Transformer
-
Motion Retargeting with Spatiotemporal Preferences for Humanoid Loco-Manipulation
2026-10-21 23:54:17Xiongbo Shi, Shihang Du, Zongyou Yu, Ao Zhou, Xudong Zhang, Ziyu Lin, Haodong Zhang, Junqiao Zhao, Guang ChenAbstract
Human motion retargeting bridges the embodiment gap between humans and humanoid robots by converting human demonstrations into reference trajectories for whole-body control and loco-manipulation. For object-interaction tasks, successful execution depends on accurately preserving body--object relations at critical stages. However, uniform per-frame iteration budgets do not account for these localized accuracy requirements, limiting the computational efficiency of retargeting. To address this limitation, we present Spatiotemporal Retargeting (STR), which uses keyframes and key body parts to allocate optimization effort. Specifically, a vision-language model selects relevant motion signals and body parts, and keyframes are identified from changes in the selected signals. Based on these selections, STR assigns additional solver iterations to keyframes and higher interaction weights to the selected parts. To further improve contact fidelity, we introduce Mechanics-Aware Contact Anchoring (MCA) to refine object-surface contact positions using an approximate object-wrench model while retaining the selected surface directions. Building on the resulting references, we develop Spatiotemporal Reinforcement Learning (STRL), which combines adaptive key-stage sampling with a contact reward based on target proximity and object contact forces for downstream control. Simulation experiments on a Unitree G1 and real-world experiments on a Unitree R1 demonstrate that STR improves geometric contact coverage and continuity at lower computational cost, while the resulting references improve downstream task completion across diverse object-interaction motions.
Abstract (translated)
URL
https://arxiv.org/abs/2610.30540
PDF
-
Behind Aggregate Scores: What Systematically Varying Arena-Hard's LLM Judge Reveals
Abstract
LLM judges are used to determine which of two answers is better. Trusting these verdicts requires knowing how repeatable they are and how much they depend on the judge's instructions. We show that similar aggregate scores can conceal substantial changes in individual verdicts, both when the same setup is repeated and when the instructions change. Using 500 Arena-Hard v2.0 prompts with two models' answers held fixed, we run GPT-4.1 as the judge under all eight combinations of criteria, output format and task framing, and repeat every setup unchanged at temperature zero on a pinned model version. Rerunning the default setup moves the aggregate score by about one point but changes the win/tie/loss outcome on 21.2% of prompts. Instruction changes introduce further sensitivity: in one setup, which asks for the verdict without reasoning and also drops several other instructions, outcomes change on 28.2 percentage points more prompts than the two setups' unchanged reruns do on average, and the aggregate score falls by 4.09 points. In another, where the criteria prioritise accuracy, outcomes change on 6.7 percentage points more prompts by the same comparison, while the aggregate score barely moves. A stable aggregate score therefore does not establish that individual verdicts are repeatable. This distinction matters when those verdicts are used as preference labels or to assess individual responses. Evaluations that rely on individual verdicts should repeat each setup they use and report both aggregate scores and prompt-level agreement.
