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

+1112新增论文
+875其中 AI 相关
+133已有论文修订
+97新增 AI 导读

arXiv 最近完成扫描:2026-11-20 03:37 · 全站最近入库:2026-11-19 19:09各信源进度 →

统计口径与数据新鲜度

统计 2026-11-19 03:47 — 2026-11-20 03:47(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:2026-11-20 03:37最近发现新增版本:2026-11-19 19:09库内论文最新更新时间:2026-11-19 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。

27,982 已收录论文23,249 AI 相关8,559 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 16 篇论文 · 按版本更新时间排序上轮扫描起点 2026-11-20 03:35(北京时间)
新论文 · v1arXiv:2611.24002

ProtoBridge-CLIP: Observation-Masked Finding Prototypes for Medical Image--Report Representation Learning

Jinrui Wu, Yun Peng

原文摘要

Radiology reports distinguish affirmed findings, explicit negations, and unmentioned concepts, but conventional global image--report contrastive learning often compresses these different evidence states into a single study-level target. We introduce \model{}, which combines symmetric global image--report alignment with observation-masked finding supervision and a soft relation objective over affirmed report concepts. Text-derived finding prototypes are aligned with local visual tokens, while uncertain and unmentioned concepts are excluded from direct finding labels. We evaluate the framework on three CT benchmarks and five chest-radiograph benchmarks, together with component ablations, paired and clinical retrieval, localization, perturbation robustness, calibration, and per-finding analyses. \model{} achieves macro-AUROC values of 73.6%, 66.5\%, and 68.8% on CT-RATE, RAD-ChestCT, and INSPECT, and 80.2%, 91.4%, 78.5%, 86.5%, and 90.8% on ChestXray14, CheXpert, PadChest, RSNA, and SIIM-ACR, respectively. Relative to the matched global-only controls, the findings and relation objectives yield consistent gains across the evaluated datasets. These results demonstrate the practical value of treating observed and unobserved report concepts differently during medical image--text representation learning.

cs.CV · cs.AI

新论文 · v1arXiv:2611.23999

ReliFlow: Reliability-Calibrated Residual Transport for Combinatorial Single-Cell Perturbation Prediction

Jinrui David, Yun Peng

原文摘要

Predicting the transcriptional response to combined perturbations requires distinguishing interactions between interventions from errors in their individual effect estimates. The latter become more pronounced when single-perturbation evidence is sparse or withheld. We introduce a conditional flow-matching model that estimates the reliability of single-gene effects from training-only cross-fitting and uses those estimates to calibrate the additive source distribution. The velocity is decomposed into a known correction of the additive reference and a learned residual transport field. On Norman genetic perturbations, reduces mean absolute error (MAE) from 0.142 to 0.131 for unseen combinations of observed singles and from 0.198 to 0.178 when constituent singles are withheld. On held-out ComboSciPlex drug combinations, MAE decreases from 0.191 to 0.176. Non-additive residual correlation improves from 0.574 to 0.638, and component ablations identify the contribution of reliability weighting. The results indicate that accounting for uncertain constituent effects improves generalization to unseen genetic and chemical combinations, with a moderate increase in training cost.

stat.ML · cs.AI · cs.LG

版本更新 · v3arXiv:2606.03303

LEAP:用智能体框架为形式数学增强大语言模型

LEAP: Supercharging LLMs for Formal Mathematics with Agentic Frameworks

Po-Nien Kung, Linfeng Song, Dawsen Hwang, Jinsung Yoon, Chun-Liang Li, Simone Severini, Mirek Olšák, Edward Lockhart 等 13 位作者

AI 速读 基于摘要

LEAP 是一个智能体框架,让通用基础模型通过与 Lean 编译器持续交互,把非形式化推理转化为可机械验证的形式化证明,并在 2025 年 Putnam 竞赛与自建 Lean-IMO-Bench 上取得作者报告的高求解率。

