Computer Vision
Plan-Bench: Can Existing VLMs Go Well Beyond Perception to Reason and Plan in Embodied Worlds?
Juntong Feng, Qingan Liu, Jingru Dong et al. (+8 more)
Vision-
Language Models (VLMs) have achieved strong performance in visual perception and spatial
reasoning, leading to growing interest in using them as high-level planners for embodied AI. However, physical tasks require models to integrate spatial information, reason over temporal states, and generate multi-step plans under various constraints, raising the question of whether current VLMs can reliably support such
reasoning and planning independently of execution. We introduce Plan-Bench, a
benchmark for evaluating the high-level
reasoning and multi-step planning capabilities of VLMs relevant to embodied worlds, while isolating the influence of action execution. It covers three complementary dimensions, spanning 16 subclasses and 699 curated samples. Extensive evaluation of 28 VLMs reveals room for improvement, with even the strongest model achieving only 75.0% accuracy. We further find that model scaling generally improves these capabilities. Ablation studies also indicate larger models show increasing potential as high-level planners. Overall, Plan-Bench provides a lightweight probe for evaluating VLMs'
reasoning and planning capabilities relevant to embodied environments and assessing their potential as decision makers for physical tasks. Our
benchmark and code are available at https://github.com/MinneFeng/Plan-Bench.
Read the full abstract ▾
All authors Juntong Feng, Qingan Liu, Jingru Dong, Yue Zhu, Weiming Zheng, Yujie Cheng, Xinzhe Han, Shuai Hao, Yunzhi Zhuge, Haiwen Diao, Huchuan Lu
Robotics
RoboSimStudio: An Interactive Toolkit for Human and Agent Rapid Real2Sim of Robot Manipulation Tasks
Kaiyuan Chen, Shuangyu Xie, Uksang Yoo et al. (+9 more)
It has been difficult to accurately model the physics of the interaction between a
robot and the objects it manipulates, including contact mechanics, coupled contact dynamics, deformations, stiffness and friction. We present RoboSimStudio (RSS), an interactive toolkit that makes choices of solver, contact model, and physical parameters explicit and editable, allowing both humans and coding
agents to quickly create and iteratively refine simulations based on observed physical behavior. We implement RSS with the open Newton physics simulation system and describe experiments with human-guided real2sim using (1) visual evaluation, (2) force-feedback and (3) agentic Real2Sim2Real. Code,
video and additional information can be found on the project website: https://robosimstudio.github.io
Read the full abstract ▾
All authors Kaiyuan Chen, Shuangyu Xie, Uksang Yoo, Xinyue Zhu, Cael Magner, Alvin Zhu, Daehwa Kim, Eric Heiden, Erwin Coumans, Miles Macklin, Spencer Huang, Ken Goldberg
Computer Vision
Game2Real-Bench: Do Toy Games Reveal and Transfer to Real-World Visual Reasoning?
Qingan Liu, Juntong Feng, Yue Zhu et al. (+6 more)
Visual
reasoning is an important capability for
vision-
language models (VLMs) to understand visual evidence, spatial structure, constraints, and state changes. Existing evaluations mainly rely on real-world data or procedurally generated tasks. Real-world data preserve natural complexity, while procedural games provide controllable generation and verifiable solutions. However, it remains unclear whether abilities measured in toy games correspond to real-world visual
reasoning and whether game examples can support transfer. We introduce Game2Real-Bench, a
benchmark that studies these connections across four capability groups, including fine-grained perception, spatial
reasoning, constraint
reasoning, and sequential planning. We construct 1,000 procedurally generated game samples with executable verification and curate 400 real-world samples aligned by shared
reasoning requirements rather than visual similarity. We evaluate 25 representative VLMs and observe strong game-to-real consistency, with Pearson and Spearman correlations of 0.935. We further study one-shot game-to-real transfer on four models. Constraint
reasoning improves for several models, while fine-grained perception declines with game demonstrations. We also construct Game2Real-
Dataset with nearly 24K verified game samples across 18 tasks and 65 QA families. Our
benchmark,
dataset, and code are available at https://github.com/QingAnLiu/Game2Real-Bench.
Read the full abstract ▾
All authors Qingan Liu, Juntong Feng, Yue Zhu, Yujie Cheng, Xinzhe Han, Shuai Hao, Yunzhi Zhuge, Haiwen Diao, Huchuan Lu
Cryptography and Security
AegisAsync: Symbolic Async-Event Simulation for Risks in Python Asynchronous Programs
Baihong Chen and Wen Li
Security and robustness risks in Python async programs depend on event order, object identity, and execution after exceptions or cancellation. Detecting these risks requires connecting related operations across function calls and tracking their effects along execution paths. We present AegisAsync, a bytecode-based analysis of Python async risks for CPython 3.10-3.14. Its Async Event Transition Graph combines bytecode control flow, resolved calls, async events, and symbolic object relations. Bounded path-local simulation tracks event order and object states and explores cancellation at await suspensions. Six detector plugins evaluate the resulting traces and states to produce candidates with supporting witnesses. On AsyncBench, our
benchmark of 180 labeled risks, AegisAsync detects all labeled risks with zero false positives on each of the five versions. Across five path budgets on 2,500 PyPI package versions, source review confirms 457 of 910 AegisAsync candidate records (50.2%), representing 457 unique bugs; across three flake8-async samples, 70 of 2,187 review judgments are confirmed (3.2%). By submission, 166 PoCs reproduce AegisAsync findings, and we have reported 157 bugs to developers, of which 16 have been confirmed and fixed.
