Text-Image Fusion for Scalable Two-Tower Retrieval

Qujiaheng Zhang, Guangyue Xu, Fengjie Li

原文摘要

E-commerce customers make purchase decisions by jointly considering product text and visual information, yet most industrial retrieval systems rely almost exclusively on textual signals. Deployed at scale on CPU-based nearest-neighbor infrastructure, these systems leave substantial visual evidence untapped. In this work, we study unified text-image fusion for two-tower retrieval models at a large-scale e-commerce platform, training on 20M query-item pairs. We demonstrate that domain-specific fine-tuning and modality-specific query alignment between user queries and both product text and image modalities are crucial for effective multimodal retrieval. Building on these insights, we propose a mixture-of-modality-experts fusion architecture with a bilinear interaction network to capture cross-modal complementary information. Our model achieves up to 4.86% improvement in nDCG@1 for desirability and 2.36% for relevance over a strong text-only baseline, while preserving the pre-computable item embedding design required for production deployment.

cs.IR

版本更新 · v3arXiv:2603.02578

How Controllable Are Large Language Models? A Unified Evaluation across Behavioral Granularities

Ziwen Xu, Kewei Xu, Haoming Xu, Haiwen Hong, Longtao Huang, Hui Xue, Ningyu Zhang, Yongliang Shen 等 11 位作者

原文摘要

Large Language Models (LLMs) are increasingly deployed in socially sensitive domains, yet their unpredictable behaviors, ranging from misaligned intent to inconsistent personality, pose significant risks. We introduce SteerEval, a hierarchical benchmark for evaluating LLM controllability across three domains: language features, sentiment, and personality. Each domain is structured into three specification levels: L1 (what to express), L2 (how to express), and L3 (how to instantiate), connecting high-level behavioral intent to concrete textual output. Using SteerEval, we systematically evaluate contemporary steering methods, revealing that control often degrades at finer-grained levels. Our benchmark offers a principled and interpretable framework for safe and controllable LLM behavior, serving as a foundation for future research.

cs.CL · cs.AI · cs.HC · cs.LG

新论文 · v1arXiv:2611.20507

VoxSAOR: Semantic-Aware Occupancy Robustification via Aleatoric Uncertainty Modeling

Mengfei Duan, Yuheng Zhang, Guoqiang Zhao, Yuanfan Zheng, Kailun Yang, Zhiyong Li

原文摘要

Vision-based 3D occupancy prediction provides a dense semantic representation of scenes for autonomous driving and embodied perception, yet remains vulnerable to occlusion, depth ambiguity, and projection errors that induce spatially varying reliability in voxel features. Existing methods typically map voxel representations to semantic predictions without explicitly modeling such observation-induced feature variations. To address this limitation, we propose VoxSAOR, a semantic-aware occupancy robustification framework that models aleatoric uncertainty through feature-space proxies. It first quantifies Local Semantic Consistency (LSC) to derive a voxel-wise structural stability prior and uses cross-view semantic consensus to filter unreliable semantic correspondences. Within each semantic category, spatially matched stable and ambiguous voxels form stable-to-ambiguous feature transitions, which are modeled by a class-conditional diagonal Gaussian distribution, capturing both feature displacement and dispersion. Guided by these proxy distributions, Aleatoric Regime Ensemble Learning (AREL) trains independently parameterized prediction heads using feature perturbations at progressively increasing scales, with perturbation strength modulated by structural reliability. At inference, proxy modeling and sampling are disabled, and all heads process the same observed features; their output probabilities are then averaged to produce a deterministic consensus prediction without requiring sampling. Experiments across multiple datasets and occupancy backbones demonstrate the effectiveness of VoxSAOR in improving semantic occupancy prediction, long-tail recognition, and predictive calibration. The source code will be made publicly available at https://github.com/MengfeiD/VoxSAOR.

cs.CV · cs.RO · eess.IV

新论文 · v1arXiv:2611.20505

How Reliable Is Visual Token Selection? Rethinking Pruning in Multimodal LLMs

Chaehee Song, Seokil Ham, KiHyun Seol, Jihyun Lee, Yubin Choi, Changick Kim

原文摘要

Visual token pruning improves the efficiency of multimodal large language models (MLLMs) by removing redundant visual tokens, yet the reliability of existing selection signals has not been systematically examined. We analyze token selection under semantics-preserving perturbations and across representative pruning criteria, revealing two limitations. First, attention-based importance shifts non-uniformly under perturbations, so magnitude alone cannot distinguish stable evidence from unstable saliency (local instability). Second, attention- and diversity-based methods select substantially different subsets despite comparable performance, showing that useful evidence is not confined to a unique subset (global inconsistency). These findings motivate reliable token-set construction. We instantiate this perspective in CoRePruner, a training-free framework combining instruction-aware diversity with reliability-aware attention. Across image benchmarks, multiple MLLM architectures, and token budgets, CoRePruner improves performance, especially under aggressive pruning, and remains competitive on video benchmarks. Our results establish reliability as a central consideration in visual token selection.

cs.CV

版本更新 · v2arXiv:2611.18464

Typing Matters: Muscle, Language and Typing Behaviour in Live Keystroke Decoding from sEMG

