Cryptography and Security
MIRROR: Multipath Quorum Integrity for LLM Multi-Agent Communication
Ryuichi Yamafuji Lun, Jingzhen Wang, Shreyas Kolte et al. (+1 more)
Inter-
agent communication is central to Large
Language Model Multi-
Agent Systems (LLM-MAS), but it introduces an underexplored vulnerability:
Agent-in-the-Middle (AiTM) attacks that manipulate messages in transit without compromising the
agents themselves. Prior work reports Attack Success Rates (ASR) approaching 100% on structured tasks. Existing defenses rely on semantic validation, which requires additional inference and can block benign outputs, or on transport-layer encryption, which does not help when an intermediary legitimately terminates TLS. We present MIRROR, a communication-layer integrity primitive that replicates a single canonicalized payload across k logical routes and accepts a message only when a strict majority of routes report the same digest. MIRROR uses unkeyed hashing and so authenticates nothing on its own, since an active on-path adversary can always recompute a digest over a payload it has modified. All integrity derives from the assumption that honest routes form a majority. The digest serves only to make witness routes constant-size and to bind the recovered payload to the quorum-agreed value under second-preimage resistance. We give the guarantee under a route-compromise bound alpha < 0.5, and extend it to correlated routes, where the quantity that matters is the size of the largest shared-failure group and not the route count. We further show that availability and integrity degrade at the same threshold: below alpha = 0.5, quorum-denial and message-dropping adversaries cannot block honest traffic. Across MMLU, HumanEval, and MBPP on two frameworks and four communication
topologies, and in a MetaGPT deployment against a production API, MIRROR reduces ASR to 0% below the threshold at 1x LLM
token cost. LLM-as-a-Judge costs 35x in the same deployment, and blocks up to 44.2% of benign outputs in the
topology sweep.
Read the full abstract ▾
All authors Ryuichi Yamafuji Lun, Jingzhen Wang, Shreyas Kolte, Ruiteng Li
Number Theory
Arithmetic growth in partially commutative semigroups of endomorphisms
Wade Hindes
We establish asymptotics for the degree and height growth rate functions of partially commutative
semigroups of endomorphisms of varieties, i.e.,
semigroups where the only relations between generators, if any, are commuting ones. To do this, we prove a general result connecting the growth rates of two abstract functions, one multiplicative and one simply nonnegative, subject to a Tate telescoping hypothesis and a sufficiently nice normal form on the underlying
semigroup. Finally, we construct several new families of endomorphisms on which our asymptotics apply. To do this, we generalize the standard ping-pong lemma for elements to a ping-pong lemma for subsemigroups.
Read the full abstract ▾
Machine Learning
From Behavior to Provenance: Attributing Tabular Foundation Models to Synthetic Pretraining Data
Mohamed Bouadi, Nassim Bouarour, Shivam Dubey et al. (+2 more)
Training-data attribution aims to identify which training examples shape model behavior, yet validating such claims is difficult because causal training influence is rarely observable. We argue that controlled synthetic pretraining makes attribution experimentally testable. Using O'PRIOR, a provenance-rich synthetic task generator for tabular foundation models, we construct a testbed in which every pretraining task carries explicit lineage over structural mechanisms, missingness, confounding, shortcuts, and distribution shift. We combine behavior-conditioned attribution with counterfactual retraining and provenance-aware interventions to test both task-level faithfulness and mechanism-level consistency. On held-out real tasks, removing the top-attributed 5% of synthetic tasks decreases mean ROC-AUC by 0.013, compared with 0.002$\pm$0.004 under random removal, while removing bottom-attributed tasks improves performance by 0.003. Within shortcut-provenance tasks, targeted removal yields an effect of 0.043 versus 0.016 for matched random removal. Provenance discrimination is more modest by ranking AUROC (0.55-0.62), despite substantial top-k enrichment, revealing that provenance association and interventional faithfulness need not coincide. Our results establish synthetic provenance as a controlled setting for verifiable contributive attribution
Read the full abstract ▾
All authors Mohamed Bouadi, Nassim Bouarour, Shivam Dubey, Aditya Tanna, Vinay Kumar Sankarapu
Soft Condensed Matter
Traveling waves, phase separation and pattern formation in the active stepping stone model
Rashmiranjan Bhutia, Prasad Perlekar and Kabir Ramola
Selective advantage leads to the proliferation of a species into an ecosystem. On the other hand, self-propulsion can lead to dynamic clustering of individuals. To analyze the intricate effects produced by such competing processes, we introduce a minimal model for a proliferating active population, where run-and-tumble migration is combined with a
stochastic birth-death process. Beginning with the microscopic dynamics of the particles, we derive the corresponding hydrodynamic equations and validate them through numerical simulations. We demonstrate that the inclusion of activity leads to the emergence of novel morphological patterns, which are sensitive to the specific rules governing migration. Furthermore, we show that activity significantly alters the characteristics of invasion waves, increasing their propagation speed. However, beyond a threshold, excessive activity disrupts the wave-like nature of the invasion process. We also extend our study to two dimensions and analyze the instabilities that lead to pattern formation in such systems.