Read the full abstract ▾
Computer Vision
OmniCompass: A Holistic Benchmark for Omni-Modal Video Grounding and Understanding
Xuyang Liu, Shihao Wang, Xinyu Wei et al. (+5 more)
With the rapid development of
multimodal large
language models (MLLMs), evaluating their
video understanding capabilities has become increasingly important. Existing
video understanding
benchmarks cover only part of the evaluation dimensions, limiting systematic diagnosis of omni-modal
video understanding. We introduce OmniCompass, a holistic
benchmark for omni-modal
video grounding and understanding. OmniCompass defines two evaluation tracks: OmniCompass-VL assesses
vision-
language models on tasks requiring visual evidence, while OmniCompass-AV assesses omni-modal models on tasks requiring joint audio-visual evidence. Each track includes three complementary tasks: temporal grounding (TG), spatial grounding (SG), and question answering (QA), assessing when a referenced target appears, where it is, and what happens. QA covers five perception and three
reasoning categories in OmniCompass-VL, and three perception and two
reasoning categories in OmniCompass-AV. OmniCompass-VL contains 820
videos and 2,460 instances across domains and three duration types: short, medium, and long. A narrative- and audio-rich subset forms OmniCompass-AV with 240
videos and 720 instances. Each instance is evaluated under two paired settings: text-only and text+image, specifying the same target through text and visual-
language guidance. All instances are manually annotated with multi-round quality verification. Evaluations of 50 MLLMs show that open-source models still lag behind proprietary counterparts, particularly in audio-visual understanding. Precise temporal and spatial grounding remains challenging even for leading models, and their predictions still exhibit inconsistencies across the paired target specifications.
Read the full abstract ▾
All authors Xuyang Liu, Shihao Wang, Xinyu Wei, Minghan Li, Zeqing Wang, Zuhao Yang, Zichen Wen, Lei Zhang
Data Structures and Algorithms
Settling the Approximability of Submodular Facility Location
Adam N. Elmachtoub, Bento Natura and Meike Neuwohner
In submodular facility location (SFL), introduced by Svitkina and Tardos (SODA 2006), clients $\cal C$ in a metric space are assigned to facilities $\cal F$. A facility serving a set $S \subseteq \cal C$ of clients incurs cost $g(S)$, where $g$ is a monotone submodular function common to all facilities. The best known approximation factor for SFL is $O(\log\log|\cal C|)$, due to Abbasi, Adamczyk, Bosch-Calvo, Byrka, Grandoni, Sornat, and Tinguely (ICALP 2024), improving on the $O(\log|\cal C|)$-approximation of Svitkina and Tardos. We show that this is best possible up to a constant factor in the value-oracle model by showing that there exists no
algorithm with only polynomially many queries to $g$ with approximation ratio $o(\log\log|\cal C|)$. This bound holds for
algorithms with unlimited computation time, and when $g$ is restricted to matroid rank functions. We also show that the natural linear programming relaxation has an integrality gap of $\Theta(\log\log|\cal C|)$. Our lower bound follows by constructing an extension of Vondrák's
symmetry-gap approach (FOCS 2009) for matroid rank functions.
Read the full abstract ▾
Information Theory · Cross-listed from quant-ph
Gaussian entanglement and quantum communication: geodesic optimization and multimode bounds
Hemant K. Mishra and Mark M. Wilde
Loss and noise limit the use of light to distribute
entanglement, transmit
quantum information, and establish secret keys. Gaussian states have a compact description in terms of means and covariance
matrices, but their
entanglement still requires global optimization. We develop a framework for computing Gaussian-restricted versions of the relative
entropy of
entanglement and Rains
entanglement. An integral representation reveals a convex geometry under
matrix geometric means and reduces Gaussian Rains
entanglement to relative
entropy minimized over Gaussian states with positive partial transpose. We obtain unique optimal reference states and additivity for independent Gaussian input states. Our convex-majorization
algorithm has an explicit
geometric convergence rate, uniform in the faithful regularization parameter and valid for pure input states. Controlled numerical integration and approximate convex subproblem solutions give a bit-complexity bound that is
polynomial in accuracy bits for each fixed rational input covariance
matrix, with explicit dependence on covariance size. These calculations bound extractable
entanglement and, through the relative
entropy of
entanglement, extractable secret key. Teleportation simulation, energy allocation, and a convex optimization of passive-basis loss cores give bounds on
quantum and private communication under an input-energy constraint, allowing arbitrary two-way classically assisted protocols. Two- and three-mode examples compare the bounds with achievable rates, including matching pure-loss
benchmarks. Finally, we separate the Gaussian-restricted quantities from their unrestricted counterparts and give a
photon-number truncation method with controlled error for computing the latter.