Abhilash Durgam, Nyle Siddiqui, Mubarak Shah, Qiushi Fu

原文摘要

Surface electromyography (sEMG) from the wrists can recover typed text without a keyboard. It is usually posed as a signal-inference problem: characters are read out of the muscle signal and character error rate is reported offline. We argue that this framing removes most of the problem. The decoder is constrained simultaneously by the muscles, by the language being typed, and by typing behaviour such as typos, corrections, backspaces and line breaks, and a usable system must predict keys live while the hand keeps moving. Existing systems fall short on two counts: (1) they process sEMG in fixed windows and carry nothing from one window to the next, discarding context accumulated over a session, and (2) they lack any understanding of the linguistic structure that drives typing, so every keystroke is inferred from the muscle signal alone. We propose a fully causal, zero-lookahead sEMG typing decoder built on a state space model (SSM), in which a single model decodes a whole session without windowed partitions, conditioned on what has already been typed. We achieve this by fusing a traditional typing decoder with an efficient autoregressive language model, instilling both sEMG and linguistic understanding in one model. To train such a model for per-character typing rather than whole-word text, we introduce a retokenization method that rewrites ordinary prose as human-like typing, with keystrokes, typos, backspaces and line breaks included, converting WikiText into a large typing corpus where none existed. Our model, having never seen a single sEMG-to-character transcript in typing, achieves a state-of-the-art character error rate of 25.13% on emg2qwerty's cross-user benchmark, a 5.4% relative improvement over the previous best with 6.7 times fewer parameters, while running strictly causally with no future context.

cs.AI

新论文 · v1arXiv:2611.20502

Pointillist Maize: 3D Reconstruction of Field-Grown Maize Using RGB UAV Imagery

Dewi Endah Kharismawati, Toni Kazic

原文摘要

Three-dimensional reconstruction can extend field phenotyping to individual maize organs, but thin, flexible, occluding structures and plant movement complicate reconstruction. We evaluated whether consumer-grade UAV RGB video processed with Structure from Motion and multi-view stereo could recover recognizable field-grown maize structures, and examined the effects of acquisition trajectories and frame sampling. COLMAP reconstructions preserved recognizable leaves, stems, tassel branches, ears, silks, and brace roots. For an equal-frame comparison, we reconstructed one plant from two sets of 295 frames from the same video: a single orbit sampled at 5 frames/second and three orbits at different altitudes (ribbon skirt) sampled at 2 frames/second. After ground removal, the reconstructions contained 520,601 and 375,318 points, respectively. The single orbit yielded greater point concentration within plant structures and fewer visible gaps, while the ribbon skirt yielded fewer scattered points and slightly lower proportions of isolated points. In densely planted nursery rows, visible voids occupied 17.57% of the rendered scene with one orbit at 1 frame/second, decreasing to 5.07% with additional serpentine passes and 2.68% when the original orbit was sampled at 3 frames/second. Additional observations also increased point dispersion associated with plant movement. A curlicue trajectory tested in a different field in 2024 produced distinguishable individual plants without visible duplicated structures, although substantial voids remained, mainly on the soil surface. These results demonstrate the feasibility of recovering recognizable maize organs from consumer UAV imagery and reveal trade-offs between acquisition strategies, reconstruction completeness, and point dispersion. Independent geometric validation is still needed before quantitative phenotyping.

cs.CV · cs.GR

版本更新 · v3arXiv:2604.14655

AgentGA: Evolving Code Solutions in Agent-Seed Space

David Y. Y. Tan, Kellie Chin, Jingxian Zhang

原文摘要

We present AgentGA, a framework that evolves autonomous code-generation runs by optimizing the agent seed: the task prompt plus optional parent archives that initialize a fresh workspace. The outer loop searches over reusable starting conditions rather than editing code directly. Each generation launches a fresh autonomous run in an isolated workspace, while selected parent archives provide inherited artifacts that descendants can inspect and reuse. AgentGA couples a population-level genetic algorithm with long-horizon agents, deterministic 1:1 elite tournaments, and online operator allocation through a modified Hedge controller. We evaluate the approach for tabular AutoML on the 16-competition Weco-Kaggle Lite benchmark, including model-matched AIDE and AutoGluon comparisons. AgentGA with DeepSeek V4 Flash (0731) achieves 71.13% mean Exceeds % of Human, compared with 57.61% for model-matched AIDE and 57.89% for AutoGluon, outranking AIDE on 14/16 tasks. A control that removes inheritance, in which every agent starts from scratch, reaches 65.76% under AgentGA's stopping rule and 67.25% at the full budget, above a single long-lived agent (61.20%) and below AgentGA. These descriptive results support agent-seed optimization as a practical design choice for autonomous code-search systems, without isolating the causal contribution of inheritance.

cs.AI · cs.LG

新论文 · v1arXiv:2611.20501

A Testing Method for Detector Error Models in Quantum Error Correction Compilers

Junjie Luo, Yasunari Suzuki

原文摘要

A quantum error correction (QEC) compiler produces, from a code and a noise model, both the physical circuit to execute and a detector error model (DEM). The DEM lists the probability of each error mechanism and the measurements it flips. Decoders consume the DEM without verification, so a defect that describes the noise incorrectly calibrates the decoder against wrong rates and silently degrades the logical error rate. Existing techniques for testing quantum software judge correctness by crashes or by statistics of sampled outputs. DEM defects that only distort rates lie below what sampling can resolve at the error rates where QEC operates. We present DEMVal, a testing method that validates each model the compiler emits against a reconstructed reference, with rates obtained from outside the compiler under test. It compares the two models exactly, so its sensitivity is independent of the physical error rate. The reference can be built in three ways that jointly cover the pipeline from source program to emitted model. Applied to NVIDIA's CUDA-Q QEC, Riverlane's Deltakit, and the deq component of Microsoft's QDK-EC, DEMVal found six defects, all confirmed by the developers and fixed. Sampling could have exposed only the four defects that drop, merge, or misplace mechanisms, and exposing all four requires millions of samples per experiment. Even analytically optimal moment tests miss the two rate-level defects, while exact model comparison detects all six with no samples. To our knowledge, DEMVal is the first systematic testing method for the error-model semantics of QEC compilers.

