FrontierPhysics: Evaluating Agents for End-to-End Physics Research
Bingran You, Wenjun Ke, Ria Rosenauer, Bo Liu, Wenbo Chen, Min Chen, Shuqi Xu, Yue Ma 等 72 位作者
AI 速览：作者提出包含56个由物理学家基于自身研究设计的端到端物理研究基准，用确定性验证器与专家评分量表同时考察智能体研究的正确性与可发表性。
AI agents are playing an increasingly important role in physics research. However, existing physics benchmarks only evaluate fixed questions, isolated sub-tasks, or paper reproduction with given solutions, and rarely check research quality alongside correctness. We introduce FrontierPhysics, an end-to-end benchmark with 56 tasks made by 42 physicists based on their own research, covering nine domains across theoretical, computational, and experimental physics. Each task requires agents to review the literature, develop a research plan, conduct research and experiments, and produce structured outputs and scientific manuscripts. Combining deterministic verifiers with expert-authored rubrics, FrontierPhysics evaluates whether agents' research is correct and publishable. Across 22 models in 26 configurations, no configuration passes more than 46 of its 168 rollouts. Manual review of agent trajectories shows that frontier agents' main weaknesses are insufficiently rigorous reasoning and limited expert intuition, rather than gaps in physics knowledge or execution errors.
Triangle finding is a fundamental problem in graph theory that has played a central role in the development of quantum algorithmic techniques. For $N$-vertex graphs given by their adjacency matrices, a long line of work has improved the quantum query upper bound from $O(N^{3/2})$ to $O(N^{5/4})$. However, the best known lower bound has remained the elementary $Ω(N)$ bound obtained from unstructured search. We prove the first superlinear quantum query lower bound for triangle finding, showing that $Ω(N^{13/12})$ queries are necessary. Our result also implies the same lower bound for triangle detection and for detecting every fixed connected non-bipartite graph. The proof builds on the compressed oracle approach of Zhandry and the transfer approaches of Belovs. It compares the computation on graphs with a planted triangle to the computation on random graphs conditioned to be triangle-free, through a map on the input registers that almost commutes with every query. We average over local changes to the input while preserving its distribution, through a procedure known as conditional resampling. Our main technical contribution is a set of estimates that let us control the query error under this averaging, despite the dependence introduced by conditioning a dense graph to be triangle-free.
Group Relative Policy Optimization (GRPO) has advanced generative model post-training in verifiable domains such as code and mathematics, but evaluating video quality, especially emotional expression, often depends on subjective aesthetics and has no uniquely correct target. Existing approaches typically rely on fixed rewards for visual quality, motion quality, and text alignment that poorly capture emotional intent and can invite reward hacking, while RLHF is difficult to scale and subject to individual aesthetic differences that complicate the calibration of facial, body motion, and chromatic cues across narrative contexts. We introduce EmoReward, a new self-evolving training mechanism that distills evidence reliability from reward traces collected during an actual training run to adaptively guide multi-reward weight allocation in subsequent training by determining which emotional evidence to trust and how much to weight it. EmoReward first plans specific grounding cues from the emotional prompt to guide evaluation, then uses heterogeneous signal primitives to score the generated video during the act stage. Reflective Reward Gating calibrates their contributions through instance-level gates for phase, face confidence, conflict, and subtlety, then renormalizes weights over the active set to produce the sample's fused reward. RewardMemory then estimates each primitive's reliability from rollout scores and outcomes within each emotional regime in the finished training run, then fixes bounded multipliers that calibrate fusion weights in the next training run without updating reward models. Experiments on HunyuanVideo 1.5 show improvements across multiple RL algorithms and compare EmoReward with other agentic reward methods under SAGE-GRPO. Ablations and transfer experiments on SD 3.5 and FLUX further support adaptive reward calibration for visual generation.
Radiologists interpret computed tomography (CT) scans by jointly determining what a finding represents and where it lies in three-dimensional (3D) anatomy. However, CT models have treated CT interpretation and 3D segmentation as separate tasks, leaving textual findings spatially unverifiable and overlooking potential synergy between clinical understanding and spatial perception. Here we present HoloCT, a generalist 3D foundation model unifying CT interpretation and voxel-level segmentation by jointly modelling whole-volume context and fine-grained anatomical detail, providing 3D masks as evidence. We curated HoloDATA, comprising 195,921 CT scans from 73 public datasets and six clinical cohorts, with 6.2 million instruction dialogues and 2.8 million CT-mask pairs. Across 346 tasks spanning segmentation, lesion screening, visual question answering, and report generation on 61 internal and 12 external datasets, HoloCT outperformed specialist and foundation models. In multicentre validation across 11,455 clinical CT scans from five hospitals, HoloCT outperformed competing models in lesion segmentation and report generation by 18.2 and 15.3 points. It also achieved 85.1% mean accuracy in screening thyroid tumours and colon lesions, exceeding all four participating radiologists in a retrospective reader study. Further analyses revealed bidirectional semantic-spatial synergy: language-guided segmentation delineated unseen anatomical structures and lesions as well as diffuse free-text findings, outperforming the leading approach by 22.1 points, whereas prompting the model to generate corresponding 3D masks alongside its textual outputs significantly improved interpretation performance. HoloCT shows that clinical understanding and spatial perception can be learned together within a single foundation model and offers a path towards more accurate and reliable CT analysis.