Abstract (translated)
URL
https://arxiv.org/abs/2610.30512
PDF
-
HeyGen Voice: Scene-Aware Voice Cloning for Talking Videos
2026-10-21 21:42:44Pavel Andreev, Andrei Gusev, Adithya Narayan, Yi Ren, Ivan Somov, Zijian Li, Sergey Demyanov, Viktor Goriachko, Alan Tan, Nik Nolte, Sam Prokopchuk, Andy Pan, David Chou, Kirill Leonov, Penghan Wang, Jiewei Yuan, Zhibin Hong, Rong YanAbstract
Most voice-cloning systems are built for audio alone: they copy the reference recording's delivery and acoustics, so a voice cloned on a phone sounds like a phone recording whether the speaker is pictured in a studio, on a street, or in a cathedral. Speech for a talking video should sound like the pictured person speaking in the pictured place. We present HeyGen Voice, a scene-aware voice-cloning model that separates who is speaking from how and where: it keeps the speaker's voice from an everyday recording while the target image sets the delivery and acoustics. Prior image-conditioned TTS does half of this job: it either adds scene acoustics without keeping a specific speaker or infers a voice from a face. HeyGen Voice is a multimodal autoregressive transformer conditioned on the image, speaker and scene captions, and the voice recording, trained in five stages that move from multilingual TTS pretraining through simulated and real cross-scene adaptation to reinforcement and preference optimization. Two clone policies trade scene adaptation against fidelity to the original recording, and an audio anchor holds long scripts in one voice and one acoustic environment. Listeners prefer HeyGen Voice over leading commercial systems across a third-party arena, production votes, and talking-video studies: it ranks first (Elo 1201) on the public Artificial Analysis controlled-voice TTS leaderboard (October 2026); in a blind test, users cloning their own voice chose it over ElevenLabs v4 and Cartesia Sonic 3.6 in 62% and 54% of votes; and volunteers rated talking videos of their own digital twin about 10 to 30 points higher (out of 100) when voiced by HeyGen Voice than by leading commercial alternatives. Try HeyGen Voice and hear samples at this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2610.30470
PDF
-
Tactile-Based Robotic Manipulation: From Mechanics-Driven to Data-Driven Perspectives
2026-10-21 20:49:25Jingyi Xiang, Yixuan Wang, Wenzhen YuanAbstract
Tactile sensing provides robots with direct access to physical interactions at the contact interface, making it a critical modality for robust and dexterous manipulation. Over the past several decades, tactile-based manipulation research has evolved from mechanics-driven approaches, which explicitly interpret tactile signals through contact mechanics and task-specific reasoning, to data-driven approaches that learn tactile representations and manipulation policies directly from data. This paper reviews the development of tactile-based manipulation from these two complementary perspectives. We first introduce the fundamentals of tactile sensing, and then review, from a mechanics-driven perspective, how those signals form manipulation-relevant state representations and subsequently influence the robot action. We then examine the recent advances in data-driven tactile manipulation, covering imitation learning, reinforcement learning, vision-language-action models, and predictive models. Finally, we discuss open challenges and future opportunities toward scalable, generalizable, and robust tactile-enabled robotic systems.
Abstract (translated)
URL
https://arxiv.org/abs/2610.30438
PDF
-
Agent Mosaic: Recognizing Confidentiality Leaks in Agentic LLM Systems
2026-10-21 20:25:11Daniel Ruiz, Luca Luceri, Emilio FerraraAbstract
Modern LLM agents can retrieve information from accessible documents, and then disclose a sensitive conclusion assembled from them. This mosaic effect creates a critical risk for agents operating over confidential data. We quantify its realization with 30 synthetic cybersecurity, medical, and military examples, yielding 10,800 final answers from three LLM agents in controlled simulation and 1,440 answers from live agents in two open-source harnesses. Even under benign queries designed to be answerable from one document, GPT-5.6-Sol, the best performing model tested, discloses forbidden composites in 7.08% of answers (95% CI: 3.75-10.67%) in simulation, with worse disclosure rates in live harnesses. Adding visible classification headers and explicit aggregation rules reduces pooled disclosure rates substantially, yet unsafe disclosures persist in 32.37% of simulated answers (95% CI: 28.33-36.19%) across all intents despite these safeguards. Furthermore, increased refusals do not guarantee confidentiality, since significant refusals reveal the prohibited conclusion regardless. These results point to a troubling conclusion: sensitive data leaks via mosaic-style events are an observed confidentiality failure mode in the evaluated settings at non-negligible rates. They motivate stronger controls on what agents may infer and disclose in sensitive data environments with mixed-clearance users.