cs.SE

新论文 · v1arXiv:2611.20498

Replay-Augmented Regularisation for Class-Incremental Network Intrusion Detection: A Three-Dataset Evaluation of EWC and LwF

Mohammed Fackruddin Tusher, Muhammad Iqbal Hossain

原文摘要

As new attack categories emerge, network intrusion detection models need to be updated; however, when fine-tuned sequentially on new classes, previous classes are forgotten, resulting in catastrophic for- getting. The three public datasets we use are UNSW-NB15, ToN-IoT and CIC-IDS2017, and for these datasets, our class-incremental intrusion detection system is built using a CNN with one-dimensional (1D) archi- tecture trained over nine sequential tasks, each of which introduces one attack category. We evaluate sequential fine-tuning without any buffe compared to Elastic Weight Consolidation (EWC) and Learning without Forgetting (LwF), both of which include a class-size-adaptive rehearsal buffer of 20% of all previous classes (from 20 to 300 samples). The hyper-parameters were selected using UNSW-NB15 and fixed for the other two datasets. On UNSW-NB15, replay improves the final accuracy of both regularisers to 73.6% (EWC) and 71.1% (LwF) compared to about 32% (no improvement over fine-tuning). LwF with replay achieves 46.1% and 91.6% final accuracy for the two frozen-configuration datasets (ToN-IoT, CIC-IDS2017) while EWC with replay achieves 23.7% and 61.9%. In order to avoid this, EWC was therefore tuned on one dataset and the other two data sets were used as test sets. It is also demonstrated that the common forgetting measure can be misleading for a model which does not learn the old classes. As the fine-tuning baseline was not re- played and all the results were based on a single seed, it is not claimed hat the regularisers provide any extra support over replay alone.

cs.CR

版本更新 · v2arXiv:2512.03976

Adapting Large Language Models to Low-Resource Tibetan: A Two-Stage Continual and Supervised Fine-Tuning Study

Lifeng Chen, Chenyang Zhong, Ryan Lai, Tianming Liu

原文摘要

Adapting large language models (LLMs) to low-resource languages remains a major challenge due to data scarcity and cross-lingual drift. This work presents a two-stage adaptation of Qwen2.5-3B to Tibetan, a morphologically rich and underrepresented language. We employ Continual Pretraining (CPT) to establish Tibetan linguistic grounding, followed by Supervised Fine-Tuning (SFT) for task and translation specialization. Empirical evaluations demonstrate a consistent decrease in perplexity (from 2.98 $\rightarrow$ 1.54) and substantial improvements in Chinese$\rightarrow$Tibetan translation quality (BLEU: 0.046 $\rightarrow$ 0.261; chrF: 2.2 $\rightarrow$ 6.6). Layer-wise analysis across 435 layers in Qwen3-4B reveals that adaptation primarily concentrates on embedding and output heads, with mid--late MLP projections encoding domain-specific transformations. Our findings suggest that CPT constructs a Tibetan semantic manifold while SFT sharpens task alignment with minimal representational disruption. This study provides the first quantitative exploration of Tibetan adaptation dynamics for LLMs, and offers an open, reproducible framework for extending multilingual foundation models to low-resource settings.

cs.CL

版本更新 · v3arXiv:2607.28591

Change2Task: From Repository Changes to Executable Coding Agent Tasks and Environments

Haomin Qi, Xingliang Wang, Xuanqi Gao, Baihui Sang, Xin Zhang, Minghua Ma, Pengfei Gao, Yu Kang 等 12 位作者

原文摘要

Coding agents learn from and are measured on executable tasks, but building such tasks from repository history usually means rebuilding an environment at an old revision for every task, which is costly and brittle. We observe that a merged pull request (PR) can instead be undone on today's code: removing the effect of a bug fix or feature from a maintained revision recreates the same task inside an environment that already builds and runs. We present Change2Task, which converts merged PRs into verified tasks on an up-to-date revision of the same repository. It reconstructs the pre-change state by patch reversal, code mapping, or agent-based rewriting, and keeps a task only if its target tests pass, fail, and pass again across the healthy, task, and restored states. From 1,130 PRs spanning bug fixing, feature addition, test generation, API migration, and security repair, Change2Task builds 900 verified tasks, recovering 29% more than SWE-smith's PR Mirror on matched bug-fix candidates. The reconstructed tasks preserve the evaluation signal: in aggregate, four coding agents solve 41.1% of tasks on both the original and reconstructed versions, with 89.7% per-task agreement. Sharing modern environments across tasks further reduces setup time by 58% and storage by 71%. On 11 private industrial repositories, Change2Task yields 1,000 tasks on which the strongest model reaches only 14.6% Pass@3.

cs.SE · cs.CL · cs.LG

新论文 · v1arXiv:2611.20493

A Temporal Convolutional Network with Multi-Head Attention Model: A Surrogate for Parallel DEVS Semiconductor Fab Simulation

Vamsi Krishna Pendyala, Hessam S. Sarjoughian, Edward J. Yellig, Bharadhwaj Balasubramaniam