Read the full abstract ▾
Robotics
WorldOracle: Benchmarking Action-Conditioned World Models against Real-World Executions
Haoyi Duan, Liliang Chen, Nan Wang et al. (+13 more)
We introduce WorldOracle, a
benchmark specifically designed to evaluate action-conditioned world models (ACWMs) against paired real-world executions. Given the same initial observation and low-level action sequence, WorldOracle asks whether a predicted rollout reproduces what happens in the real world, including physical consequences and task-state transitions. We evaluate models along four complementary dimensions: physical state correctness, interaction spatial
alignment, key interaction area fidelity, and multi-view action coherence. To support this evaluation, we curate real-world training and evaluation suites that cover different interaction dynamics and embodiments. The evaluation suite contains long, multi-segment episodes with both successful and failed executions, enabling evaluation across varied action outcomes and behavior distributions. Experiments on seven representative ACWMs reveal distinct model strengths and failure modes. WorldOracle provides a systematic foundation for measuring and advancing action-conditioned world modeling for manipulation. We will release the
datasets, evaluation code, and leaderboard.
Read the full abstract ▾
All authors Haoyi Duan, Liliang Chen, Nan Wang, Yang Zhang, Yuhang Lu, Xiaotong Chen, Wei-Cheng Tseng, Shenyuan Gao, Zhaoshuo Li, Jingyi Jin, Pooya Jannaty, Dhruv Shah, Yuke Zhu, Ming-Yu Liu, Maoqing Yao, Guanghui Ren
Machine Learning · Cross-listed from stat.ML
Joint Neural Posterior Estimation and Model Selection for Chemical Reaction Networks
J. Arturo Esquivel, Jorge A. Campos-Gonzalez-Angulo, Sergio Pablo-García et al. (+2 more)
Simulation-based
Bayesian inference for Chemical Reaction Networks (CRNs) must simultaneously solve two computationally demanding problems: discriminating between overlapping reaction pathways and estimating reaction rate constants. Classical approaches, such as Sequential Monte Carlo ABC, rely on handcrafted summary statistics and scale poorly across large parameter spaces. To overcome this, we introduce a neural inference architecture. Our pipeline utilizes shared 1D-CNNs and a permutation-invariant Set
Attention mechanism to learn low-dimensional summary statistics from variable-length, multi-experiment time-series data. These statistics simultaneously condition a model selection network to compute posterior model probabilities and model-specific conditional Invertible Neural Networks (cINNs) that estimate parameter posterior distributions. We demonstrate the practical utility of our approach on a real protodeboronation example, illustrating how it handles structural ambiguity and extracts kinetic parameters from competing catalyst deactivation pathways.
Read the full abstract ▾
All authors J. Arturo Esquivel, Jorge A. Campos-Gonzalez-Angulo, Sergio Pablo-García, Alán Aspuru-Guzik, Vianey Leos-Barajas
Robotics
EgoMatrix: Physically Faithful Dexterous Robot Demonstrations from Egocentric Human Videos
Ziheng Wang, Yanming Shao, Ting Mao et al. (+4 more)
Egocentric human
videos offer a scalable alternative to
robot teleoperation for collecting dexterous manipulation data. However, two challenges limit their use as reliable training data: accurate hand-object interaction (HOI) reconstruction from monocular RGB and high-quality
robot demonstrations, especially visual observations. We present EgoMatrix, an end-to-end data engine that converts monocular egocentric human
videos into bimanual dexterous trajectories and paired
robot observation
videos, with the trajectories directly replayable on real
robots. It reconstructs HOI with emphasis on relative geometry and contact rather than exact metric scene recovery, then uses retargeting and reinforcement learning (RL) to reproduce the demonstrated tasks in simulation. We jointly render
robot hands and objects from these simulated trajectories to produce physically faithful observations synchronized with state and control records. Rendering the same trajectories in diverse, task-compatible backgrounds expands paired training data while preserving
robot-object motion and contact. Across TACO, OakInk2, and self-collected data, EgoMatrix achieves 54.84% end-to-end success in simulation versus 14.52% for the strongest baseline. Reconstruction reaches state-of-the-art performance on most metrics of the two public
benchmarks. Qualitative results show more faithful hand-object interactions than prior methods, while feature-space analysis shows closer
alignment with real-
robot observations than GPT-6 Astra. Our project page is available at https://spikew726.github.io/EgoMatrix/.
Read the full abstract ▾
All authors Ziheng Wang, Yanming Shao, Ting Mao, Kaihan Chen, Yiqun Wang, Xuanye Wu, Yao Mu
Artificial Intelligence
Read the Room Before Your Model Generates
Muyao Kong and Zhihao Lin
Large
language models (LLMs) generally rely on sampling temperature to regulate the amount of randomness in their outputs. Mainstream methods for selecting temperature typically involve fixed, heuristic global settings and context or
token-level calibration, rather than a geometrically interpretable representation of the communicative scenario. In this work, we introduce a simple and interpretable temperature selection framework, the Scenario-Actor, which adapts decoding to the inferred situational characteristics. Scenario-Actor extracts scenario-relevant structures from
Transformer hidden states and determines the temperature from scenario
embedding, a structured
geometric representation of the communicative setting. Empirical results show that on Llama-3.1-8B-Instruct, Scenario-Actor achieves strong goal-completion performance in interactive social scenarios, significant gains on open-ended instruction following, and competitive performance on multi-turn conversation compared to multiple baselines.