Read the full abstract ▾
Machine Learning
Mitigating Convergence Collapse in Fixed-Target Anomaly Detectors via Kernel-Anchored Locality Regularization
José Lucas De Melo Costa, Fabrice Popineau, Arpad Rimmel et al. (+1 more)
A family of tabular anomaly detectors trains a neural map toward a fixed target under squared-error loss and scores anomalies by the test-time residual; contraction matching, one-step rectified flow, and reconstruction autoencoders all fit this template. We characterize a convergence collapse: better optimization makes the detector worse. At convergence, the learned map tracks the target even off-distribution, so the residual signal vanishes on anomalies as well as on normal data. These detectors therefore rely on implicit non-convergence (early stopping, capacity caps) to retain signal. We argue this is structural: effective anomaly detection requires a locality constraint that blocks unconstrained extrapolation. Classical detectors (kNN, KDE, isolation forests, LOF) enforce locality explicitly; fixed-target neural detectors do not. We formalize the connection by showing that the kernel-regression analog of a fixed-target detector is a finite-bandwidth Nadaraya-Watson smoother, which we call Kernel Contraction Matching (KCM). KCM is closed-form, training-free, and CPU-efficient, yet matches established neural baselines on ADBench. Building on this bridge, we introduce the Kernel-Anchored Regularizer (KAR), which penalizes deviation of the neural prediction from a kernel-weighted average of training targets. Across collapse-prone ADBench
datasets and three backbones, KAR mitigates collapse and improves AUROC under prolonged training.
Read the full abstract ▾
All authors José Lucas De Melo Costa, Fabrice Popineau, Arpad Rimmel, Bich-Liên Doan
Machine Learning
Drive vs. Decay: On the Training Dynamics of Joint-Embedding Predictive Architectures
José Lucas De Melo Costa, Seong Woo Ahn, Fabrice Popineau et al. (+2 more)
Joint-
Embedding Predictive Architectures (JEPAs) are prone to representation collapse, typically mitigated through empirical heuristics. We develop an early-training stability theory that unifies these heuristics. Linearising the coupled JEPA gradient flow around the trivial fixed point reveals two competing effects: a driving force ($\gamma$) and a decay effect ($\sigma$). Under approximate spectral decoupling, a per-mode stability ratio $\mu_i = \gamma_i / \sigma_i$ factorises into independent data-side and predictor-side terms and the count of unstable modes tracks the rank of representations that can emerge. The framework predicts a phase boundary, which we confirm empirically across more than 800 Tabular-JEPA configurations. It also unifies predictor scaling, masking ratio, and EMA as distinct mechanisms for shifting $\mu$. Guided by this analysis, we introduce ResidualPred, a
transformer predictor whose
attention is biased toward the identity at initialisation; it improves both effective rank and downstream accuracy on tabular
benchmarks and in I-JEPA pretraining on CIFAR-10, CIFAR-100, STL-10, and ImageNet. Our framework connects empirical collapse-avoidance heuristics to an explicit dynamical picture, yielding theory-driven stabilizers. Code is available at https://github.com/jose-melo/drive-vs-decay.