原文摘要

Semiconductor factories process many kinds of wafers organized into lots of various sizes and configurations. Different kinds of machines, each with its own operations, process wafers for hours to days, and the wafers undergo many incremental fabrication steps (with re-entry) and flows spanning days to weeks across connected machines. Fast and accurate predictions of throughput ($TH$) and turnaround time ($TA$) dynamics for individual machines and their combinations are essential for factory operations. To use machine learning methods for this purpose, this paper presents a Temporal Convolutional Network with Multi-head Attention (TCNA) architecture. It uses simulated datasets from a benchmark Parallel DEVS (PDEVS) semiconductor fabrication factory model. The factory processes configurable wafer lots with uniform and sinusoidal wafer arrival profiles and may undergo preventive maintenance. The wafer lot size, arrival profile, and factory maintenance policy are used as static covariates. TCNA is applied to a selected set of simulated discrete-event throughput and turnaround time trajectories. The $TH$ and $TA$ time series predicted by TCNA, evaluated with the common $MAPE$, $MSE$, and $R^2$ measures, show comparable or higher accuracy relative to other model architectures, with similar parameter counts and epoch times; the TCNA model size and execution time are similar to those of other architectures such as the Temporal Fusion Transformer. Although time series predictions are less accurate than high-resolution simulations, they can be deployed at a much lower computational cost. The TCNA achieves orders-of-magnitude speedup over PDEVS simulation with an acceptable loss of accuracy, enabling frequent product scheduling, capacity planning, and maintenance planning.

cs.LG

新论文 · v1arXiv:2611.20492

Weighted Sum-Rate Optimization of Near-Field STAR-RIS-Aided ISAC Systems Under Hardware Impairments

Huy T. Nguyen, Nhu-Thinh Bui Doan, Minh-Tuan Pham, Thai-Hoc Vu, Tien-Hoa Nguyen, Lajos Hanzo

原文摘要

This paper investigates weighted sum-rate (WSR) optimization for a simultaneously transmitting and reflecting reconfigurable intelligent surface (STAR-RIS)-assisted multi-user multiple-input multiple-output (MIMO) integrated sensing and communication (ISAC) system operating under the energy splitting (ES) protocol. A hybrid propagation model is considered, where the base station (BS)-STAR-RIS channel follows far-field propagation, whereas the STAR-RIS-user channels are modeled using near-field spherical wavefronts. Practical hardware impairments (HWIs), including STAR-RIS phase perturbations and signal-dependent additive distortion at the BS and user equipment (UEs), are incorporated into the system model. Under this framework, the WSR maximization problem is formulated subject to radar beampattern, quality-of-service, transmit-power, and STAR-RIS ES constraints. The resulting problem is highly non-convex due to the coupling among the active beamformers, passive STAR-RIS coefficients, and HWI-dependent signal-to interference-plus-noise ratio expressions. An alternating optimization (AO)-based successive convex approximation framework is developed to iteratively solve the active and passive beamforming subproblems. Numerical results demonstrate stable convergence of the proposed algorithm and consistent WSR improvement over direct-link transmission. Besides, severe HWIs also lead to WSR saturation in the high transmit-power regime, whereas stringent sensing constraints reduce communication throughput. Increasing the STAR-RIS aperture and the number of BS antennas improves the achievable WSR, although the performance gain becomes limited under strong hardware distortion.

cs.IT

版本更新 · v2arXiv:2610.05162

Memadapter: Counterfactual Adaptation Against Memory-induced Sycophancy

Ruqing Ning, Haibo Meng, Zhishang Xiang, Zerui Chen, Jinsong Su, Xin Wang, Qinggang Zhang

原文摘要

Long-term memory enables LLM-based agents to retain and reuse information across tasks and sessions, supporting personalization and long-horizon interactions. However, persistent memories can also induce sycophancy, causing agents to over-align with users' historical beliefs even when they are inaccurate, outdated, or inconsistent with objective evidence. Existing mitigation methods assume that memory-induced sycophancy originates from biased or incorrect memories and attempt to reduce this risk by filtering such memories at different stages of the memory pipeline. However, in the real world, objective and correct memories can still induce sycophancy, and the same memory can warrant different influence across different contexts. To this end, we propose MemAdapter, a novel framework that adaptively integrates retrieved memories to support objective and reliable reasoning. Specifically, MemAdapter consists of three components: (i) Counterfactual Induction, which leverages counterfactual reasoning to uncover the potential risk of retrieved memories; (ii) Context-Aware Reflection, which calibrates the inferential influence of each retrieved memory in light of the current task via self-reflection; and (iii) Evidence-Based Reasoning, which grounds the final response in appropriate evidence while preserving the legitimate influence of memory. Extensive experiments on three benchmarks demonstrate that MemAdapter consistently improves memory reliability across diverse scenarios. Our code is available at https://github.com/DEEP-JLU/MemAdapter.

cs.AI

版本更新 · v4arXiv:2610.21390

PriorityFEC: Deadline-Aware Streaming FEC for KV-Cache Handoff in Edge Multi-Agent Systems

Han Shen, Qifu Tyler Sun, Fanyang Meng, Ye Wang, Yongsheng Liang

原文摘要

Collaborative edge large language model (LLM) systems often require repeated prefill computation when task contexts are transferred between agents. Reusing previously computed key--value (KV) caches can eliminate this overhead, but requires KV-cache delivery before the receiving agent begins decoding. Under strict deadlines and unreliable edge links, complete KV cache transfer may be infeasible, making the timely delivery of useful KV information critical. This paper proposes PriorityFEC, a utility-aware KV-cache handoff scheme that combines prioritized source transmission with adaptive loss recovery. PriorityFEC first organizes KV blocks into a utility-aware source queue to prioritize important information. It then employs ACK-sliding Streaming FEC and a lightweight Double Deep Q-Network (Double-DQN) scheduler to dynamically balance source delivery and repair transmission under deadline constraints. We implement PriorityFEC in a distributed edge LLM-serving prototype and evaluate it using Qwen2.5-7B-Instruct and Llama-3.1-8B-Instruct across multiple workloads. Experimental results demonstrate that PriorityFEC preserves more useful KV information and achieves substantially higher task accuracy than competing transport schemes, particularly when the deadline is insufficient to deliver the complete KV cache.