Read the full abstract ▾
Computer Vision
DuplexBridge: Coordinating Understanding and Generation for Unified Image Editing
Yidi Liu, Jie Xiao, Chaojie Mao et al. (+2 more)
Complex image editing requires faithful task interpretation and accurate execution, but a failed edit does not reveal which responsibility failed. We present DuplexBridge, which uses a unified model's own understanding capability to enhance
prompts for its own generation capability, without an external understanding module. A source-grounded bridge specifies targets, final-state changes, and preservation constraints. Understanding is supervised on fidelity to the original request without observing the generated image; generation is supervised on execution of the condition it actually receives. Reciprocal trust checks selectively admit global task feedback to each role while retaining independently valid local supervision. We introduce CE-Bench, a 1,200-task
benchmark measuring joint satisfaction of task-derived change and preservation requirements. Across three model configurations, the full system improves complete compliance over direct editors by 21.75-25.75 percentage points, with gains also observed on KRIS-Bench, RISEBench, ImgEdit, and GEdit-Bench. From a common supervised Wan-Image initialization, gated joint training reaches 77.08% complete compliance, compared with 70.92% without gates and 75.00% when generation-local supervision instead checks the original task with both gates retained. These results support distinct local responsibilities and selective shared feedback for joint understanding-generation adaptation.
Read the full abstract ▾
All authors Yidi Liu, Jie Xiao, Chaojie Mao, Jingfeng Zhang, Xueyang Fu
Computation and Language
Back-Translation Still Matters: Leveraging Monolingual Data for Low-Resource Machine Translation with LLMs
Divyanshu Bathla, Parag Singla and Jitin Singla
Pre-trained Large
Language Model (LLMs) are effective on many tasks in zero-shot and few-shot settings. For machine translation of low-resource
languages, the obvious options are few-shot prompting and fine-tuning on a small parallel set. We ask: what if moderate amounts of low-resource monolingual data are also available? We argue that, rather than continued pretraining, a more effective use of this data is the pre-LLM era technique of back-translation (BT). The current fine-tuned LLM translates monolingual sentences in each
language, and the resulting pairs train it in the reverse direction. Updates on the parallel set are interleaved for stable learning. With 1K parallel pairs on Odia and Swahili, BT beats continued pretraining into the low-resource
language by 7.7 to 16.4 XCOMET-XXL points in domain. With beam-decoded intermediate translations, it also beats a model with 10 time more parallel data. In-context examples retrieved from the parallel data further improve BT. We also analyze the source, quality and amount of monolingual text. Across five low-resource
languages, with Hindi as a high-resource control, and two 3B backbones, BT improves translation into the low-resource
language by 16 to 173% XCOMET-XXL over parallel-only fine-tuning. Blind human evaluation on Odia and Sanskrit agrees with the automatic metrics.
Read the full abstract ▾
Artificial Intelligence
Self-Evolving Agent Harness
Chuyang Wei, Haoxiang Guan, Minghang Zhu et al. (+5 more)
Harness self-
evolution revises an
agent's software scaffolding, including
prompts,
memory, workflows, and executable components, while keeping model weights fixed. We formalize a feedback-driven process in which accepted revisions persist across runs. The framework separates update targets from proposal and admission mechanisms. It distinguishes changes to interpreted state from revisions to tools and code, repository-level control logic, and plugin composition. We characterize search through evolutionary optimization and examine its reliability through continual PAC-learning. Under explicit assumptions, the analysis yields an expected generalization bound, a retention bound conditional on historical regression budgets, and a high-probability admission rule using fresh, independent validation data. We examine how persistent harness revisions affect knowledge sharing and organizational workflows, with implications for scientific reliability, physical safety, and institutional accountability. Cases from prior work illustrate model–harness co-design, distributed execution, and plugin reuse. Reliable deployment requires continual validation as task distributions and evaluators change, with independent safeguards for learning, runtime privileges, and consequential actions. We provide a comprehensive 66-page discussion of related work, theory,
algorithms, evaluation, applications, and the broader impact of Harness RSI, together with continuously updated code and supplementary materials, at https://github.com/zgcagi/Harness-RSI.
Read the full abstract ▾
All authors Chuyang Wei, Haoxiang Guan, Minghang Zhu, Kexin Zhou, Jiyan He, Xiren Zhou, Huanhuan Chen, Yifei Shen
Game Theory
The Geography of (Concurrent) Block Proposing
Ciamac C. Moallemi, Mallesh M. Pai, Dan Robinson et al. (+1 more)
The resilience and censorship resistance of a blockchain depend on where the validators and proposers of that blockchain physically operate. We study the economic incentives behind this location choice, both with a single proposer and with multiple concurrent proposers. Formally, we study a game where proposers first choose locations and then compete in a first-price auction for an opportunity whose common value is driven by information from geographically dispersed sources. A single proposer minimizes variance-weighted delay, locating at a weighted median of the sources and co-locating with any dominant one. By contrast, with two proposers, information both observe equally early earns no rent, and each proposer's expected profit is proportional to the standard deviation of the information it receives before its rival. We show that the location game admits the jointly unobserved variance as an ordinal potential, so global solutions of a weighted two-median problem are
equilibria. With two sources, one proposer locates at each, however weak one is. With more proposers and at least as many sources,
equilibrium locations are distinct under regular bidding. Multiple concurrent proposers thus do more than limit any single proposer's control over inclusion: by rewarding exclusive rather than shared information, they create an economic force toward geographic decentralization.