Read the full abstract ▾
All authors José Lucas De Melo Costa, Seong Woo Ahn, Fabrice Popineau, Arpad Rimmel, Bich-Liên Doan
Computer Vision
SCOPE-4D: Endoscopic 4D Geometry Foundation Models
Chaoyi Zhou, Zhongpai Gao, Anwesa Choudhuri et al. (+6 more)
Geometric understanding supports endoscopic
navigation and robotic assistance, but learning reliable endoscopic geometry faces two challenges: scarce
geometric annotations and ambiguity between camera motion and tissue deformation. We present SCOPE-4D, an endoscopic 4D geometry foundation model that jointly predicts camera parameters, dense geometry, and 3D tissue trajectories from
monocular RGB
video in a single forward pass. Our curation and annotation pipeline constructs SCOPE-5K, a collection of approximately 5,000 clips spanning real and synthetic gastrointestinal endoscopy and laparoscopy. The collection provides rich
geometric supervision and includes newly collected phantom and real-colonoscopy evaluation sets.
Geometric supervised fine-tuning on SCOPE-5K learns endoscopic priors that improve camera and depth estimation. Common–Residual Motion (CRM) further constrains local deformation relative to common tissue movement. Together with
geometric supervision, CRM and trajectory supervision further improve camera and depth estimation over
geometric fine-tuning alone while enabling dense 3D tissue tracking. Evaluations on public and newly collected
benchmarks demonstrate strong in-domain and out-of-domain geometry, superior 3D tracking, and more stable long-sequence colon reconstruction. A blinded user study further supports the perceived reconstruction quality on real clinical
video. Together, these results demonstrate the value of large-scale endoscopic supervision and motion constraints for joint geometry estimation and tissue tracking.
Read the full abstract ▾
All authors Chaoyi Zhou, Zhongpai Gao, Anwesa Choudhuri, Meng Zheng, Benjamin Planche, Run Wang, Terrence Chen, Siyu Huang, Ziyan Wu
Artificial Intelligence
A Multi Method Importance and Performance Efficiency Analysis of Topological Metrics for Natural Visibility Graph Based Cyber Attack Detection
Ali Melih Kanca and Ilker Turker
Natural Visibility Graph (NVG) based analysis characterizes network traffic through topological descriptors reflecting different structural properties. However, not all descriptors contribute equally to cyber-attack classification, and extracting a large metric set can increase computational cost. This study evaluates 21 NVG derived topological metrics and investigates whether a compact subset can preserve classification capability while improving computational efficiency. Four importance analysis methods SHAP, grouped Permutation Importance, Boruta, and Recursive Feature Elimination (RFE) are integrated through a Consensus Ranking strategy. Based on this ranking, Full21, Top15, Top10, Top7, Top5, and Top3 configurations are evaluated using the CICIDS2018
dataset, a CNN classifier, and stratified 5 fold cross validation. The three highest ranked metrics are avg_clustering_coeff_median, avg_clustering_coeff_std, and avg_clustering_coeff_mean. Top3 achieved the highest observed mean performance, with 97.148% accuracy, 97.055% weighted F1 score, and an MCC of 0.9675, compared with 95.999%, 95.521%, and 0.9549 for Full21, respectively. It also reduced total runtime from 14,961.39 s to 589.22 s (96.06%). These results indicate that importance guided metric reduction can provide a compact NVG representation with higher observed mean predictive performance and substantially lower computational cost under the evaluated setting.
Read the full abstract ▾
Robotics
Filter-Aware Fine-Tuning for Safe Humanoid Whole-Body Tracking
Pranit Mohnot, Christian Helten, Daniele Gammelli et al. (+1 more)
Safe whole-body motion is essential for deploying
humanoid robots in unstructured environments. Modern
humanoid control commonly separates reference specification from execution, with a planner, teleoperator, or motion generator providing a reference that a reinforcement-learning policy tracks through dynamically feasible whole-body control. Runtime safety filters, such as control barrier functions (CBFs), offer a promising approach for enforcing newly introduced constraints via interventions on the tracker's outputs. We show, however, that treating the tracking policy and safety filter independently induces fundamental mismatches, as filtering alters both the executed actions and the induced state distribution. We study this policy-filter interface through case studies that isolate dynamics, objective, and information mismatches, highlight their root causes, and use these insights to develop CoFiT (Constrained Filter-aware Tuning), a filter-aware fine-tuning method for pretrained trackers. Across diverse constraint scenes, CoFiT reduces violation time relative to filter-only training by 91% on TWIST2 and 21% on SONIC, while requiring smaller safety filter corrections. On Unitree G1 hardware, CoFiT reduces violation time by 83% for TWIST2 and completes every trial without operator intervention, whereas 50% of baseline trials require an operator stop. Together, these results provide actionable insights into policy-filter interactions and establish design principles for integrating learned trackers with runtime safety filters.