cs.NI · cs.AI

版本更新 · v3arXiv:2607.26643

SkillBoost: A Constrained Exploration-Exploitation Process for Mitigating Skill Overfitting

Hongqiang Lin, Chao Liu, Xiaofan Bai, Xuan Jin, Yuhong Li, Nenggan Zheng, Xipeng Cao

原文摘要

Enabling large language model (LLM) agents to accumulate and reuse experience from past interactions remains a central challenge in real-world applications. A promising solution is to treat skills as trainable states and optimize them in the same way as model parameters in neural network training. However, data-driven skill optimization is prone to overfitting to the limited trajectories collected from real environments. Overexploiting these trajectories overfits the current batch, while unconstrained exploration causes regression on previously solved cases. This tension motivates a constrained search view of skill self-evolution, governed by an exploration--exploitation trade-off. We propose SkillBoost, a three-stage framework that mitigates both risks: structured exploitation localizes observed failures to editable skill components, prior-guided exploration draws on prior knowledge in the LLM to generate diverse repair candidates, and verified acceptance commits a candidate only when it improves performance within a regression bound. Experiments across 23 model--benchmark configurations show that SkillBoost achieves state-of-the-art performance while mitigating overfitting, outperforming both human-crafted and LLM-generated skills. Transfer experiments further show that optimized skills can be reused by other agents on similar tasks. The code is available at https://github.com/HQ-Lin/SkillBoost

cs.AI · cs.LG

版本更新 · v2arXiv:2611.16223

We Are What We Remember: Mental-State-Driven Memory for Long-Term Character Dialogue

Zhicheng Jia, Shuchen Wu, Zhuoshi Tong, Haisen Liao, Sicheng Zhou, Chao Yang

原文摘要

Shared history in human dialogue is not merely a repository of facts, but the foundation of evolving relationships and mutual understanding. However, mainstream conversational memory systems rely on passive semantic similarity over external fact stores, frequently retrieving topically similar records while missing relationally pivotal experiences. Inspired by cognitive psychology's Self-Memory System (SMS), we propose MindMem, a mental-state-driven memory framework that asynchronously consolidates shared experiences into source-linked mental states of the character, the user, and their relationship. At each turn, a specialized mental-state reasoner (Mind) contextualizes these persistent states to anticipate latent user needs, form proactive intentions, and recover supporting episodic evidence that surface-level matching overlooks. To evaluate history-sensitive adaptation beyond factual recall, we introduce MemReal-Bench, a longitudinal benchmark grounded in month-long real-user interactions across dyadic and group scenes. By pairing Direct Question Answering (DQA) with Situated Memory Realization (SMR), it tests whether agents bridge the gap between knowing past facts and naturally using shared history, world rules, and relational trajectories. On MemReal-Bench, MindMem reaches 59.03% SMR and 82.58% DQA, exceeding the strongest memory baseline by 24.19 and 9.19 percentage points, respectively, and outperforming oracle target-memory injection on SMR. Code and dataset are available at https://github.com/jiazhicheng7-ctrl/MindMem.

cs.CL

新论文 · v1arXiv:2611.20485

PromptPoseFX: Authoring Pose-Driven Visual Effects through Augmented Prompting and Interactive Refinement

Kexin Zheng, Hui Ye, Hongbo Fu

原文摘要

Human pose-driven visual effects offer a compelling way for enhancing video expressiveness and storytelling. Authoring such effects with traditional video editing tools is a complex and time-consuming task that requires considerable technical expertise and manual effort. While generative AI models can simplify the workflow by reducing manual scripting through language-driven generation, existing approaches suffer from imprecise alignment between visual effects and pose data, failure to preserve original video content, and limited support for fine-grained, iterative refinement. To address these challenges, we present PromptPoseFX, an interactive system that allows creators to author pose-driven visual effects through augmented language prompts and interactive refinement. Our system leverages Large Language Models (LLMs) to translate high-level user intent into parameterized procedural graphics scripts, automatically mapping pose-tracking data to specific effect parameters, allowing precise alignment with human motion. This script-based approach manages effects as independent layers, thereby preserving the original content of the underlying video. To reduce language ambiguity, our interface offers controls across spatial, temporal, and parameter domains, enabling the precise, intuitive, and iterative authoring of pose-driven visual effects. We evaluated PromptPoseFX through a comparative study against a text-only LLM interface, followed by an exploratory study. In the comparative study, participants rated PromptPoseFX significantly higher than the text-only baseline on measures of perceived control and precision. The exploratory study examined usability and creativity support in open-ended authoring tasks.