cs.AI

查看原文摘要

Large Language Models (LLMs) exhibit strong informal mathematical reasoning but struggle to generate mechanically verifiable proofs in formal languages like Lean. We present LEAP, an agentic framework that enables general-purpose foundation models to achieve state-of-the-art performance on automated formal theorem proving. LEAP leverages foundation model capabilities, such as informal reasoning, instruction following, and iterative self-refinement. By decomposing complex problems into smaller units, the system bridges formal proof construction with informal blueprints through continuous interaction with the Lean compiler. To provide a rigorous evaluation beyond increasingly saturated benchmarks, we introduce Lean-IMO-Bench, a benchmark of IMO-style problems formalized in Lean, with short statements yet highly non-routine and multi-step proofs across a wide range of difficulty levels. Empirically, on the latest 2025 Putnam Competition, an annual mathematics competition for undergraduate students in North America, LEAP solves all 12 problems, matching recent breakthroughs by frontier formal mathematical models. On Lean-IMO-Bench, LEAP boosts the one-shot formal solve rate of general-purpose LLMs from below 10% to 70%, notably surpassing the 48% benchmark set by a specialized, gold-medal-caliber IMO system. Furthermore, we demonstrate LEAP's research-level utility by autonomously formalizing complex proofs for open combinatorial challenges, including a verified proof for a key subproblem in Knuth's Hamiltonian decomposition of even-order Cayley graphs.

版本更新 · v2arXiv:2610.02945

面向AI辅助数学发现的持续图记忆与社区基础设施

Continual Graph Memory and Community Infrastructure for AI-Assisted Mathematical Discovery

Eric Hanchen Jiang, Junyi Zhang, Jinxi Yu, Jiachen Lu, Zhi Zhang, Michael Simon, Justin Baker, Richard Cai 等 30 位作者

AI 速读 基于摘要

作者提出Ansatz研究框架与Ansatze社区平台,用持续图记忆区分模型接受的结论与人工整理探索,并支持数学研究的检查、引导与形式化验证。

cs.AI · cs.CL

查看原文摘要

Mathematical discovery requires preserving partial ideas and the evidence supporting their reuse. We present Ansatz, a research harness whose continual graph memory separates model-accepted claims and dependencies from curated exploration, and Ansatze, a community platform for inspecting, guiding and checking this work. On First Proof, Ansatz recorded nine local verifier closures in each of two ten-problem batches; a partial result in the Second Batch also counted as a pass under the study's criterion. These model-assessed outcomes include continued searches and differ from expert validation. Component studies and re-review expose disagreements and their dependency reach. Open-problem investigations produced candidate results, with five Lean formalizations reported. The platform connects problem discovery, research records, manuscript analysis and formal checking while preserving distinct evidence states. Together, these systems support cumulative human-AI research through persistent investigation and inspectable contributions.

新论文 · v1arXiv:2611.09061

Does Reasoning Improve Psychological Depth in Large Language Models?

Ruichen Zheng, Yihe Wang, Fabrice Harel-Canada, Sara Khosravi, Zeynep Senahan Yildiz, Amit Sahai, Nanyun Peng

原文摘要

Benchmark gains from reasoning-enhanced large language models encourage the assumption that these models should broadly perform better on complex generative tasks, but this remains underexplored in open-ended, creative domains shaped by nuances of reader experience. We tested this assumption through the lens of psychological depth in short stories. We compared DeepSeek-R1 vs. DeepSeek-V3, our closest available contrast for reasoning-oriented post-training, and GPT-5 vs. GPT-4o, through 60 blinded pairwise evaluations in which seven human annotators and an ensemble LLM-as-a-Judge rated prompt-matched story outputs. Human evaluations showed no universal reasoning advantage: GPT-5 was slightly preferred over GPT-4o (60.0-62.9% across two reasoning-effort settings), whereas DeepSeek-R1 fell behind DeepSeek-V3 (42.9%). Although this pattern was marked by low inter-annotator agreement (Krippendorff's $α= 0.070$), our annotation-quality checks suggested genuine evaluative subjectivity, plausibly reflecting distinct annotator decision schemes rather than annotation failure. In contrast, the LLM-as-a-Judge showed a strong pro-reasoning bias (98.3% of pairs vs. 55.2% of human votes), consistent across its five configurations and partly tracking sentence length. These findings suggest that, at the frontier, reasoning-related quality differences are small enough to be dominated by reader subjectivity, and that LLM-as-a-Judge evaluators can obscure this subjectivity in nuanced creative-generation tasks.