Read the full abstract ▾
All authors Ciamac C. Moallemi, Mallesh M. Pai, Dan Robinson, Christoph Schlegel
Machine Learning
Statistical Vignettes on the Grokking Phenomenon
Baye Malick Gning and Elvis Dohmatob
In this work, we study grokking in a simple, analytically tractable setting whose rich phenomenology allows us to disentangle the effects of initialization, signal strength, weight decay, sample size, and input dimension. These factors can produce superficially similar learning curves through distinct mechanisms, complicating their interpretation. Our analysis distinguishes early fitting through nuisance features, core feature learning, and delayed generalization. We identify sharp conditions under which grokking provably occurs and derive analytic asymptotic expressions for the lengths of the generalization plateaus. Our results also characterize regimes of early generalization and failure to reach a prescribed test accuracy target. We show that the interpolation edge constitutes a distinct dynamical regime, where vanishing spectral gaps separate core learning from generalization, helping clarify the role of proximity to interpolation in existing accounts of grokking. The resulting picture is analogous to fast and slow phases in chemical kinetics and suggests that the lazy-to-rich transition discussed in the literature may reflect a similar separation of dynamical time scales. Our theoretical results are faithfully supported by extensive experiments.
Read the full abstract ▾
Machine Learning
David González-Martínez, Philipp Nazari, T. Konstantin Rusch et al. (+1 more)
Residual connections propagate information through a network by additively accumulating layer outputs. Recently proposed alternatives, such as
Manifold-Constrained Hyper-Connections (mHC) and
Attention Residuals (AttnRes), instead use more expressive depth aggregation mechanisms that improve
language modeling performance at scale. In this work, we consider the mixing of representations over depth as a problem of storing and retrieving information across layers. Formalizing this perspective, we introduce Depth
Memory (DM), a unified framework that describes the standard pre-layer-norm (PreLN)
transformer, mHC, and AttnRes in terms of streams that carry representations through the network. One such stream, for example, could be an explicit
memory stream that is queried to enrich the input to every layer, and whose output is subsequently written back into the
memory for use by later layers. For AttnRes, this
memory stream is a growing cache of explicit layer representations. For mHC, it is an expanded residual stream. Building on this view, we propose Linear Depth
Memory (LDM), which formulates information
retrieval across depth as a regression problem solved using recursive least squares. The resulting architecture propagates three streams across depth: the standard residual stream, a linear associative
memory stream, and an inverse covariance stream. For high-throughput training and inference, we propose a hardware-aware implementation of LDM that reduces peak activation
memory and
memory traffic. LDM consistently achieves the lowest perplexities among PreLN, mHC and AttnRes baselines at the 400M and 1.5B parameter scales, alongside competitive zero-shot accuracy.
Read the full abstract ▾
All authors David González-Martínez, Philipp Nazari, T. Konstantin Rusch, Shiwei Liu
Multiagent Systems
Optimizing Team Behavior via Extremum-seeking Control in Multi-agent Reinforcement Learning
Svar Rajankumar Patel and Kshitij Jerath
In cooperative Multi-
Agent Reinforcement Learning (MARL),
agents must learn to coordinate their actions to achieve shared objectives. A key indicator of effective coordination is behavioral diversity, i.e., the
agents' potential to enact distinct yet complementary control policies given the same set of observations. Although behavioral diversity can enable role specialization and richer joint behaviors, its appropriate level is highly context- and task-dependent and often costly to tune. We propose Adaptive Diversity Control (ADiCo), a two-timescale framework that adapts behavioral diversity online during training. On a fast timescale, Diversity Control (DiCo) regulates system diversity toward a target value. On a slower timescale, an Extremum Seeking Controller (ESC) adjusts this target using only mean episodic returns, without requiring an explicit model or analytical gradient of the diversity-performance landscape. We evaluate ADiCo on six configurations across two cooperative MARL tasks, where it maintains task-appropriate diversity levels: low diversity when homogeneous policies are beneficial and higher diversity when heterogeneous policies are required. Through this adaptation, ADiCo matches or exceeds the mean performance of DiCo (best grid-searched diversity target) and Unconstrained DiCo (diversity allowed to emerge without target regulation). These results demonstrate that incorporating classical feedback control into MARL training enables effective coordination by providing a systematic alternative to task-specific manual selection of the diversity target.