Abstract (translated)
URL
https://arxiv.org/abs/2610.30418
PDF
-
SIM-UT: A User Simulation based Evaluation Framework for Medical Triage Chatbots
2026-10-21 20:14:16Deepali Mittal, Dheeraj Mundhra, Yulia Sidorova, Tersia de Kock, Oli Munns, Sofie S MeyerAbstract
LLM-based chat agents are increasingly used for complex, multi-turn, multilingual medical triage in low- and middle-income countries (LMICs), yet evaluation methods focus on single-turn, English-only benchmarks that primarily test the chatbots' knowledge. In practice, users are often confused, overwhelmed, or emotionally vulnerable, giving unclear, misspelled single-word answers or long messages in regional dialects, and the context of a conversation can shift rapidly. We propose SIM-UT (Simulation of User Behaviour for Triage), a multi-turn, simulation-based evaluation method that identifies ecologically valid, contextually relevant failure modes of LLM-based triage chat agents. Using SIM-UT, we evaluated three real-world WhatsApp-based health chatbots: a respiratory screening programme in India, a hearing health support programme in South Africa, and a diabetes and lifestyle coaching agent in Kenya. We compared a medically fine-tuned open model (MedGemma) against general-purpose frontier models (ChatGPT and Claude) across 5,000+ multilingual simulated conversation turns grounded in the personas, constraints, and communication patterns of the real user populations, over 2-4 iterative rounds. A key failure mode emerged in both model families: de-escalation risk. The agents initially identified medical emergencies correctly, but when the user minimised the symptoms without introducing new clinical information, they de-escalated, exhibiting clinically dangerous sycophancy, for example agreeing that vaginal bleeding in pregnancy was "not a problem" or that decreased fetal movements at 32 weeks "can be normal". We recommend user simulation-based behavioural evaluation, grounded in real user context and covering adversarial and edge cases, as standard practice for triage chatbots in production, so that safeguards can be added before deployment or in subsequent iterations.
Abstract (translated)
URL
https://arxiv.org/abs/2610.30412
PDF
-
Climbing the Ladder: Automating Data Pipelines and Reproducing Baselines in Low-Resource Settings
2026-10-21 19:36:05Donato Meoli, Eryk WdowiakAbstract
The massively multilingual translation systems that now include Sicilian learned it from noisy, non-standardised web text. To obtain better translation quality, we redevelop the small neural machine translation (NMT) system that Wdowiak customized for Sicilian. We rebuild on maintained software the extraction of the parallel text from the PDFs of the bilingual journal Arba Sicula, whose sentences are aligned by a language-agnostic sentence encoder with thresholds tuned against an existing hand alignment and then filtered for the typical errors of automatic alignment. Furthermore, the tokenization and the desinence-biased subword segmentation of the original system are ported to a Perl-free implementation, and all the results are measured on a frozen test set drawn from the journal. We compare small Transformers trained from scratch with the pretrained NLLB model adapted by low-rank fine-tuning, in both directions and with Italian as a closely related bridge language, and we adapt to the pretrained model the reverse-training recipe of the original system, i.e., a stage on forward-scored back-translations of the web Sicilian followed by a stage on the curated literary text. The pretrained model is by far the better of the two, and the staged training improves the translation into Sicilian more than the translation from it. On the test set of the original system, and in the tokenized space in which it was scored, the staged model with a larger adapter and more back-translations is above the reverse-training recipe in the directions that involve English and below it between Sicilian and Italian. Yet, a multilingual adapter degrades the directions on which it was not trained, which limits its use as a general multilingual model.
Abstract (translated)
URL
https://arxiv.org/abs/2610.30374
PDF
-
CiteAnchor: Source-Aware Grounding for Hallucinated Citation Detection and Resolution
2026-10-21 19:13:15Chenyi Tong, Ping Nie, Yu ZhangAbstract
Hallucinated citations make it harder to trace sources and verify the evidence behind scientific claims. Detecting them is difficult because bibliographic records can be incomplete or differ across source types and publication versions. A plausible match alone neither validates the citation nor establishes its underlying source. We introduce CiteResolve, a benchmark of 423 real-world hallucinated citations annotated for source recoverability. We further propose CiteAnchor, a source-aware grounding framework that adapts retrieval and field-level verification to each source type and separately determines citation validity and whether a unique source can be recovered. On an existing detection benchmark, CiteAnchor achieves 0.841 F1 with GPT-5.6 Luna and 0.784 with an open-weight backbone, outperforming the baseline with the same backbones by 7.9 and 36.8 points. On CiteResolve, it achieves 74.7% and 70.7% end-to-end task success, outperforming the corresponding web search baselines by 16.3 and 65.3 percentage points. Our code and data are available at this https URL.