cs.CL

版本更新 · v4arXiv:2607.07779

From Solvers to Research: Large Language Model-Driven Formal Mathematics at the Research Frontier

Eric Jiang, Xiao Liang, Junyi Zhang, Justin Baker, Zhi Zhang, Yingjia Wan, Yubei Li, Jinxi Yu 等 41 位作者

原文摘要

Large language models have made rapid progress in generating machine-checked proofs for well-defined mathematical problems. Extending this progress to research mathematics requires more than solving isolated theorems. Research involves choosing questions, developing definitions and conjectures, connecting results, and combining formal proofs with expert judgment. This survey examines how LLM-driven systems can support these broader activities. We review formal-mathematics tasks, datasets, and autoformalization, and organize proof-generation methods into three overlapping areas: training strategies, inference-time search and adaptation, and agentic workflows. We then examine benchmark results, selected research projects, and platforms for collaborative mathematical work. A central distinction runs throughout the survey: checking a formal proof does not by itself establish that it captures the intended problem or makes a new mathematical contribution. We therefore distinguish theorem benchmarks, repository tasks, formalization projects, and research claims, and discuss the evidence needed to assess each. Finally, we identify five priorities for progress: better data and evaluation, representations of mathematical relationships, conjecture and concept generation, reliable tool integration, and human--AI collaboration. Together, these perspectives connect advances in proof generation to the practical demands of mathematical research.

cs.CL · cs.AI

新论文 · v1arXiv:2611.05919

Routes of Reasoning: Reading and Reshaping Test-Time Compute in MoE Models

Shengjie Wang, Mohsen Fayyaz, Nanyun Peng

原文摘要

Reasoning models allocate test-time compute across distinct behaviors such as planning, deriving, and verifying. However, models frequently "overthink," continuing to verify and re-derive long after finding a correct answer. This wastes expensive compute and can even degrade performance. Existing interventions, such as token budgets or steering, primarily regulate reasoning duration rather than targeting specific behaviors. In this work, we demonstrate that Mixture-of-Experts (MoE) routers provide a natural, training-free interface for behavioral control. We first split reasoning traces into sentence-level segments and label each with one of six reasoning behaviors (planning, execution, restatement, fact, reflection, and conclusion), so that every token carries the behavior of its segment. Linear probes on router logits predict this label nearly as well as probes on the residual stream, the model's full hidden state and the standard target for probing, while using about half as many features. These signals generalize: experts associated with a reasoning behavior on one dataset predict the same behavior on the others, across six datasets spanning five domains (math, code, science, knowledge, and commonsense). Building on this, we selectively edit expert router scores during inference to curb overthinking. Across four models and four benchmarks, this intervention reduces the number of generated tokens by a median of 30.4% (and up to 63.3%). Crucially, it encourages the model to stop once the problem is solved, rather than skipping essential reasoning steps. Accuracy never drops by more than 0.5 percentage points, and frequently improves. Ultimately, MoE routing offers both an interpretable lens into reasoning states and a lightweight mechanism for optimizing test-time compute.

cs.AI · cs.CL · cs.LG

新论文 · v1arXiv:2611.03197

Finding the Pivots to Repair Broken Source-Test Evolution Using LLMs

Xiao Chen, Songqiang Chen, Jialun Cao, Yun Peng, Chun Yong Chong, Shing-Chi Cheung

原文摘要

Repairing obsolete tests in response to source code evolution is critical for test suite reliability and software quality assurance. While large language models (LLMs) have emerged as promising tools for automating many software development tasks, the effective supply of context and the orchestration of repair pipelines in this source--test co-evolution scenario remain poorly understood. Moreover, although reasoning is found helpful for LLMs on some complex coding tasks, it remains unclear which reasoning steps actually contribute to effective test repairs and how the insights learned from LLMs' behaviors benefit coding agents. In this paper, we present a systematic empirical study of test repair for source-test co-evolution empowered by LLMs, investigating the effective setups and typical LLM behaviors. To support this study, we construct a new benchmark of 500 real-world source-test co-evolution cases from commits submitted between May 2024 and April 2026 across 15 highly-starred open-source Java projects. Our experimental results reveal several key insights for effective LLM-based test repair: the complementary benefits of combining test failure symptoms, source code changes, and supportive code context, as well as the effectiveness of self-repair rounds. We further derive a common reasoning action taxonomy from LLMs' reasoning traces, which proves instrumental in guiding LLMs to repair tests more effectively while at a lower financial cost. Applying our findings to the Codex agent further improves its repair success rate on challenging cases by 66.7%.