cs.HC

新论文 · v1arXiv:2611.20484

Agent-Guided Configuration Tuning for SMT Solvers

Cong Wang, Qiaoyi Zheng, Zhaoguo Wang, Fan Long

原文摘要

SMT solvers expose many configuration choices that determine how well their existing algorithms perform on a workload. We present, to our knowledge, the first workflow that uses general-purpose LLM agents to tune SMT solver configurations. Agents read solver documentation, source code, and persistent experiment records to propose option changes, and a deterministic harness measures each proposal, validates the returned models, and selects changes separately for each competition Division. Over roughly nine weeks, our tuning campaign recorded 4,545 proposals for cvc5 across 13 Divisions and 36 logics. We evaluate the tuned configurations on the 2026 and 2025 SMT-COMP ModelValidation benchmarks under a 24-second cutoff. On problems not used for training, they validate more problems than official cvc5 in ten of the 13 Divisions in 2026 and in eight in 2025, with geometric-mean gains of 7.8% and 6.8% across Divisions. On the complete official selections, which include training problems, the tuned floating-point counts exceed those of every individual competition participant in both years. Further studies show how configuration changes make existing algorithms effective by changing access to strategies, scheduling, and option compatibility, and how benchmark families trade off against one another. The agents' experiment traces contain hypothesis-based proposals in every tuning group, and 64.3% of the extracted hypotheses draw on prior measurements. These results show that agent-guided tuning can specialize a mature SMT solver and that accumulated experimental evidence informs sustained search.

cs.AI

新论文 · v1arXiv:2611.20483

Parallelizable Interior-Point ADMM for Block-Structured Nonconvex Programs

Zherong Pan

原文摘要

Physics simulation is an increasingly important computational primitive for AI, and local--global/ADMM splitting exposes its parallel structure but lacks matching finite-time convergence guarantees. We close this gap with two ADMM variants, Interior-Point ADMM (IP-ADMM), for block-structured nonconvex composite programs with smooth nonlinear inequality and affine constraints. A one-sided finite-activation wall keeps each constraint split state strictly feasible while the original-space infeasibility vanishes, giving closed-form constraint updates, one proximal call per objective block, and one equality-constrained quadratic consensus solve per iteration. \rev{To our knowledge, this makes IP-ADMM the first single-loop method that updates nonconvex inequality constraints locally and in parallel with finite-time convergence guarantees.} Ordinary IP-ADMM attains a CQ-free best-iterate weak-KKT rate of $O(K^{-1/3})$ or a banded Fritz--John rate of $O(K^{-1/4})$, which converts to KKT at $O(K^{-1/4})$ under a uniform constraint qualification; these are nearby-point certificates that reduce to conventional KKT for smooth blocks. A checkpoint variant, IP-ADMM-Ckpt, improves the asymptotic KKT exponent to $O(K^{-1/3})$. We also show how the bounds scale with the objective- and constraint-block counts and the coupling norm.

math.OC · cs.LG

新论文 · v1arXiv:2611.20478

Beyond Seeing: Fine-Grained Preference Validation and Severity-Aware Alignment for Mitigating Visual Hallucinations in Vision-Language Models

Abdulaziz Alqahtani, Alwaleed Alharthi, Muzammil Behzad

原文摘要

Large vision-language models can produce fluent responses containing objects, attributes, or relations unsupported by the image. We study an implemented alignment pipeline that normalizes released fine-grained hallucination supervision, validates chosen-rejected preference pairs with vision-language judges, repairs rejected chosen responses, and trains LLaVA adapters with standard or severity-aware direct preference optimization. The study uses 16,143 normalized feedback records and 8,386 released preference pairs, comparing direct training with zero-shot and two-shot judged severity-margin variants. The zero-shot variant reduces Object HalBench CHAIRs from 53.00 to 36.00 and CHAIRi from 15.72 to 10.41. Across aligned models, AMBER-Hal decreases from 35.90 to 25.10-26.90, accompanied by reduced visual coverage. In contrast, the base model retains the highest valid POPE adversarial F1, 84.18 versus 83.87-83.91 after alignment. These results support benchmark-dependent reductions in generative hallucination rather than uniform superiority of severity-aware alignment. We distinguish implemented pipeline components from the preference paths used in the reported experiments and discuss the limits of attributing gains to judging, repair, or the training objective individually.

cs.CV · cs.AI

新论文 · v1arXiv:2611.20476

Improving X-Ray Baggage Instance Segmentation with Shared Point-Based Mask Refinement

Abdulaziz Alqahtani, Qasim Umer

原文摘要

Overlapping projections and complex object contours make prohibited items difficult to delineate in X-ray baggage images. This study investigates shared point-based mask refinement to improve instance segmentation within Cascade Mask R-CNN. The framework combines a ResNet-101 backbone with an attention-based feature pyramid and progressively updates uncertain mask locations, increasing mask resolution from $28 \times 28$ to $112 \times 112$. A single refinement module is shared across cascade stages and refinement levels, with transformer and multilayer perceptron (MLP) variants used to examine the role of interactions among selected points. Ten configurations were evaluated on the official PIXray split across three training seeds using the original target-instance annotations for evaluation. The transformer and MLP refiners achieved mean mask average precision (AP) of 71.79 and 72.47, respectively, compared with 70.50 for the unrefined detector, yielding gains of 1.29 and 1.97 AP points. The MLP achieved the highest mean mask AP among the evaluated configurations. Both variants also improved boundary agreement on commonly detected objects, supporting improved mask delineation, while detection diagnostics did not establish a corresponding positive gain. These findings demonstrate that shared point-based refinement can improve segmentation quality in the tested cascade architecture using existing target-mask annotations, without additional occluder supervision. Evaluation on independent X-ray datasets is needed to determine how well these improvements generalize.