Abstract (translated)
URL
https://arxiv.org/abs/2610.30355
PDF
-
AQL-Lite: Enabling Efficient Knowledge Acquisition for Arabic LLMs via Data-Centric Continual Pretraining
2026-10-21 19:04:28Houssam Eddine Boukhalfa, Mourad Bouache, Mourad DahmaneAbstract
Training Arabic large language models faces challenges from noisy web data, dialectal variation, and the high cost of continual pretraining (CPT). We introduce AQL-Lite (AQL: Arabic Quality Language), a time- and resource-efficient framework for adapting a multilingual LLM to Modern Standard Arabic (MSA). AQL-Lite couples a quality-first curation pipeline (structural cleaning, heuristic and model-based filtering, MSA selection, and deduplication) with a partial-parameter CPT recipe that trains RS-LoRA adapters across the network together with the last eight transformer blocks and the embedding matrix, under layer-wise learning rates. At a matched 1B-token budget, CPT on AQL-curated data raises a four-benchmark Arabic mean by 3.51 points over CPT on raw web text, which yields no gain over the base model. In a controlled 5B-token comparison, AQL-Lite recovers 94% of the gain of full-parameter CPT (mean 73.19 vs. 73.58) in 52% of the wall-clock time and outperforms adapter-only baselines on every benchmark, and a component ablation shows that adapters, unfrozen blocks, and embedding training each contribute. Scaled to 40B MSA tokens and instruction-tuned on one million examples, the resulting 12B model scores 75.3 on the seven-benchmark Open Arabic LLM Leaderboard v2 (OALL v2) suite. These results suggest that partial-parameter CPT on a carefully curated corpus is a practical route to efficient MSA adaptation.
Abstract (translated)
URL
https://arxiv.org/abs/2610.30344
PDF
-
EgoHand: Monocular 3D Hand Reconstruction from Egocentric Video
2026-10-21 18:09:27Chao-Yeh Chen, Tiffany Chen, Lin Duan, Tianyi Hu, Yu-Hsiang Huang, Umar Iqbal, Jiefeng Li, Edy Lim, Yanchen Liu, Yihan Pang, Hang Qiu, Jun Saito, Shao Su, Qingsi Wang, Xiufeng XieAbstract
We introduce EgoHand, a monocular 3D hand reconstruction model that achieves state-of-the-art performance on egocentric video. Central to our approach is the curation of a high-quality training corpus of 14 million images, combining synthetic data, in-the-wild video, and multi-camera motion capture to balance scene diversity and supervision quality. The model adopts SOMA-Hand as its parametric representation and processes 2D hand crops with a transformer encoder, incorporating camera cues and an auxiliary 2D branch to improve geometric consistency and image alignment. Zero-shot evaluations on HOT3D, ARCTIC, and H2O demonstrate strong out-of-domain generalization, with lower joint-position, root-translation, and acceleration errors than competing baselines at an inference latency of approximately 9.69 ms. A blinded user preference study further shows that participants favor EgoHand's reconstructed hand poses over those from Apple EgoDex and Ropedia Xperience-10M. We release the EgoHand model and source code under the OpenMDW license, allowing free commercial use of the model.