cs.SE

新论文 · v1arXiv:2610.38160

ATI-OPD: Adaptive Teacher Intervention for Stable Long-Sequence On-Policy Distillation

Jiannan Guan, Youxin Jiang, Yutai Hou, Qiguang Chen, Yirong Zeng, Haitao Wang, Dengyun Peng, Zheng Yan 等 12 位作者

原文摘要

On-policy distillation (OPD) trains a student on its own responses using token-level feedback from a teacher. As responses grow longer, the student's prefix can depart from the teacher's own reasoning, weakening its feedback on later tokens. Updates based on this feedback can reinforce the same deviations in later responses and destabilize training. Prior approaches such as length control and loss reweighting limit learning from problematic responses but leave their prefixes unchanged. Fixed cutoffs can also discard later reasoning. We propose \textbf{Adaptive Teacher Intervention for On-Policy Distillation (ATI-OPD)} to improve subsequent teacher guidance by revising the student's reasoning prefix. ATI-OPD replaces the earliest high-disagreement segment with a short teacher segment, then lets the student resume its own reasoning from the revised text. To handle renewed disagreement, it excludes persistently divergent suffixes from the distillation loss while retaining useful student continuation. Training diagnostics show less teacher-student disagreement, fewer gradient spikes, and fewer extremely long responses. Across five mathematical reasoning benchmarks and two teacher-student pairs, ATI-OPD improves average student-only accuracy over standard OPD.

cs.CL

新论文 · v1arXiv:2610.35616

Hierarchical Skill Evolution for Coding Agent

Jun Gao, Yun Peng, Changhai Zhou, Yuhan Zhu, Xiangyu Zeng, Shiyang Zhang Yuhua Zhou, Luozheng Qin, Qian Qiao 等 13 位作者

原文摘要

Skills provide agents with reusable experience to guide future decisions without modifying the underlying model. However, coding trajectories are often entangled with task- and repository-specific contexts, making it challenging to distill procedural skills that generalize across diverse coding environments. To address this challenge, we present Hierarchical Skill Evolution (HSE), which progressively distills procedural knowledge from coding trajectories through hierarchical evidence aggregation. HSE first extracts task-level evidence from individual trajectories, then aggregates recurring evidence within the same repository to form repository-specific skill candidates, and finally consolidates consistent patterns across repositories into transferable skills across repositories. The evolution is driven by an external writer-evaluator loop that generates and assesses skills while keeping the underlying coding model frozen. The loop externally updates the writer and evaluator policies to improve subsequent skill generation and assessment. This hierarchy progressively filters task-specific details while preserving procedures that generalize across diverse coding environments. Successful trajectories yield reusable guidance for future tasks, while failed trajectories reveal recurring failure modes that guide avoidance and recovery. We evaluate HSE on SWE-bench Verified and DeepSWE, using GLM-5.2 across Pi and Claude Code harnesses and GPT-5.5 and Opus 4.8 with Pi. Across both benchmarks, 10 skills evolved solely from SWE-Gym improve coding performance across all evaluated settings, demonstrating transfer across benchmarks, coding models, and agent harnesses. We further evolve the skill library from target-benchmark trajectories without verifier feedback, relying solely on the learned evaluator for candidate assessment and achieving additional gains through benchmark-tailored skills.

cs.CL

新论文 · v1arXiv:2610.35604

RoboDoctor: Training-Free Recursive Self-Improvement for Robot Policies

Heyang Yu, Haomin Li, Sihang Li, Jixin Yan, Shiyun Peng, Wanqi Zhong, Tianyu Xin, Yuan Yao 等 10 位作者