cs.CV · cs.AI

新论文 · v1arXiv:2611.20475

Rounded Potentials Certify Exact Maximum Flow

Ohad Trabelsi

原文摘要

The textbook certificate for maximum flow is lopsided: the cut takes one bit per vertex, while the flow takes one number per edge. Indeed, some graphs have $Θ(n^2)$ terminal pairs for which every maximum flow uses $Θ(n^2)$ edges. Yet we show that one short number per vertex suffices for both cut and flow: $n$ vertex potentials, each rounded to $O(\log(nW))$ bits, certify the exact maximum $st$-flow value in any directed graph with integer capacities at most $W$. Exact potentials can have exponentially large representations. Rounding them may violate flow conservation, but the total error across any cut is less than one unit, so integrality of cut capacities absorbs it. Checking conservation is local and linear in the capacities, allowing the same certificate to yield new bounds in four models (capacities are polynomially bounded unless stated otherwise): - Online annotated streaming. $\widetilde{O}(n)$ total cost (help plus space) for the maximum $st$-flow value, optimal up to logarithmic factors, answering a question of Thaler (ICALP 2016) on whether $o(n^2)$ is possible. - Broadcast proof-labeling. $O(\log(n+p))$-bit labels certifying a maximum $st$-flow value of $p$ in directed and undirected graphs with arbitrary capacities, optimal for $p\ge n^{Ω(1)}$, settling that regime of a question of Korman, Kutten, and Peleg (Distributed Computing 2010). - Merlin-Arthur verification. $\widetilde{O}(n^3)$ time and $O(n^3\log n)$ proof length for directed All-Pairs Max-Flow, improving on previous verification times of $n^{4+o(1)}$ for general capacities and $n^{3.5+o(1)}$ for unit capacities, and answering a question of Akmal (TheoretiCS 2024). - Nondeterministic algorithms. $O(n^{3.614})$ time for directed All-Pairs Max-Flow, or $n^{3.5+o(1)}$ if $ω=2$: the first subquartic bound for general edge capacities, answering a question of Trabelsi (SODA 2025).

cs.DS

版本更新 · v4arXiv:2307.07735

Faster Algorithms for Structured Linear and Kernel Support Vector Machines

Yuzhou Gu, Zhao Song, Lichen Zhang

原文摘要

Many machine learning problems are quadratic programs, including Support Vector Machines (SVMs). A general quadratic program has input size $Θ(n^2)$, and assuming the Strong Exponential Time Hypothesis (SETH), no $O(n^{2-o(1)})$ time algorithm exists when the objective matrix is positive semidefinite (Backurs, Indyk, and Schmidt, NeurIPS'17). SVMs, however, are given by $n$ data points of dimension $d$, often much smaller than $n$, and have only $O(1)$ equality constraints. In this work, we design the first nearly-linear time algorithm for solving quadratic programs whenever a low-rank factorization of the quadratic objective matrix is given, the number of linear constraints is small, the coefficients of the objective and some optimal solution are bounded by $\mathrm{poly}(n)$, and some feasible point has all entries between $1/\mathrm{poly}(n)$ and $\mathrm{poly}(n)$. For linear SVM on $d$-dimensional data, our randomized algorithm runs in expected time $\widetilde O(nd^{(ω+1)/2}\log(n/ε))$ when $(d+1)^{ω-1}\le n$ and the data points have norms at most $\mathrm{poly}(n)$ and margin at least $1/\mathrm{poly}(n)$, where $ω<2.258$ is the matrix multiplication exponent. Its output satisfies the equality constraint only approximately. For Gaussian kernel SVM, when $d=O(\log n)$, the squared dataset radius is $o(\log n/\log\log n)$ and the dual variables lie in a box $[0,R]^n$ with $R\le\mathrm{poly}(n)$, our randomized algorithm runs in expected time $n^{1+o(1)}$ for any accuracy $ε\ge 1/\mathrm{poly}(n)$. We also prove that, assuming SETH, when the squared dataset radius is $Ω(\log^2 n)$, $Ω(n^{2-o(1)})$ time is required on some inputs to reach error $\exp(-Θ(\log^2 n))$ for the program without the bias term. This improves upon the prior best lower bound in both the dimension $d$ and the squared dataset radius.

math.OC · cs.LG · stat.ML

新论文 · v1arXiv:2611.20473

Online and Fully-Dynamic Sorting

Yossi Azar, Sujoy Bhore, Devdan Dey, Debmalya Panigrahi, Or Vardi, Ian Zhang

原文摘要

Online sorting asks how to place $n$ sequentially arriving elements in an array to minimize the sum of absolute differences between consecutive elements. Fully-dynamic sorting further extends the problem to allow both element insertions and deletions. Without recourse, these problems face strong lower bounds on the competitive ratio: $Ω(\sqrt{n})$ for online sorting and $Ω(n)$ for fully-dynamic sorting. In sharp contrast, we show that allowing only poly-logarithmic recourse yields at least exponential improvements in the competitive ratio for both problems:(1) Our main algorithmic result is a deterministic fully-dynamic sorting algorithm that is $O(\log^2 n)$-competitive and has $O(\log^4 n)$ amortized recourse per update. We complement this with a lower bound showing that any deterministic algorithm with any $O(\mathrm{polylog}(n))$ competitive ratio has amortized recourse at least $Ω(\sqrt{\log n})$.(2) For the insertion-only setting, we establish a family of approximation--recourse tradeoffs, including $O(\log\log n)$ competitive ratio with $O(\log\log n)$ amortized recourse. Given the lower bound in the fully-dynamic case, this result shows that the insertion-only problem yields exponentially better bounds than the fully-dynamic one.(3) Finally, we show that the stronger approximation-recourse tradeoff for the insertion-only case extends to the fully-dynamic setting when the array has $(1+\varepsilon)n$ space.