Abstract (translated)
URL
https://arxiv.org/abs/2610.30293
PDF
-
Agentic World Action Models
2026-10-21 17:59:56Zhuoyang Liu, Dantong Niu, Botong Zhang, Boning Shao, Jing Wang, Ruijie Zheng, Yuvan Sharma, Runyu Lu, Jiaming Liu, Renrui Zhang, Roei Herzig, Yutong Bai, Ken Goldberg, Danfei Xu, Yuke Zhu, Guanzhi Wang, Linxi "Jim" Fan, Trevor DarrellAbstract
Recent world action models provide visual foresight and future prediction for robot manipulation, but lack explicit mechanisms for compositional skill reuse. Agentic robotics combines code-based planning and reasoning with reusable skills, but often lacks predictive grounding in physical dynamics. We introduce AW-0, an agentic world action model framework that integrates code-based planning with joint video and action prediction. We jointly model code, future video, and robot actions as three coupled generative streams within a unified mixture-of-transformers architecture. Within this architecture a code expert generates parameterized skill routines and control logic while video and action experts generate future visual states and low-level motor commands. Code serves two complementary roles: it structures multi-step prediction and control within the world model and represents skills as reusable programs for subsequent agentic execution. We further introduce an automated curation pipeline that labels robot and human video trajectories with aligned programs and consolidates recurring routines into reusable skill libraries. Evaluations on RoboTwin 2.0 and physical bimanual manipulation tasks show improved task performance over VLA and WAM baselines, alongside gains in language following, semantic reasoning and robustness, demonstrating that agentic code enables connection between planning, prediction and execution within WAMs.
Abstract (translated)
URL
https://arxiv.org/abs/2610.30136
PDF
-
Quant-Arena: Benchmarking LLM Agents for Quantitative Trading Research
2026-10-21 17:54:01Houde Qian, Zheng Jiang, Wenyi Li, Chenxi Xu, Xinjie Lin, Yuzhao Peng, Ruoyu Han, Youjie Zheng, Kaisen Yang, Qinhuai Na, You Fu, Calvin XiaoAbstract
Quantitative research is an iterative loop: a researcher proposes a market hypothesis, implements it as an executable signal, tests it on historical data, and decides what to keep. Existing benchmarks score formulas, code, or individual trading decisions, leaving open whether large language models (LLMs) can complete this loop and produce strategies that work beyond the research period. We introduce Quant-Arena, a benchmark and public leaderboard for end-to-end quantitative research on real A-share market data. A model constructs features and stock-scoring code, receives feedback only on 2022--2023, and freezes one strategy for assessment on the sealed year 2024. A cash-account evaluator applies trading constraints and costs, including market impact estimated from a simulated order book. Scores are dominated by realized excess return and complemented by signal quality and the diversity of features used by the strategy. We evaluate nine frontier models, among which GPT-6 Astra scores highest; yet no model beats the equal-weight market in 2024. Quant-Arena is open to submissions from any model or agent harness.
Abstract (translated)
URL
https://arxiv.org/abs/2610.30113
PDF
-
AtomicRL: Atomic Reinforcement Learning for Long-Horizon Robotic Manipulation
2026-10-21 17:45:24Likui Zhang, Zhe Zhu, Zhihao Zhan, Yibo Bai, Hanlin Wang, Dongwei Li, Bin Wang, Tianle Zhang, Xiaodan Liang, Liang LinAbstract
Vision-language-action (VLA) models have shown promise in robotic manipulation, but imitation learning (IL) alone often struggles to master complex long-horizon skills. Reinforcement learning (RL) offers an effective avenue for further improving VLA policies. However, existing VLA-RL methods typically suffer from delayed and biased credit assignment due to trajectory-level sparse rewards, while imbalanced subtask mastery creates bottlenecks that limit overall performance. To address these challenges, we propose AtomicRL, an RL post-training framework that leverages atomic task structures for reliable credit assignment and adaptive exploration. AtomicRL introduces Semantic-Aware Policy Optimization (SAPO), which uses semantic-aware gating to constrain TD residuals from failed atomic subtasks from propagating across task boundaries, mitigating credit contamination while preserving intra-subtask credit assignment and value bootstrapping. In addition, Bottleneck-Adaptive Target Exploration (BATE) dynamically coordinates task sampling and local exploration incentives based on the mastery of individual atomic subtasks, directing learning toward the remaining bottlenecks. Finally, we develop a Real2Sim2Real pipeline that bridges online RL in simulation with real-world deployment, facilitating practical transfer to physical robots. On LIBERO-LONG, AtomicRL achieves absolute performance gains of 3.6% and 4.0% over PPO across two VLA architectures. On CALVIN ABC-D, it improves full-task performance by up to 9.5% over PPO. Combined with the Real2Sim2Real pipeline, AtomicRL further improves real-world task performance by an average of 21.3%. These results demonstrate that AtomicRL effectively improves the learning efficiency and manipulation performance of VLA policies on long-horizon tasks.