原文摘要

Generalist robot policies and language-model agents provide complementary capabilities, but completing a task also requires adapting their use to the consequences of physical interaction. We study execution as a locus of adaptation: feedback guides the current action and supplies evidence for revising subsequent decisions under fixed model weights. We introduce RoboDoctor, a training-free framework that connects diagnosis, grounded action, and persistent experience across three revision targets. The harness allocates control between policy and agent; intervention determines when to assist while preserving useful policy behavior; and operation refines physical methods through observed effects. Linked experience records retain the reason for a decision, the attempted method, its outcome, and conditions for reuse. On a curated archive of 54 RoboDojo-Sim task entries, RoboDoctor achieves a five-dimension mean score of 57.10, with a native success rate of 49.44%. Recorded interactions illustrate strategy changes, selective correction, and grasp revision. These findings characterize execution-time adaptation in the evaluated system; controlled trials remain necessary to isolate the contribution of cross-episode experience.

cs.RO · cs.CV

首次观测 · 记录 #1ACL Anthology · ACL 2026:acl:2026.acl-long.1558ACL Anthology · ACL 2026

MM-PoisonRAG: Disrupting Multimodal RAG with Local and Global Knowledge Poisoning Attacks

Hyeonjeong Ha Department of Computer Science, University of Illinois at Urbana-Champaign, Qiusi Zhan, Jeonghwan Kim, Dimitrios Bralios, Saikrishna Sanniboina Microsoft, Nanyun Peng Google and University of California, Los Angeles, Kai-Wei Chang University of California, Los Angeles and Amazon, Daniel Kang 等 9 位作者

原文摘要

Retrieval-augmented generation (RAG) has become a common practice in multimodal large language models (MLLM) to enhance factual grounding and reduce hallucination. Yet, its reliance on retrieval exposes MLLMs to knowledge poisoning attacks, in which adversaries deliberately inject malicious multimodal content into external knowledge bases to steer models toward generating incorrect or even harmful responses. We present MM-PoisonRAG, a framework to systematically study the vulnerability of multimodal RAG under knowledge poisoning. Specifically, we design two novel attack strategies: Localized Poisoning Attack (LPA), which implants targeted, query-specific multimodal misinformation to manipulate outputs toward attacker-controlled responses, and Globalized Poisoning Attack (GPA), which uses a single, untargeted adversarial injection to broadly corrupt reasoning and collapse generation quality across all queries. Extensive experiments on diverse tasks, multimodal RAG components, and attacker access levels reveal severe vulnerabilities: LPA achieves up to 56% attack success rate even under restricted access, and transfers effectively across four different retrievers without re-optimizing the adversaries. GPA completely disrupts model generation to 0% accuracy with just one poisoned content. Moreover, both LPA and GPA bypass existing defenses, underscoring the fragility of multimodal RAG and establishing MM-PoisonRAG as a foundation for future research on securing RAG frameworks against multimodal knowledge poisoning.

cs.CL · cs.CV

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.391ACL Anthology · ACL 2026

MulVul: Retrieval-augmented Multi-Agent Code Vulnerability Detection via Cross-Model Prompt Evolution

Zihan Wu, Jie Xu, Yun Peng The Chinese University of Hong Kong, Chun Yong Chong Monash University Malaysia, Xiaohua Jia

原文摘要

Large Language Models (LLMs) struggle to automate real-world vulnerability detection due to two key limitations: the heterogeneity of vulnerability patterns undermines the effectiveness of a single unified model, and manual prompt engineering for massive weakness categories is unscalable.To address these challenges, we propose MulVul, a retrieval-augmented multi-agent framework designed for precise and broad-coverage vulnerability detection. MulVul adopts a coarse-to-fine strategy: a Router agent first predicts the top- coarse categories and then forwards the input to specialized Detector agents, which identify the exact vulnerability types. Both agents use evidence retrieved from vulnerability knowledge bases to mitigate hallucinations. Crucially, to automate the generation of specialized prompts, we design Cross-Model Prompt Evolution, a prompt optimization mechanism where a generator LLM iteratively refines candidate prompts while a distinct executor LLM validates their effectiveness. This decoupling mitigates the self-correction bias inherent in single-model optimization. Evaluated on 130 CWE types, MulVul achieves 34.79% Macro-F1, outperforming the best baseline by 41.5%. Ablation studies validate cross-model prompt evolution, which boosts performance by 51.6% over manual prompts by effectively handling diverse vulnerability patterns.

cs.CL · cs.MA

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.694ACL Anthology · ACL 2026