Jingyang Zhao, Trung Q. Duong, Een-Kee Hong, Hyundong Shin
The traveling salesman problem (TSP) is one of the fundamental problems in combinatorial optimization that asks for a minimum-cost Hamiltonian cycle in a complete undirected graph with $N$ vertices. For the metric TSP, where edge costs satisfy the triangle inequality, a long line of work has produced constant-factor approximation algorithms, culminating in the best-known approximation ratio $α<3/2-10^{-36}$. The biased TSP (BTSP) generalizes the metric TSP to bipartitioned networks: the vertex set is partitioned into two subsets, edge costs satisfy the triangle inequality within each subset, while edges crossing the partition may violate it by a certain bias factor. This model captures heterogeneous routing scenarios in which distances are metric only locally but may be distorted across different regions or vertex types. Previously, the only known result for the BTSP was a $7/2$-approximation algorithm with running time $O(N^3)$. In this paper, we present two new approximation algorithms that significantly improve the state of the art. The first achieves an approximation ratio of $2α+\varepsilon$ (strictly less than $3$) in polynomial time $N^{O(1)}$ for any constant $\varepsilon>0$, where the exponent grows at least linearly with $1/\varepsilon$. The second achieves a $4$-approximation in only $O(N^2)$ running time, reducing the previous cubic-time bound by a full polynomial factor. These advances directly support emerging networked systems---including multi-tier 6G architectures and hybrid quantum-classical networks---where cross-region transitions incur biased and non-metric overheads.
Achieving both high pronunciation accuracy and fine-grained controllability remains a central challenge in Japanese text-to-speech (TTS), as Japanese is an unsegmented language mixing logographic Kanji and phonographic Kana with extensive polyphony. Traditional cascaded systems convert text into phoneme sequences via a front-end, ensuring pronunciation control but discarding lexical semantics and prosodic cues in surface orthography. Conversely, recent zero-shot TTS models directly consume raw orthographic text to improve naturalness, yet suffer from polyphone errors and lack explicit pronunciation control. To bridge this gap, we propose a two-stage framework for flow-matching TTS. First, we investigate Ruby text representations with orthography perturbation to jointly preserve surface forms and target pronunciations for reliable control. Second, we introduce On-Policy Distillation (OPD) on large-scale text-only corpora to enhance pronunciation accuracy on raw orthographic text without paired speech. On the Joyo-Kanji-Yomi benchmark, our model achieves 97.13% Accuracy, 2.02% Target PER, 0.70% Sentence PER, and 4.21% Sentence CER under raw text while retaining strong controllability.
Data-driven pre-fault dynamic security assessment (DSA) rapidly evaluates the dynamic risk of credible contingencies on a power system using machine learning (ML). Existing approaches face two limitations. First, they require a large labelled database for training, with a separate model trained, tuned, and maintained for each contingency in a potentially long list of credible contingencies. Second, the trained models generalize poorly to unseen contingencies. This work addresses these limitations by using a tabular foundation model (TFM) that assesses stability through in-context learning, and requires no retraining or hyperparameter optimization. We use a single TFM to assess many contingencies at once, removing the need to train one classifier per contingency. We also characterize when using electrical distance coordinates (EDC) as continuous features enables the TFM to generalize to unseen contingencies and when it does not, showing how a few labelled samples can reliably improve generalization. Through comprehensive case studies on the IEEE 68-bus system, we show that a single TFM attains an average Macro F1 score of about 90\% with only 120 labelled samples per contingency, outperforming both classical ML and deep learning baselines, without any model retraining or hyperparameter tuning. For held-out contingencies, we show that using just 10 labelled samples of the new contingency with a modified EDC encoding matches the balanced accuracy of the best achievable transfer learning oracle model, which requires fully labelled data and is not deployable in practice. Overall, this initial study paves the way for developing and deploying foundation models for power system operations, with possible applications across multiple operational tasks.
Physics sensing in evolving environments requires reconstructing dense spatiotemporal fields from sparse observations while adapting where future observations should be collected. Existing reconstruction and sensor-placement methods have made substantial progress, but they typically rely on static sensing assumptions: they either use a single layout across all frames or optimize placements independently at each frame, which limits their ability to produce temporally coherent sensing trajectories. We introduce DyPSI, a dynamic physics sensing framework that formulates this problem as the joint generation of future fields and sensor trajectories conditioned on historical observations. DyPSI lifts discrete sensor coordinates into continuous Sensor Position Fields, encodes fields and sensing configurations with a shared Functional Tucker representation, and trains a joint diffusion model to generate future fields and sensing trajectories simultaneously in the resulting latent space. A Gaussian-process temporal kernel correlates the perturbation injected during diffusion training, biasing the denoiser toward temporally smooth recoveries, and we further provide an analysis linking the EDM training objective to a sequence-aggregated A-optimality criterion and a kernel-induced bound on discrete trajectory variation. Experiments on turbulent flow, GLORYS12 sea-surface temperature, and 3D car aerodynamics show that DyPSI consistently outperforms static and frame-wise placement strategies, with substantial error reduction under tight sensor budgets. The code is available at https://github.com/mengqi132/DyPSI.