cs.DS

版本更新 · v3arXiv:2603.12344

Towards Generalist Models for Molecular Property Prediction: Distilling Verbalized Knowledge from Decision Trees

Khiem Le, Marcos Martínez Galindo, Vanessa Lopez, Sreejata Dey, Ting Hua, Nitesh V. Chawla, Hoang Thanh Lam

原文摘要

Molecular Property Prediction (MPP) is a fundamental problem in drug discovery that has recently attracted growing attention. Large Language Models (LLMs), known for their impressive proficiency across domains, show promise as generalist models for MPP. However, their current performance remains below the threshold needed for practical adoption. To bridge this gap, we propose TreeKD for distilling the knowledge of tree-based specialist models into LLMs to complement the internal knowledge of LLMs and improve their predictive accuracy. For each property, we train a specialist decision tree using features derived from 40K functional groups in the input molecules. Then, the predictive rule learned by the decision tree, which encodes its knowledge, is verbalized and incorporated into the prompts for training LLMs. In addition, by replacing a single decision tree with a Random Forest, we introduce a test-time scaling technique called rule-consistency, which aggregates predictions generated from different prompts constructed with different rules. An extensive evaluation with two LLMs, Gemma-2-2B and Granite-3.3-2B, on the TDC benchmark with 22 prediction tasks shows that our method substantially enhances the performance of LLMs, advancing the development of generalist models for MPP.

cs.LG

新论文 · v1arXiv:2611.20468

Tabular Foundation Models as Emulators for Forecasting Terrestrial Ecosystem Dynamics

Zhihao Wang, Ruohan Li, Lei Ma, George Hurtt, Xiaowei Jia, Yiqun Xie

原文摘要

Process-based terrestrial biosphere models such as the Ecosystem Demography (ED) model underpin global carbon budgets, but their cost limits global, high-resolution and scenario-rich simulation. Deep emulators such as DeepED reduce that cost but require task-specific architectures and training. Tabular foundation models offer a different route: TabPFN predicts new rows by in-context learning, without being trained on the task. We compare TabPFN-3.5 with two DeepED baselines, $DeepED_{LSTM}$ and $DeepED_{Transformer}$, on CarbonGlobe's overall evaluation, which trains on a uniform 1/16 sample of sites and tests on the other 50,779, predicting the 40-year state of all seven ED outputs. We then test every model, trained on present-day data only, on ED runs forced by SSP1-2.6 and SSP5-8.5 for 2021--2060 in six continental forest regions. With no training, TabPFN-3.5 has the lowest present-day height error of the three models and is within 4% of $DeepED_{LSTM}$ for soil carbon, but is less accurate for AGB, LAI and the carbon fluxes. In 2060, errors are several times larger than at the same sites under present-day forcing, and every model underestimates the biomass that ED gains. Under SSP5-8.5, TabPFN-3.5 is the most accurate model for height, AGB, soil carbon and LAI, with the lowest AGB error in all six regions, although, unlike ED, it predicts no larger biomass gain than under SSP1-2.6. Under SSP1-2.6, no model is best overall.

cs.LG · cs.AI

新论文 · v1arXiv:2611.20467

On the Applicability of $p$-adic Numbers in Computational Modules for Neural Networks

Vladimir Evgrafov

原文摘要

The applicability of $p$-adic arithmetic as an intermediate computational module in an otherwise unchanged machine-learning pipeline is investigated. A one-feature linear regression on the scikit-learn Diabetes dataset serves as the test model. The fitted coefficients are represented in $\mathbb{Q}_p$, predictions are computed in that field, and rational reconstruction is applied to the outputs. In the pyAdic experiments, rational reconstruction either returns no result or produces values that differ from the baseline predictions; changing computational precision can change both the magnitude and sign of the reconstructed value. In SageMath, forward conversions succeed, but the reconstructed predictions differ from the baseline. An alternative representation based on digit reversal fails to preserve ordinary addition because of carry propagation. Under the tested settings, none of the test predictions is reconstructed correctly. These results show that the tested conversion and reconstruction procedures do not provide a compatible intermediate module even for linear regression. Models defined directly over $p$-adic data and operations offer a direction for further investigation.

cs.LG

新论文 · v1arXiv:2611.20465

ReRead: Wrong Predictions, Useful Representations - Correcting Frozen Robot Policies by Re-reading Their Own Representations

Xu Su, Shunlong Xiao, Qintao Yi, Mingkuan Zhao, Xinmiao Yu, Yizhe Li, Zihan Wang, Bokui Chen

原文摘要

When a pretrained robot policy fails on challenging tasks, common remedies include retraining the backbone, adding an external visual encoder, or introducing a memory module. We argue that wrong predictions do not imply wrong representations: a policy can retain the required information yet fail to read it out as the correct action. To test this view, we propose ReRead: with the backbone frozen, a corrector re-reads the current and two previous caches to predict an action residual. The corrector sees no images, is trained solely on offline demonstrations from all 50 RoboTwin 2.0 tasks, and adds only 4.17% latency. Ablations and interventions show that the correction relies on the backbone's own representations: none of four external visual encoders adds further gain, and swapping in past caches from other trajectories sharply increases action error. On the 11 hardest RoboTwin 2.0 tasks, ReRead improves the success rates of LingBot-VLA, $π_{0.5}$, and Fast-WAM by 23.45, 13.09, and 8.00 percentage points, respectively. In paired evaluation on all 50 tasks, it raises Fast-WAM's average success rate from 90.73% to 92.66%. On a Flexiv dual-arm robot, it raises success on a phase-ambiguous task from 0% to 36%.

cs.RO · cs.AI