Abstract (translated)
URL
https://arxiv.org/abs/2610.30097
PDF
-
Looped Uniform Diffusion Language Models
2026-10-21 17:24:35Saiteja Utpala, Chieh-Hsin Lai, Yuki Mitsufuji, Praneeth VepakommaAbstract
Uniform diffusion language models (UDLMs) generate text by iteratively denoising sequences of ordinary vocabulary tokens, any of which may be corrupted and subsequently revised. During training, UDLMs must denoise from a context that may itself be unreliable, whereas masked diffusion language models (MDLMs) always condition on a clean context. This challenging training objective makes UDLMs more FLOP-intensive to train than MDLMs, but it yields stronger downstream reasoning performance. Recently, selective looping, which recursively applies a few early Transformer layers within each denoising step, has been shown to improve training efficiency in MDLMs under an iso-FLOP budget. We investigate looping in UDLMs while keeping the training objective and sampler unchanged. We find that looping significantly improves training efficiency: looped UDLMs reach the validation NLL of a non-looped UDLM trained on about 0.5 trillion tokens with up to 2.8x fewer training FLOPs on OpenWebText and 3.3x fewer on FineWeb-Edu. We also find that, at a matched NFE budget, looped UDLMs reach up to 9.8% lower generation perplexity than the non-looped UDLM, with most gains achieved at moderate loop counts. Finally, we study different loop-count distributions and find that sampling smaller loop counts more frequently during training improves inference efficiency: on downstream tasks, this strategy uses 12% fewer loops than uniform loop-count sampling without degrading accuracy.
Abstract (translated)
URL
https://arxiv.org/abs/2610.30066
PDF
-
From Reasoning to Action: Benchmarking Frontier AI Models on LIBERO-10 Robot Manipulation Tasks
2026-10-21 17:11:52May Espinola, Qifan Zhang, Yu XiangAbstract
Recent advances in frontier AI models have demonstrated remarkable capabilities in visual understanding, reasoning, planning, and tool use, raising the possibility of using general-purpose AI models for robot manipulation without task-specific training. However, their manipulation capabilities have not been systematically evaluated on standardized robotics benchmarks. In this report, we present a unified closed-loop manipulation framework for evaluating frontier AI models on the LIBERO-10 benchmark. Our framework enables models to control a robot directly through visual observations, robot proprioception, and structured tool calls, primarily by specifying target end-effector poses and gripper commands, without task-specific fine-tuning or pretrained manipulation skills. We evaluate six frontier AI models, including GPT-6 Astra, GPT-6.1 Sol, GPT-6 Luna, Claude Opus 5, Claude Opus 5.5, and Gemini-4 Argon, across three reasoning effort levels, resulting in 18 configurations and 1,800 evaluation episodes. Our results show that GPT-6 Astra achieves the highest success rate of 44%, followed by Gemini-4 Argon at 43%. Increasing reasoning effort generally improves manipulation performance, but the benefits vary across models and come at additional inference cost. We also observe substantial differences across tasks, with models performing relatively well on object placement tasks but struggling with precise spatial arrangements and complex multi-step interactions. These findings demonstrate both the potential and limitations of frontier AI models for robot manipulation, highlighting the challenges of translating reasoning capabilities into reliable physical actions. We release our evaluation framework and experimental results to support reproducible benchmarking and future research on reasoning-based robot manipulation.