Jailbreak attacks remain a critical threat to the safe deployment of large language models (LLMs). While prior work has primarily studied attacks and defenses at the prompt level, we show that this prompt-centric paradigm overlooks a structural vulnerability in stateful, function-calling environments. In such applications, developer-defined schemas, structured arguments, and untrusted tool outputs are interleaved into a single shared model context. This architecture expands the attack surface by blurring the boundary between trusted control logic and untrusted data, allowing adversarial intent to be distributed across a multi-turn execution path. We exploit this architectural flaw through SMT, a black-box attack framework based on Simulated Moderation Traces. Departing from purely prompt-based interactions, SMT constructs a multi-turn trajectory that simulates a legitimate moderation-auditing workflow. Within this trajectory, a fabricated moderation frame leverages red-team testing as a pretext to elicit harmful generations. The subsequent validation feedback treats safety refusals as execution failures, prompting refinements that gradually weaken the model's safety constraints and ultimately trigger harmful outputs. Extensive empirical evaluations on prominent commercial LLMs from five different providers across two standardized safety benchmarks show that SMT consistently achieves the highest average attack success rate and HarmScore while requiring a near-minimal number of queries, substantially outperforming existing baselines. These findings demonstrate that prompt-level sanitization alone is fundamentally insufficient for defending tool-enabled LLM systems and highlight the urgent need for context-aware validation across schemas, arguments, tool outputs, and accumulated conversation state. The code is available at https://github.com/liujlong27/SMT.
We give a new algorithm for finding balanced separators in minor-free and induced-minor-free graphs. A key technical idea is a new connection to online learning: we formulate a continuous, non-convex optimization problem and solve it using the Follow the Regularized Leader (FTRL) framework. Our implementation of each FTRL step yields a purely combinatorial algorithm that assigns and adjusts vertex weights over multiple rounds. For graphs with $n$ vertices and $m$ edges, we obtain the following results. (1) For $H$-induced-minor-free graphs, we find a balanced separator of size $O(\log(|V(H)|)\sqrt{|V(H)|+|E(H)|}\sqrt{m})$ in expected linear time for fixed $H$. As a corollary, we find balanced separators of size $O(\sqrt{m})$ in string graphs in expected linear time, without requiring string representations. (2) For $K_h$-minor-free graphs, we find a balanced separator of size $O(h\log h\sqrt{n})$ in expected linear time for fixed $h$. Our algorithm unifies two long lines of work on separators in minor-free graphs, dating back to Alon, Seymour, and Thomas (1990): one optimizing separator size and the other optimizing running time. It also gives a natural interpretation of the vertex weighting scheme in a recent linear-time algorithm by Bonnet, Korhonen, Le, Li, and Masarik (2026). The separator bounds for both graph classes are flow-optimal: techniques based on flow-cut duality, including ours, cannot yield balanced separators of asymptotically smaller size.
\emph{Text-to-Pattern Hamming Distances} is one of the most studied problems in string algorithms: given a text $T$ of length~$n$ and a pattern~$P$, the task is to compute the Hamming distance between~$P$ and every length-$|P|$ substring of~$T$. As exact algorithms have been stuck at time~$\tilde O(n^{3/2})$ for decades, a long line of work has turned to the $(1+ε)$-approximate version of the problem. Milestones include Karloff's $\tilde O(ε^{-2} n)$-time algorithm [IPL '93], followed by randomized improvements to $\tilde O(ε^{-1} n)$ by Kopelowitz and Porat [FOCS '15] and recently to $\tilde O(ε^{-0.915} n)$ by Chan, Jin, Vassilevska W., and Xu [FOCS '23]. In this work, we significantly improve this bound to $\tilde O(ε^{-2/3} n)$, using a considerably simpler algorithm. Our algorithm follows the framework of Chan, Jin, Vassilevska W., and Xu, which reduces the problem to a variant of 3SUM that we call \emph{Sumset Estimation}. We solve Sumset Estimation optimally (up to logarithmic factors) by replacing their matrix multiplication techniques with ideas from sparse recovery. Specifically, we adapt the classic Count Sketch by replacing its random signs with random complex roots of unity---a simple idea that may find further applications for other additive problems. Finally, we give evidence that our running time might in fact be optimal: we show that the $ε^{-2/3}$ dependence is \emph{best possible} for a certain natural class of algorithms, including ours and all previous algorithms.