Read the full abstract ▾
Machine Learning
SkillScaffold: Scaffolding Agentic Reinforcement Learning with Policy-Adaptive Skill Credit Assignment and Selective Injection
Wenlin Zhong, Hao Wang, Kun Yuan et al. (+6 more)
Agentic reinforcement learning (RL) enables large
language models (LLMs) to learn from multi-turn interactions with verifiable outcomes. Textual skills can improve exploration by distilling prior interactions into reusable guidance. However, existing skill-based RL approaches often either assume skill injection is beneficial or rely on indirect feedback or rollout-intensive evaluation to assess its utility. Our pilot studies show that rollout-based utility estimation for skill credit assignment is costly and unstable under limited sampling, while skill benefits vary across candidates and depend on the current policy's competence on each query. To address these challenges, we introduce SkillScaffold, a policy-adaptive training scaffold that combines low-cost skill-utility estimation with selective skill injection. By reusing outcome-verified training trajectories, SkillScaffold generates candidate skills and estimates their utility through outcome-aware likelihood comparisons, without candidate-specific environment rollouts. The resulting utility guides whether and which skill to inject, retaining skill-free execution when no candidate is estimated to help. Joint training enables the Generator to produce skills tailored to the evolving Executor. Scaffolded RL encourages skill internalization for skill-free inference. Experiments on ALFWorld, WebShop, ScienceWorld, and $\tau^2$-Bench demonstrate gains over strong agentic-RL and skill-based RL baselines across both short- and long-horizon tasks. Ablations and further analyses support the importance of utility-based selection and abstention. Our code is available at https://github.com/Vaidurya00/SkillScaffold.
Read the full abstract ▾
All authors Wenlin Zhong, Hao Wang, Kun Yuan, Fan Yang, Jinluan Yang, Dian Jin, Xu Tan, Yiquan Wu, Kun Kuang
Computer Vision
Registration is a Representation Problem: Solving Whole-Body PET/CT Registration
Neel Dey and Adrian V. Dalca
Raw image intensities are a poor surrogate for similarity across images in correspondence problems. Further, while auxiliary segmentation labels can provide semantic identification and improve matching, they can assign only coarse identities to fine-grained anatomy and often cover only a subset of structures, particularly in whole-body settings. We present an approach that instead establishes correspondence using complementary representations of anatomy and demonstrate its utility for longitudinal whole-body PET/CT registration in the Learn2Reg 2026 PSMAReg Challenge. Our 3D foundation model, anatomix, is used to derive fine-grained, per-voxel semantic representations that are consistent across volumes, and we further employ off-the-shelf segmentation models to obtain coarse anatomical identities. Together, these representations drive a GPU-accelerated iterative registration solver (FireANTs) in a hybrid multistage framework: the first stage estimates
alignment from dense multi-channel features, after which a second refines the deformation using both anatomix- and segmentation-based losses. Lastly, because registration should not introduce non-biological tumor volume changes across time, we develop lesion-specific component-wise volume preservation schemes. Without any task-specific training, our system won first place in the PSMAReg Challenge leaderboard. Code and interactive visualizations are available at http://www.neeldey.com/reps-for-reg.
Read the full abstract ▾
Cryptography and Security · Cross-listed from quant-ph
$\mathsf{EFI}\Leftrightarrow\mathsf{1PRS}$
Rishabh Batra
A series of recent works has established EFI pairs as the "minimal" assumption for
quantum cryptography. Recently, Cavalar et al. [CCC+26] showed that EFI pairs are equivalent to non-uniform expanding single-copy pseudorandom state generators (1PRS), and obtained a non-uniform meta-complexity characterization of EFI pairs. We remove the non-uniformity in their construction and prove that EFI pairs are equivalent to expanding 1PRS. As a corollary, we obtain the first uniform meta-complexity characterization of EFI pairs. Together with the equivalence of statistically verifiable one-way state generators and commitments proved by Batra and Jain [BJ24], this gives a
quantum analogue of the classical equivalence between one-way functions, commitments, and pseudorandom generators.
Read the full abstract ▾
Logic in Computer Science
Non-Wellfounded and Cyclic Proofs for LTL: A Syntactic Correspondence with Linear Nested Sequents
Tim S. Lyon and Lukas Zenger
We introduce the formalism of non-wellfounded and cyclic linear nested sequent calculi, developing concrete systems for linear temporal logic (LTL). The paper addresses two central problems, which we call "cycle recognition" and "unraveling." Cycle recognition concerns identifying cycles in non-wellfounded proofs in order to extract corresponding cyclic proofs, while unraveling studies the converse transformation, from cyclic proofs to non-wellfounded ones. Although these processes are well understood for Gentzen sequents, they have received little
attention for more expressive sequent formalisms and become more challenging in the linear nested sequent setting. To address cycle recognition, we show the completeness of non-wellfounded proofs relative to a particular normal form exhibiting a property we call "saturation recurrence," which enables the systematic extraction of cyclic proofs. To address unraveling, we introduce a specialized procedure that shifts rule applications forward along linear nested sequents, allowing non-wellfounded proofs to be reconstructed from cyclic ones. Overall, our work provides new proof-theoretic techniques for cycle recognition and unraveling in expressive multisequent formalisms.