Abstract (translated)
URL
https://arxiv.org/abs/2610.30051
PDF
-
SP-HLM: Sparse-Pose-Driven Motion Generation and Whole-Body Control for Humanoid Loco-Manipulation
2026-10-21 17:03:34Zhaohui Yang, Quanbao Lin, Chengqi Shi, Zhenglang Weng, Jicheng Lin, Boyu Chen, Tengjie Zhu, Guanzhu Ren, Jiahao Xu, Yang Feng, Xiaokang Yang, Yao MuAbstract
Humanoid loco-manipulation requires an effective interface between high-level task policies and whole-body control. Existing interfaces typically rely on a single action representation for both task intent and whole-body coordination. This may burden the high-level policy or leave one aspect underspecified. We present SP-HLM, a framework built on a sparse-pose-driven interface. The interface combines five task-critical body poses with a learned latent motion prior to represent spatial intent and whole-body coordination in complementary forms. SP-VLA, a vision-language-action model, first predicts task-relevant sparse poses and finger joint positions from multimodal observations and language instructions using asynchronous action-chunk inference. Given these sparse targets, SP-MGM, a motion generation model trained through two-stage distillation, reconstructs coordinated full-body motion in real time. A pretrained motion encoder maps this motion to a compact latent representation. SP-GMT, a general motion tracker, then integrates the explicit sparse targets with this implicit motion prior for whole-body control. Experiments in simulation and on a real humanoid robot demonstrate improved motion tracking, whole-body coordination, and loco-manipulation success over existing approaches. Additional results and videos are available on our \href{this https URL}{\textcolor{blue}{project page}}.
Abstract (translated)
URL
https://arxiv.org/abs/2610.30044
PDF
-
What the Author Wrote, What the Agent Reads: Silent Cue Loss in Enterprise Document Ingestion
2026-10-21 16:51:25Yikai Zhao, Saurabh Pandey, Pradeep Kumar MisraAbstract
Enterprise agents act on documents. Before an agent can follow a policy, answer a compliance question or file a ticket, some ingestion layer must turn a Word file or PDF into tokens - and that layer is where an agent deployed in the wild silently loses information. We show that one layer is lost by every mainstream converter - the meaning an author encodes in how text looks - and that the loss makes agents confidently wrong rather than visibly incomplete. We build a task from 76 real policy documents: recover the requirements the author singled out, scored against the file's own OOXML properties by a deterministic matcher with no LLM judge. Formatting-blind ingestion collapses from micro F1 0.690 to 0.150 (docling) and 0.075 (plain text), a gap that replicates across six answering models from two vendors - five Claude models spanning the Opus, Sonnet and Haiku tiers, and OpenAI's GPT-5.6. An agent that cannot see the cue guesses, and docling's false-positive rate on requirements the author did not flag is four times ours. A span-level leave-one-out localises the loss to two cues - highlight (+0.302) and colour (+0.262), both Holm-adjusted p < 1e-4 - while bold, italic, monospace and strikethrough each sit within +/-0.003 of zero. The same asymmetry appears in PDF. Colour is also the cue no mainstream converter preserves and no parsing benchmark scores, and it is common: 32.8% of 6,460 documents carry a deliberate colour or highlight, 10.3% on an obligation. A pre-registered test shows that typing a cue for the agent adds +0.085 beyond merely preserving it (Holm-adjusted p = 0.049). We validate the labels against rendered page images through an independent modality (kappa = 0.72) and bound what wording alone can achieve (AUC 0.664). For agent builders the practical finding is that ingestion, not reasoning, is where this reliability is won or lost, and that the fix is cheap and measurable.