AI 速览：作者通过凸对偶将硬核引理推广为基于掩蔽信道保真度的通用硬度认证框架，覆盖非布尔函数与多种损失。
We prove a broad learning-theoretic generalization of the celebrated hardcore lemma concerning the average-case complexity of computing a Boolean function (Impagliazzo, 1995). Our starting point is the observation that the classic notion of a hardcore measure, which reweights a function's input distribution, is equivalent to an input-dependent masking channel that injects randomness into the function's output distribution until it appears fully pseudorandom. The analogue of density of the measure is the fidelity of the channel. From this viewpoint, we develop a general framework for certifying hardness that allows for non-Boolean functions $f$ and a wide range of notions of hardness besides the classic 0-1 loss, by choosing the right notion of fidelity via convex duality. We show the difficulty of learning $f$ using a class $\mathcal{C}$ to minimize a convex loss is tightly characterized by the maximum fidelity possible for a masking channel, which measures the least amount of noise the channel must add to hide the structure of $f$ and make it $\mathcal{C}$-indistinguishable from random noise. Using our framework, we derive a number of multiclass generalizations of the hardcore lemma, and reveal some unexpected subtleties in going to larger alphabets. We construct a universal masking channel that applies to many natural notions of hardness simultaneously. Such channels are dual to omnipredictors, which optimize multiple losses at once (Gopalan, Kalai, Reingold, Sharan, and Wieder, 2022). We observe a symmetry between masking channels and generalizations of the hardcore lemma using pseudoentropy. We present a tight characterization of pseudoentropy using the MaxEnt algorithm from machine learning. A long line of research has shown that learning and hardness are two sides of the same coin; our framework gives a crisp and general formulation of this principle via convex duality.
Infrared and visible image fusion is a pivotal technology in low-altitude Unmanned Aerial Vehicle (UAV) reconnaissance missions, enabling robust target detection and tracking by integrating thermal saliency with environmental textures. However, the advancement of fusion algorithms is hindered by a critical evaluation bottleneck. In this paper, we identify a systematic failure in traditional no-reference metrics (specifically Statistics-based and Gradient-based metrics) within complex low-light environments, termed as ``Noise Trap''. It is mathematically proven that these metrics are positively correlated with high-frequency sensor noise, paradoxically assigning higher scores to degraded images and misguiding algorithm optimization. To resolve this dilemma, this paper proposes the Weber-inspired Local Contrast (WLC) metric. Grounded in the psychophysical principle of Weber's Law, WLC shifts the evaluation paradigm from global statistical distribution to local semantic contrast. By leveraging infrared priors, it effectively decouples target saliency from global background noise. Extensive experiments on the DroneVehicle dataset demonstrate that WLC exhibits high ``Semantic Discriminability'' in distinguishing thermal targets from background clutter. Furthermore, it achieves remarkable computational efficiency, thus, establishing itself as a reliable and real-time standard for intelligent UAV systems.
In many applications of $k$-median and $k$-means, including load balancing and redistricting, each center can serve only a bounded number of clients. This leads to capacitated clustering, in which a client may be forced to use a distant center because all nearby ones are full. This loss of locality defeats the techniques behind approximation schemes, and no constant-factor approximation is known in general metrics even for capacitated $k$-median. A long line of work therefore opens $K=k+\lfloorεk\rfloor$ centers for a fixed $ε\in(0,1]$, while respecting the capacities exactly. We give a deterministic polynomial-time algorithm for uniform soft capacities in low dimension, that is, when the clients or the center locations have bounded doubling dimension. With $K$ centers, its ratio is $3-2ε$ for $k$-median and $9-8ε$ for $k$-means, or $5-4ε$ in $\mathbb{R}^d$. Apart from $k$-median in the plane, these are the first polynomial-time ratios bounded by a constant independent of $ε$; previously, the best ratio was $O(ε^{-2}\log(1/ε))$, even for $k$-median. We also obtain ratios that stay bounded as $ε\to0$ for non-uniform and hard capacities in low dimension, and ratios polylogarithmic in $1/ε$ in general metrics. Our approach starts from a simple structural observation: a near-optimal uncapacitated solution can be split into fewer than $2k$ capacity-respecting blocks, so $2k$ centers already suffice, one per block. Our main tool, the rescue lemma, shows that once every block that sends clients to a center of the optimal solution is rescued (given a private center), that center can be closed, so each extra center effectively rescues about two blocks. This lets $\lfloorεk\rfloor$ extra centers rescue nearly an $ε$-fraction of the blocks, yielding the linear dependence on $ε$ in our ratios.
Lin Liu, Zhicheng Bao, Lu Zhang, Ziying Song, Wu Yang, Yuzheng Zhuang, Shuai Tao, Wulong Liu 等 10 位作者
Vision-Language-Action (VLA) or World Action (WAM) models have recently demonstrated remarkable performance in robotic manipulation. On LIBERO, SOTA method have achieved nearly 100\% success rates, seemingly suggesting that the models are ready for deployment in real world. However, near perfect performance on existing benchmarks can be misleading: success under ideal conditions does not imply real world robustness. Existing benchmarks primarily evaluate task completion from predefined initial states, while real world interactions inevitably involve failures such as failed grasps, collisions, and unintended object movements. A robot must therefore not only execute tasks successfully, but also recognize and recover from failures to continue the task. Yet this capability remains largely unmeasured, revealing a critical gap between benchmark performance and real world reliability. To address this gap, we introduce LIBERO-Recover Benchmark, a large scale benchmark for failure recovery in robotic manipulation. Built upon LIBERO, we collect real execution failures from SOTA embodied models and construct 1,000+ scenarios across four recovery levels: (1) Action Retry, (2) Action Adaptation, (3) Object State Recovery, and (4) Environmental Recovery. We evaluate four core capabilities: spatial understanding, object structure reasoning, interaction understanding, and topological reasoning. As the first large-scale benchmark for embodied failure recovery, LIBERO-Recover shifts evaluation from \emph{Can the robot succeed?''} to \emph{Can the robot recover after failure?''}, promoting robust and generalizable embodied agents. The project will be avaible in \textcolor{blue}{https://liulin815.github.io/LIBERO-Recovery/}.