Read the full abstract ▾
Robotics
SEAR: Self-Evolving Agentic Robotics
Xinyi Wang, Bangzheng Li, Hanyu Wang et al. (+5 more)
Frontier
language-model
agents can now improve
robots on their own: they write control code, test it, and keep what works as skills and
memories for the next task. We introduce SEAR (Self-Evolving Agentic Robotics), a framework and
benchmark built around two questions. Is the
agent really learning, or just trying more often? And does what it learns still help on new tasks, in a new simulator, and on a real
robot? The framework runs an
agent in three modes: it writes code that becomes the policy (code-as-policy), acts live in the control loop (
agent-as-policy), or works as a researcher that fine-tunes a
vision-
language-action model (auto-research). The
benchmark adds controls that separate retained experience from repeated attempts, familiar tasks, privileged simulator state, larger budgets, and luck. Across seven frontier
language models, 330 simulated and five physical tasks, and over four years of
agent runtime, we highlight three findings. First, the learning curve of a robotic
agent depends both on the overall time budget and the task difficulty relative to model capability. Harder tasks make longer sessions worthwhile; on easier tasks, re-rolling short attempts does just as well. Second, control code and shared routines transfer across tasks and simulators, while task-specific solutions do not. On a real
robot, we observe a performance gap between
agent-designed practice and novel human-designed scenes. Third, different models evolve more effectively in different ways. Some do best acting live, others writing code first. And a text-only model keeps pace with a comparable
multimodal one: perception helps with error assessment, but agentic
reasoning and tool use can matter more. We release the
benchmark and codebase to support future research.
Read the full abstract ▾
All authors Xinyi Wang, Bangzheng Li, Hanyu Wang, Lin Zhang, Andrea Yaoyun Cui, Zhenggang Tang, Shyamal Buch, Dejia Xu
Robotics
EqC: Equilibrium Energy Control for Introspective Lifelong Robot Learning
Haodong Yang, Runhan Huang, Bowen Zhang et al. (+7 more)
Long-term
robot autonomy depends on more than effective action generation: a policy must recognize when its behavior is no longer sufficiently supported and use assistance to improve. We introduce
Equilibrium Energy Control (EqC), a framework for introspective lifelong
robot learning built around an explicit conditional energy landscape. EqC casts generative action refinement as
equilibrium-seeking optimization, while the geometry of the refinement process reveals whether the generated behavior remains consistent with the learned landscape. This competence signal enables Energy-Gated DAgger, where persistent energy-consistency violations trigger expert takeover and the resulting recovery trajectories are aggregated for subsequent policy updates. EqC integrates naturally into both
Diffusion Transformer policies and flow-matching-based
vision-
language-action (VLA) policies. Across simulation and real-world bimanual manipulation, EqC matches or improves control performance and robustness while enabling timely intervention near the policy's competence boundary. A real-world study further shows how these interventions progressively expand the range of situations handled autonomously, providing a practical path from static generative controllers toward self-improving
robot systems.
Read the full abstract ▾
All authors Haodong Yang, Runhan Huang, Bowen Zhang, Iofai Lei, Zunzhe Zhang, Xiangcheng Zhang, Shaoting Zhu, Linzhan Mou, Yicheng Liu, Hang Zhao
Computer Vision
Trust Before Autonomy: Rethinking Agentic Remote Sensing for Crisis Response
Georgios Angelidis, Aggelos Psiris, Spyridon Evangelatos et al. (+4 more)
Agentic remote sensing is evaluated on how many questions it can answer, whereas crisis response depends on whether its answers can be relied on and how much of the decision can safely be handed to the
agent. This paper argues that autonomy should be earned, not assumed: An Earth-Observation (EO)
agent should be granted autonomy in proportion to demonstrated reliability, epistemic honesty and verifiability. The position is supported by an evidential meta-analysis of twelve agentic RS/EO/GIS
benchmark studies from peer-reviewed venues. Only one repeats runs, three contain unanswerable tasks and none measures calibration. Their own tables show that completed workflows are often not correct (task success exceeds tool-call correctness by up to 54pp on identical queries), that the gap to an expert trajectory widens overall with trajectory length (from 7 to 56pp), and that capability does not predict abstention (Spearman r from -0.25 to 0.50, all n.s.). The paper then contributes three EO-specific trust properties, an autonomy ladder for crisis EO tasks, and a readiness protocol that replays archived emergency activations before any live use. In this context, current evidence justifies assistive roles only.
Read the full abstract ▾
All authors Georgios Angelidis, Aggelos Psiris, Spyridon Evangelatos, Thomas Lagkas, Panagiotis Sarigiannidis, Iraklis Varlamis, Georgios Th. Papadopoulos
Cryptography and Security
Bytes to Threats: Byte-Sequence Transformer For Malware Detection
Abdelrahman Yassin Eljamal and Abdeltawab Hendawi
Existing malware detection approaches heavily rely on handcrafted features derived from static or dynamic analysis of the executable files. These methods require specialized domain expertise, dedicated analysis tools, and time-consuming analysis, which limits their scalability and robustness in the face of the increasing complexity and diversity of modern malware. Consequently, there is a growing interest in approaches that operate directly on raw bytes, which contain rich semantic and structural information. However, capturing long-range dependencies within executable files remains challenging for conventional static and CNN-based methods. This paper presents MalwareFormer, a
transformer-based, byte-level malware detection framework that operates directly on raw bytes, without requiring domain knowledge or feature extraction. MalwareFormer achieves an accuracy of 98.35%, a balanced accuracy of 98.32%, a macro-F1 of 98.32%, and a false positive rate of 1.94%, outperforming both baselines across all threshold-dependent classification metrics and demonstrating that
transformers provide a powerful and scalable approach for byte-level malware detection.