Abstract (translated)
URL
https://arxiv.org/abs/2610.30026
PDF
-
TIDE: Tree-Induced Fine-Grained Diagnostic Evaluation from Model Behaviors
2026-10-21 16:27:48Jiayi Chen, Hui Li, Yujiong Shen, Qiyuan Peng, Yanxin Li, Chenhao Huang, Shaofan Liu, Guoqiang Zhang, Junjie Ye, Shihan Dou, Ming Zhang, Maxm Pan, Tao Gui, Qi Zhang, Xuanjing HuangAbstract
Benchmark performance is the dominant standard for comparing large language models (LLMs), typically summarized by a single aggregate score, such as accuracy, pass rate, or rubric score. This aggregate score provides limited information about a model's underlying capabilities, yet understanding these capabilities is crucial for model development and evaluation. To this end, we propose Tree-Induced Diagnostic Evaluation (TIDE), which recasts diagnostic evaluation as behavior-driven decision-tree induction to perform fine-grained capability diagnosis. TIDE recursively builds a hierarchical capability tree from benchmark samples and model scores: at each node, an LLM judge proposes candidate splits and performs sample-to-group classification, and behavior-based information gain selects among them. We apply TIDE to five models across four benchmarks to demonstrate its effectiveness, and find fine-grained capability contrasts invisible at the aggregate level: Claude Opus 4.6 achieves 100% on microeconomic market reasoning where GPT-5.4 scores 83.3%, but trails GPT-5.4 on contract law reasoning with 77.8% versus 88.9%. We hope TIDE can be extended to past and future benchmarks so that the rich behavioral signal in evaluation results is fully leveraged for understanding model capabilities.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29985
PDF
-
Training-free Acceleration for Blind Compressed Image Restoration via Activation-Calibrated Singular Value Decomposition
2026-10-21 16:16:41Suhang Yao, Xin Li, Xijun Wang, Yufei Wang, Yeying Jin, Zhibo ChenAbstract
Transformer-based models achieve strong performance on blind compressed image restoration (CIR) under unknown compression settings, but their densely stacked projection operators make deployment costly. We study training-free acceleration for such models through an empirical, operator-level analysis. Specifically, we profile where computation is concentrated, characterize channel-wise activation distributions under compressed inputs, and examine the singular-value decay of representative projection weights. These measurements identify the recurring projections in the attention, feed-forward, convolutional, and prompt-conditioning pathways as effective low-rank replacement targets that preserve the original inference graph. They further establish truncated SVD (TSVD) as a principled training-free baseline for blind CIR and reveal why a uniform low-rank objective is mismatched to the channel-imbalanced feature distributions of compressed images. Accordingly, we formulate Activation-Calibrated Singular Value Decomposition (ASVD) for this restoration setting, deriving low-rank factors entirely offline from pretrained weights and lightweight activation statistics gathered from only a few compressed calibration images. By weighting the decomposition with channel-wise activation magnitudes, our ASVD assigns larger penalties to errors on more active channels, requiring neither retraining nor online calibration and leaving the restoration pipeline intact. When instantiated on PromptCIR at rank 64, ASVD consistently improves PSNR-B over TSVD across four standard benchmarks under identical parameter and MAC budgets. At this operating point, the factorized model cuts parameters by 59% (34.75M to 14.34M) and computation by 16.1% (258.79 to 217.16 GMACs for a 256 x 256 input), enabling acceleration without retraining.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29959
PDF
-
ToM2Vec: Continuous Theory-of-Mind Tokens for Multi-Agent Coordination
2026-10-21 16:05:54Nithin Sivakumaran, Ali Payani, Hyunji Lee, Mohit Bansal, Elias Stengel-EskinAbstract
Forming an accurate theory-of-mind (ToM) is key to successful multi-agent collaboration and requires reasoning under uncertainty to capture multiple possible partner states. This is hard to do in the text representation space native to large language model (LLM) agents, which more easily describe a single partner state. To address this challenge, we introduce ToM2Vec, which learns to form abstract theories-of-mind in a rich latent space via a fixed budget of continuous reasoning tokens. During training, supervision for these tokens comes from the partner's actual thoughts, which the ToM tokens are trained to abstract and match. At test-time, ToM tokens are generated by each agent and act as conditioners for that agent's thoughts and actions. Across several environments covering collaborative, competitive, and mixed-motive settings (Overcooked-AI, multiparty negotiation, Decrypto, Hanabi, and Collab Escape/Capture), we evaluate ToM2Vec's ability to support multi-agent interactions and collaboration, finding consistent improvements over the base model and a variety of competitive trained and prompted baselines. We show that latent reasoning is critical, with performance dropping when reasoning is expressed in text. Moreover, we show that ToM2Vec training generalizes to partner model families not seen at train-time, including a proprietary model (GPT-5.4). In our analysis, we provide evidence that the latent tokens encode a ToM, rather than task-specific information: ToM2Vec's tokens improve the agent's ability to directly answer questions about partner beliefs, desires, and next actions.
Abstract (translated)
URL
https://arxiv.org/abs/2610.29934
PDF