Qinglin Yang, Jiepeng Wang, Yuxing Huang, Haibin Huang, Kai Xu, Chi Zhang, Ruizhen Hu, Xuelong Li
Recent advances in video modeling have driven substantial progress in world action models (WAMs). Existing approaches exploit pretrained video priors by decoding actions from latent representations or inferring them from generated videos. The underlying link is causal: robot actions produce changes in the scene, and video observations of these changes provide evidence for inferring the actions that produced them. However, existing training formulations do not by themselves ensure that actions and their predicted visual consequences are consistent, potentially limiting the use of pretrained capabilities. We treat video and action as paired modalities linked by this causal relationship and learn their correspondence through a unified training framework designed to better exploit pretrained video priors. Cycle WAM first augments video-action co-training with bidirectional conversion to reinforce cross-modal correspondence and recover forward and inverse dynamics. It then introduces a differentiable cycle-reconstruction objective to further couple the two mappings. The resulting model supports both forward and inverse dynamics, with the potential to improve performance through more effective use of pretrained video priors. Cycle WAM achieves state-of-the-art performance among comparable methods on LIBERO-Pro and substantially improves DreamZero on DROID. Comparisons with UWM characterize its inverse-dynamics performance and show sharper visual detail in the displayed forward-dynamics predictions.
Yulong Li, Rong Xia, Yuxuan Zhang, Jianxu Chen, Xiwei Liu, Haochen Xue, Maosheng Li, Yuhang Liu 等 15 位作者
We introduce DoAtlas-2, a foundation for self-evolving causal biomedical discovery that organizes knowledge around causal mechanisms and advances through external evidence from human populations. DoAtlas-2 integrates 771 research resources covering more than 720,000 participants in 48 countries, from longitudinal clinical phenotypes, medical imaging, and continuous physiological signals to eight molecular layers, together with an evidence network of approximately 4.7 million literature-derived records over 93,566 concepts and 149,383 candidate causal relations. DoAtlas-2 autonomously formulates research questions from evidence gaps and unresolved mechanisms, prespecifies their causal designs, and generates validated analyses. Supporting, challenging, and unresolved results continuously revise mechanistic interpretations, the causal evidence state, and the discovery frontier, so that DoAtlas-2 self-evolves within a closed loop of hypothesis generation, empirical testing, and renewed discovery. DoAtlas-2 has systematically evaluated 2,031 research questions. In the Human Phenotype Project (HPP), it formulated 4,014 candidate pathway questions across vascular, early-glycemic, and hepatic-metabolic systems, and screening of the first 1,079 yielded statistical support for 756. Representative studies identify blood pressure as a convergence node linking adiposity, hepatic, and lipid phenotypes to vascular outcomes, and show that an adiposity-inflammation-blood-pressure pathway is largely attenuated by joint adjustment for body mass index (BMI) and smoking. The discovered vascular network constitutes a completely interpretable predictive foundation, admitting exact attribution of every prediction and closed-form mediation effects. DoAtlas-2 thereby unifies causal mechanism discovery, population-evidence testing, and interpretable prediction within one continuously evolving foundation.
Jianglong Liu, Zhicheng Zhu, Jia Hao, Liangyue Jia, Wenxuan He
Neural operators offer an efficient data-driven approach to predicting complex automotive aerodynamic flows. However, spectral bias limits their ability to reconstruct fine-scale flow variations around geometric features such as side mirrors and wheel arches. Reliable assessment of local geometric changes therefore requires models that capture fine spatial detail together with global flow interactions. Inspired by iterative refinement and state evolution, we propose CIS-Transolver, a cross-layer incremental-state neural operator that integrates local multiscale information into global flow modeling. The method extracts local feature variations and encodes them as zeroth- to second-order geometric moments. These encodings are combined with information from earlier layers to form a cross-layer memory. Memory-derived corrections enter the value branch of Physics-Attention, allowing local details to inform global physical-state interactions throughout the network. We systematically evaluate CIS-Transolver on three public automotive aerodynamics benchmarks, ShapeNet-Car, AhmedML, and DrivAerNet++, covering surface pressure fields and volumetric pressure and velocity fields. CIS-Transolver achieves the lowest mean relative L2 error among the evaluated methods for all five prediction metrics across the three benchmarks. Relative error reductions average 8.72% compared with the best baseline for each metric. Also achieves the lowest high-frequency spectral errors across all five metrics, with an average relative reduction of 16.06% compared with the best baseline. These results support cross-layer reuse of local information for fine-scale automotive flow prediction.