Read the full abstract ▾
Computation and Language
Where Do LLM Ensemble Gains Come From? Decomposing Candidate Value, Diversity, and Aggregation
Donghwan Kim
LLM ensemble gains are often measured against a single baseline, but the value attributed to additional candidates can change with the comparison used. We study an LLM aggregator under four matched conditions: Direct (no candidates), Single-slot (one candidate in a multi-slot interface), Duplicate (one candidate repeated), and Multi-candidate (separately generated candidates). Under either non-diverse control, overall candidate gain (CIV) splits algebraically into an additional-candidate contrast (DSV) and a candidate-context contrast (CCV); the split, but not CIV, depends on the chosen control. On MultiHopRAG model diversity (k=2), the repetition gap Rep=-8.33 pp [-13.25, -3.75] quantifies the difference between gains measured against Duplicate and Single-slot. On a matched FRAMES stronger-panel extension (n=200, k=3, aggregator output limit 4096), paired Delta Rep=+9.67 pp [+6.00, +13.50]; within panels, Rep changes sign (original -6.33; stronger +3.33). In FRAMES evidence aggregation, candidate-context contrasts remain strongly negative under both non-diverse controls. Our results show why ensemble evaluations should state the reference control before attributing gains to additional candidates.
Read the full abstract ▾
Information Theory · Cross-listed from quant-ph
Relative-Entropy Geometry and Convergence in Nonconvex Wiretap Optimization
Farzin Salek and Masahito Hayashi
We develop and analyze an entropic descent method for a nonconvex optimization problem on probability simplices arising in classical and
quantum secure communication. The optimization variables describe a randomized encoder, allowing its input distribution and
stochastic preprocessing to be optimized jointly. An exact relative-
entropy identity yields a channel-independent majorization bound and an explicit mirror-descent iteration with a fixed unit step size. We establish quantitative descent, convergence of objective values, and stationarity of accumulation points on their support faces for arbitrary finite classical channels, without assuming degradedness or less noisiness. For strictly positive transition
matrices, a square-root parametrization and a Kurdyka–Łojasiewicz argument establish finite length of the transformed trajectory and convergence of the entire sequence, including limits on the simplex boundary. For less-noisy channels, a restricted convex formulation admits global convergence with an explicit \(O(1/t)\) objective-error bound. We extend the relative-
entropy construction to noncommuting
quantum outputs and derive monotone alternating updates for covariant encoders, including correlated preprocessing over a fixed block. Numerical experiments compare the method with analytic
benchmarks and constrained optimization, reveal multiple attracting values, and demonstrate strict improvements from
stochastic preprocessing.
Read the full abstract ▾
Artificial Intelligence
Pack Light, CarryOn: Building Object-Oriented Context for Long-Horizon Agents
Yuchen Lu, Julian McAuley and Lei Zhang
Long-horizon
agents accumulate progress in persistent task state, yet repeatedly process interaction histories describing how that progress was achieved. This raises a natural question: can long-horizon context instead be organized around reusable progress rather than the history that produced it? In this paper, we introduce CarryOn, an object-oriented context runtime that organizes continuation around reusable task objects. Rather than treating compressed histories or reconstructed accounts of progress as the primary basis for continuation, CarryOn builds a context view from continuation carriers, complementary execution context, and a live tail of recent interactions. A core challenge is keeping this reusable basis valid after its surrounding history is retired: how can CarryOn update the basis without reconstructing the full representation at every fold? CarryOn maintains carrier records across folds, revising affected records against bound evidence and applicable environment checks while carrying forward unaffected records and jointly updating execution context. Experiments on OSWorld 2.0 and BrowseComp-Plus show improved gains of 5.8% and 7.0% while reducing total
token usage by 51.9% and 22.9%, respectively. Project page: https://ethannju.github.io/carryon
Read the full abstract ▾
Artificial Intelligence
BLADE: Auditing Information Use for Failure Attribution in LLM Multi-Agent Systems
Xiaoguang Guo, Yaohao Long and Chuxu Zhang
Large
language model multi-
agent systems extend individual
agents' capabilities through collaboration across specialized roles, tools, and workflows. Yet collaboration makes failures harder to diagnose: one
agent may misinterpret a correct result while others carry the mistake forward. Failure attribution therefore seeks to identify the responsible
agent and decisive error step. Recent methods trace dependencies, but knowing where information came from is not the same as knowing where it went wrong. We introduce BLADE (Blame Localization via Audited Dependency Evidence), a training-free framework that moves beyond tracing information to auditing its use. Using Jev for structured probabilistic judgments, BLADE uses dependency probabilities to select relevant source-decision handoffs for auditing. At these handoffs, it examines whether receiving decisions use their sources faithfully and within task requirements, distinguishing error introduction, propagation, and repair. These findings inform a global diagnosis that considers all execution steps. Without reference answers, BLADE achieves the best or tied-best step accuracy among eight strategies under a shared Jev decision interface on 1,228 failed runs across five Who&When and Who&When Pro subsets, outperforming All-at-once by 2.2-11.1 percentage points. Making responsibility evidence explicit offers a path toward more interpretable and reliable agentic systems.
Read the full abstract ▾