LiveCLKTBench: Towards Reliable Evaluation of Cross-Lingual Knowledge Transfer in Multilingual LLMs

Pei-Fu Guo, Yun-Da Tsai, Chun-Chia Hsu, Kai-Xin Chen Department of computer science and informational engineering, National Taiwan University, Ya An Tsai National Taiwan University, Kai-Wei Chang University of California, Los Angeles and Amazon, Nanyun Peng Google and University of California, Los Angeles, Mi-Yen Yeh Academia Sinica 等 9 位作者

原文摘要

Evaluating cross-lingual knowledge transfer in large language models (LLMs) is challenging, as correct answers in a target language may arise either from genuine transfer or from prior exposure during pre-training. We present LiveCLKTBench, an automated generation pipeline specifically designed to isolate and measure cross-lingual knowledge transfer. Our pipeline identifies self-contained, time-sensitive knowledge entities from real-world domains, filters them based on temporal occurrence, and verifies them against the model’s knowledge. The documents of these valid entities are then used to generate factual questions, which are translated into multiple languages to evaluate transferability across linguistic boundaries. Using LiveCLKTBench, we evaluate several LLMs across five languages and observe that cross-lingual transfer is strongly influenced by linguistic distance and often asymmetric across language directions. While larger models improve transfer, the gains diminish with scale and vary across domains. These findings provide new insights into multilingual transfer and demonstrate the value of LiveCLKTBench as a reliable benchmark for future research.

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.764ACL Anthology · ACL 2026

Decoupling Task-Solving and Output Formatting in LLM Generation

Haikang Deng UCLA Computer Science Department, University of California, Los Angeles, Po-Nien Kung Google and UCLA Computer Science Department, University of California, Los Angeles, Nanyun Peng Google and University of California, Los Angeles

原文摘要

Large language models (LLMs) are increasingly adept at solving complex problems, such as mathematical reasoning and automatic evaluation. However, performance often degrades when prompts intertwine task instructions with rigid formatting requirements. This entanglement creates competing goals for the model, hindering its reasoning capabilities. To address this, we introduce Deco-G, a decoding framework that explicitly decouples format adherence from problem solving. Deco-G delegates format adherence to a separate Format Estimation Module (FEM), which performs probabilistic lookahead to estimate future format compliance rate and reweighs token probabilities, allowing the LLM to focus solely on task resolution. To make this approach both practical and efficient, we introduce three key innovations: instruction-aware distillation, a flexible trie-building algorithm, and HMM state pruning. Experiments across mathematical reasoning, event argument extraction, and LLM-as-a-judge demonstrate that Deco-G constantly gains over prompting or structured generation baselines, with guaranteed format compliance.

cs.CL

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

版本更新 · v2arXiv:2505.22657Google Research

3DLLM-Mem: Long-Term Spatial-Temporal Memory for Embodied 3D Large Language Model

Wenbo Hu, Yining Hong, Yanjun Wang, Leison Gao, Zibu Wei, Xingcheng Yao, Nanyun Peng, Yonatan Bitton 等 10 位作者

原文摘要

Humans excel at performing complex tasks by leveraging long-term memory across temporal and spatial experiences. In contrast, current Large Language Models (LLMs) struggle to effectively plan and act in dynamic, multi-room 3D environments. We posit that part of this limitation is due to the lack of proper 3D spatial-temporal memory modeling in LLMs. To address this, we first introduce 3DMem-Bench, a comprehensive benchmark comprising over 26,000 trajectories and 2,892 embodied tasks, question-answering and captioning, designed to evaluate an agent's ability to reason over long-term memory in 3D environments. Second, we propose 3DLLM-Mem, a novel dynamic memory management and fusion model for embodied spatial-temporal reasoning and actions in LLMs. Our model uses working memory tokens, which represents current observations, as queries to selectively attend to and fuse the most useful spatial and temporal features from episodic memory, which stores past observations and interactions. Our approach allows the agent to focus on task-relevant information while maintaining memory efficiency in complex, long-horizon environments. Experimental results demonstrate that 3DLLM-Mem achieves state-of-the-art performance across various tasks, outperforming the strongest baselines by 16.5% in success rate on 3DMem-Bench's most challenging in-the-wild embodied tasks.

cs.CV · cs.AI · cs.CL · cs.LG