Aggregate scores for continual learning algorithms rest on a distribution over task orders that is rarely stated. We formulate algorithm comparison as an expectation over a declared order distribution, which separates training variation from order variation, and apply it to motion forecasting with four road-layout tasks from the INTERACTION dataset, all 24 task orders, four algorithms, and three training replicates (288 runs). Task order accounts for 65\% to 91\% of the variation in current-task error and backward transfer, and 90\% to 99\% of this order effect comes from the identity of the starting task: the smallest task cannot be learned from a cold start. Beyond the first stage, the order effect is comparable to training variability. In most orders, the leading algorithm in current-task error and backward transfer is within one replicate standard deviation of the runner-up. Backward transfer depends on the order mainly through its reference term, which weights the first task most heavily, so relative improvement and final error identify different leading algorithms for displacement error. A transition-direction descriptor computed from task statistics tracks the position of the smallest task, and weighting orders by it leaves the algorithm ranking unchanged. These results call for benchmarks that declare their order distribution and report order-resolved, starting-task-resolved, warm-start, and final errors alongside aggregate scores. The code and data are available \href{https://anonymous.4open.science/r/task-order-cl-4f7b/}{here}.
Yueh-Hua Wu, Kei Ota, Hanlong Li, Hayata Yuasa, Jason Wang, Kengo Ikeuchi, Kensuke Chiyoma, Makoto Sugiura 等 20 位作者
AIRoA-0 (A-0) is a bimanual manipulation system learned from 16,151 hours of handheld demonstrations, with no robot data collected at any stage. The demonstrations are recorded with YUBI, our open-source handheld gripper, which lets people record manipulation data anywhere with no robot present. From a single instruction, A-0 sorts more than a dozen small parts into their numbered cells and changes the mop pads and brush of a robot vacuum; among other tasks, it unzips a pencil case and folds a towel. It performs these tasks using only a camera on each wrist, with no head or scene camera, at a site where none of its training data was recorded. Across the ten tasks of our evaluation, it averages a success rate of 78%, against 27% for the strongest of four other vision-language-action models post-trained on the same task demonstrations. We collect no teleoperation data and use no reinforcement learning or on-robot corrections, and the robot appears only at test time. The system learns which regions of each wrist view matter for the task, predicts the stage the task has reached, and decodes actions as a continuous stream. A-0 is embodiment-agnostic: it runs on robots that mount the YUBI gripper, such as a bimanual Franka, TX-G2, OpenArm and UR. We are releasing the dataset, to our knowledge the largest open dataset of handheld-gripper demonstrations with task and subtask instructions, together with the pre-trained weights and the post-training recipe. Videos of A-0 on every task are on the project page, https://a-0.airoa.io. Anyone can build the gripper, record a task of their own and post-train the model on it in under four hours on a single node of eight H200 GPUs.
Video dubbing aims to synthesize lip movements that match target speech while preserving the speaker identity, head pose, and background of a source video. Existing editors are predominantly trained via self-reconstruction under original paired audio, even though the source mouth becomes a conflicting target when driving speech is replaced. We introduce replacement-aware cross-audio supervision: paired-audio samples use full-frame reconstruction, whereas unpaired replacement-audio samples mask out source-mouth reconstruction losses and supervise the mouth solely through audio-visual synchronization while preserving unmasked facial regions and background. Because offline dubbing provides the complete source video and target speech beforehand, we present VidDub, a full-context diffusion editor distilled into two denoising steps (2 NFE). Our replacement-aware teacher first learns accurate lip synchronization without regressing toward conflicting source mouth movements. Matched trajectory regression then transfers the teacher's CFG-guided trajectories to a two-step student, and distribution matching distillation recovers visual quality after step compression. We will release the curated training dataset, VidDub-Bench, fixed paired- and cross-audio evaluation manifests, preprocessing outputs, and unified evaluation code. Across three training seeds, masking out conflicting source-mouth reconstruction improves cross-audio Sync-C by 0.833; the final two-step model combines strong lip synchronization with superior visual quality, identity preservation, and visual speech recognition across four benchmarks.
Yang Qu, Qiang Fu, Zonghao Ying, Jiayi Zhao, Yaoyuan Zhang, Jinyang Guo, Jian Yang, Aishan Liu 等 10 位作者
Embodied agents are increasingly deployed in open-world environments, yet current robustness evaluations remain biased toward hazardous instructions and jailbreak vulnerabilities. A more fundamental issue is largely overlooked: agents are rarely tested under semantically sufficient but contextually minimal instructions (i.e., common in real-world human communication), where auxiliary contextual cues are absent. For the first, we identify a novel phenomenon termed Ambient Context Illusion, where agents fail to execute tasks correctly when non-essential contextual details are omitted, despite unchanged task semantics. To systematically study this, we establish VagueBench, a benchmark built upon the AI2-THOR interactive simulation environment for the evaluation of context dependency in embodied agents. We systematically consider key semantic components of instructions (entities, procedures, constraints, and prior knowledge) and derive a unified taxonomy with four categories and nine subtypes to reduce auxiliary information. VagueBench contains 1,303 instruction pairs across diverse indoor environments, constructed via a human-agent co-generation framework that integrates semantic redundancy estimation, taxonomy-guided transformation, and rigorous semantic-invariance verification. Through extensive evaluation on 6 state-of-the-art embodied agents, we observe significant degradation on VagueBench in both task performance and executability (relatively -19.65% in task success rate and relatively +54.10% in empty plan rate) revealing a strong reliance on auxiliary linguistic signals rather than core task semantics. Furthermore, we demonstrate that this phenomenon persists in real-world robotic systems, where it can lead to incorrect decisions and potential safety risks.
On-policy distillation (OPD) mitigates the student's exposure bias by optimizing on student-generated trajectories. However, student-generated prefixes may contain reasoning errors or follow patterns unfamiliar to the teacher, undermining the reliability of its token-level supervision. This degradation becomes more pronounced as prefixes grow longer. To address this teacher-side distribution mismatch, we propose Student-Prefix Teacher Adaptation (SPTA), which adapts the teacher to the states on which it is queried during distillation. Specifically, before OPD starts, we collect student rollout prefixes with random lengths and continually train the teacher to continue from the prefixes using reinforcement learning with verifiable rewards (RLVR). By treating each prefix as context and rewarding correct final answers, the teacher learns to recover from student-induced states rather than imitate the student's reasoning. Then, the adapted teacher is used for standard OPD, with no changes to the distillation objective. Experiments on mathematical reasoning benchmarks show that our SPTA improves average student performance over standard OPD and OPD with task-level RLVR-adapted teachers. These results highlight the importance of adapting teachers not only to solve problems independently, but also to provide reliable guidance on the states their students actually visit.
Xin Zhang, Yabo Chen, Yijie Fang, Wanying Qu, Haibin Huang, Chi Zhang, Feng Xu, Xuelong Li
Recent generative video models achieve impressive visual quality but remain constrained by limited physical consistency and controllability. Existing video generation methods provide minimal physical control, and single-image-to-3D conversion approaches often suffer from object interpenetration. Furthermore, physics-based scene-level 3D generation methods exhibit spatial misalignment, stylized artifacts, and inconsistencies with the input data, restricting their use in realistic interactive video synthesis. We propose PhysOmni, a training-free framework that converts a single image into a physically consistent and controllable video through holistic scene-level 3D reconstruction. By rep?resenting the full scene geometry in a unified spatial coordinate system, PhysOmni resolves object penetration and alignment ambiguity. Unlike prior methods, this formulation enables accurate scene?level multi-object interactions and introduces richer, complex control types for advanced mechanics?based manipulation. By decoupling simulation from rendering, PhysOmni bypasses latency-heavy priors, achieving real-time physical interaction previews paired while preserving photorealistic visual fidelity. Experimental results demonstrate that PhysOmni substantially outperforms prior methods in physical fidelity, spatial coherence, and controllability. Project Page: https://zephipage.github.io/PhysOmni/
MIT Hardness Group, Josh Brunner, Erik D. Demaine, Fabian Frei, Timothy Gomez, Markus Hecher, Jeffery Li, Jayson Lynch
Impagliazzo and Paturi's 1999 Exponential Time Hypothesis (ETH) is false, but Stearns and Hunt's 1990 Satisfiability Hypothesis (SH) remains plausible. We prove that SH is equivalent to the recently proposed Weak ETH which rules out $O(2^{n^{1-\varepsilon}})L^{O(1)}$-time 3SAT algorithms for every constant $\varepsilon>0$, where $L$ is the input length. We show that SH (or Weak ETH) preserves the typical process for lower bounds: a polynomial-time reduction from a 3SAT formula of size $s=Θ(n+m)$ to an instance of size $N\le s^{b+o(1)}$, for fixed blowup exponent $b>0$, rules out $O(2^{N^{1/b-\varepsilon}})$-time algorithms for every constant $\varepsilon>0$. By comparison, for an $N=O(s^b)$ blowup, ETH would have ruled out $2^{o(N^{1/b})}$-time algorithms. The framework also yields parameterized lower bounds ruling out running times of the form $O(2^{k^{1/b-\varepsilon}})L^{O(1)}$. We also prove equivalence of the analogous counting hypotheses, #SH and Weak #ETH, and that they rule out $O(2^{(n+m)^{1-\varepsilon}})$-time algorithms for #2SAT for every constant $\varepsilon>0$.
Xiao Yang, Aishan Liu, Siyuan Liang, Zonghao Ying, Le Wang, Jiakai Wang, Xianglong Liu, Dacheng Tao
Large language model (LLM) agents increasingly operate in open-world environments where safety risks cannot be fully predefined. However, most existing guardrail mechanisms remain static, relying on fixed rules, prompts, or classifiers that generalize poorly to emerging risks. Here we present EVOLVE, a tool-level self-evolving defense framework that formulates agent security as a process of cumulative security evolution. Inspired by how human societies transform experiences into transferable protective knowledge, EVOLVE externalizes security knowledge into executable guardrail tools that are continuously inherited, independently audited, and iteratively refined through lifelong interaction. Across multiple benchmarks, backbone models, and embodied environments, EVOLVE consistently outperforms existing defenses across diverse risk scenarios with +22\% average F1 score. We further validate the framework in real-world robotic and autonomous agent systems. Our results suggest that scalable agent safety may require continuously evolving defense systems rather than static guardrails.
Li Kong, Zilong Liu, Haoyang Zhou, Hanbing Liu, Qi Qi
Neural auction design has emerged as a powerful paradigm within automated mechanism design for modern Web platforms. Because these learned mechanisms directly dictate critical economic outcomes, their lack of internal transparency raises significant auditability and trust concerns. Opening this black box \textit{post-hoc} is therefore essential, yet presents two distinctive challenges: (1) activation contexts are dense, correlated numerical profiles devoid of explicit semantic structure, and (2) feature interpretations can only be falsifiably audited if they map to measurable auction states or outcomes. In this paper, we propose a post-hoc white-box auditing framework for unpacking trained neural auctions. Using CITransNet as a representative testbed, we first employ sparse autoencoders to decompose hidden representations while faithfully preserving downstream auction behavior. Anchoring feature interpretations to native economic targets, we validate them through activation prediction and causal interventions to isolate verified features from the reconstruction dictionary. Finally, applying this pipeline across varied auction settings, we show that the verified features can serve as analytical instruments to examine how internal representations modulate strategic misreporting incentives, align with Myerson's optimal auction, and support the diagnosis of individual auctions. Our work highlights the potential of white-box mechanism auditing to build the foundational trust necessary for neural auctions to be deployed as widely adopted Web technologies. To enhance research transparency and facilitate independent inspection, we publicly release the experimental results supporting our findings at https://github.com/Li-Beverly-Kong/opening-neural-auctions.git
Experimental design is critical for evidence-based decision-making in healthcare, marketing, and public policy. However, designing efficient experiments across heterogeneous subgroups presents significant challenges. Existing methods often optimize for statistical power or overall sample efficiency, overlooking crucial fairness considerations across these different subgroups. To address this gap, we introduce a Fairness-Aware Contextual Track-and-Stop Design (F-CTSD) algorithm. The proposed F-CTSD algorithm provides statistical guarantees on subgroup fairness while minimizing required sample sizes. We quantify the fairness-efficiency trade-off and derive the exact sample complexity for the proposed F-CTSD algorithm under its fairness constraints. We further theoretically prove that the proposed F-CTSD algorithm consistently produces accurate treatment effect estimates even under fairness requirements, enhancing statistical reliability. Numerical experiments show that the proposed F-CTSD algorithm outperforms existing methods, achieving higher sample efficiency while reducing subgroup fairness violations by 4.95%.
Language agents read code, tool output, and dialogue in contexts that grow at every step. Dynamic sparse attention lets a small indexer choose the earlier tokens each query attends to, but the indexer still scores the whole prefix at every layer. Cross-layer reuse methods such as IndexCache rerun the indexer at only some layers, with one fixed pattern for every input. In an agent, the relevant evidence can change abruptly when a traceback or test result arrives, so fresh selections matter more at some steps than at others. We propose Adaptive Index Reuse (AIR), which keeps IndexCache's execution and learns, for each segment of an agent episode, how many layers refresh their selection. A small controller reads signals from completed computation and the latest tool event, chooses a pattern from a nested library, and learns from terminal task reward under an expected compute budget. On Qwen3-4B-Instruct-2507 with a 128K context, AIR improves success over the strongest static schedule at matched cost on SWE-bench Verified (27.0% to 32.4%), Terminal-Bench 2.0, and three $τ^2$-bench domains, and it lowers time per successful task on all five. Across 50 backbone, context, and benchmark settings, it matches or exceeds that schedule in 44.
Yiqi Nie, Fei Wang, Junjie Chen, Kun Li, Yudi Cai, Dan Guo, Chenglong Li, Meng Wang
Multimodal Emotional Reappraisal (MER) aims to shift affective interpretation by rewriting and editing a situation without altering its underlying semantic context. Despite advances in controllable editing and sentiment transfer, MER remains challenging under coupled multimodal constraints, especially in internet memes where images, embedded text, and layout jointly encode affective meaning. To bridge this gap, we present MER-Bench, a systematic benchmark for MER on image-text memes, containing over 3,000 real-world memes with fine-grained annotations for emotion transformation and structure-preserving multimodal editing. MER-Bench spans diverse visual, textual, and structural settings and is paired with a structured MLLM-as-a-Judge evaluation suite, comprising eight fine-grained metrics and three aggregated objectives for emotional alignment, content fidelity, and transformation strength. Extensive evaluation of 14 advanced multimodal models reveals that: (1) current models struggle to jointly balance affect, semantics, structure, and text; (2) strong visual editing does not imply effective MER, with poor embedded text rendering and affect-aware rewriting remaining key bottlenecks. Our results identify MER as a critical capability gap in current multimodal generation systems and highlight affective image-text reasoning as a key challenge for future work. Project Page: https://github.com/one-seven17/MER-Bench