Read the full abstract ▾
All authors Pranit Mohnot, Christian Helten, Daniele Gammelli, Marco Pavone
High Energy Physics (Theory)
Holographic thermal propagator at finite energy from modularity and triality
Borut Bajc, Mendrit Latifi and Katarina Trailović
We extend the modularity-based approach of arXiv:2509.02226 for the
holographic thermal propagator to the physically relevant case of nonzero energy $\omega$. Via the AGT correspondence, the problem is mapped to $\mathcal{N}=2$ $SU(2)$ gauge theory with $N_f=4$ hypermultiplets. Turning on $\omega$ activates all four hypermultiplet masses, so that the full $SO(8)$
flavor symmetry constrains the low-temperature expansion, rather than the subgroup left unbroken at $\omega=0$. As a result, the building blocks of the expansion are no longer just Eisenstein series but also Jacobi theta functions. We use this fact, together with the Molien series of the resulting invariant ring and the modular anomaly equation, to fix the low-temperature expansion coefficients order by order, matching only a finite number of terms in the Zamolodchikov's $q$-recursion at each order. We obtain the low-temperature expansion of the propagator and find agreement with the result of a completely different method, the near-boundary expansion of the bulk wave equation.
Read the full abstract ▾
Robotics
NEEDLEWORK: Offline Rewriting of Robot Data with Verified Local Stitches
Juntao Ren, Yifan Hou and Shuran Song
Robot demonstrations may contain useful behavior even when individual episodes are inefficient or unsuccessful. Trajectory stitching offers a way to compose these behaviors into improved training data, but identifying useful connections and verifying their feasibility is difficult in high-dimensional
robot data, where many prior methods rely on low-dimensional state representations. We introduce NEEDLE, an offline
dataset-augmentation
algorithm that addresses these challenges by adding short, verified action bridges between recorded observations in high-dimensional
robot demonstrations. First, NEEDLE identifies and creates connections that bypass suboptimal detours, broaden action coverage, and augment the original
dataset with failed trajectories, using only RGB images, proprioception, and episode-level outcomes, without new environment interaction or privileged object state. Next, we present a sampling technique that incorporates accepted bridges into policy training without synthesizing intermediate images or discarding the original demonstrations, allowing policies to learn alternative actions while retaining the original
dataset's coverage. On real-
robot tasks, NEEDLE improves success rate over the strongest baseline on each task by an average of 21 percentage points.
Videos and supplementary materials are on https://needle-work.github.io/.
Read the full abstract ▾
Robotics
SoTa: Soft Tactile Skins for Dexterous Manipulation
Jingyun Yang, Baiyu Shi, Timothy Yu et al. (+6 more)
A growing body of work suggests that tactile sensing gives
robot policies contact information that complements
vision in dexterous manipulation. However, visuo-tactile
robot data remains scarce: dexterous demonstrations require teleoperating
robots, which limits
dataset scale. Human demonstrations are far cheaper to collect and offer a path to scale this data, but only if human and
robot hands carry tactile sensors with corresponding signals. This requires sensors that conform to different hand geometries, cover the full hand, and share a common layout across embodiments. We present SoTa, a low-cost capacitive tactile skin that provides full-hand coverage on humans and
robots while preserving a shared layout of 202 taxels across corresponding finger and palm regions. Our multilayer design with fabric electrodes enables in-house fabrication of thin, soft skins with customizable geometry for under $10 in materials per skin. The sensor retains over 97% of its initial response span after 10,000 loading-unloading cycles with traces retaining continuity through 1,280 tight-fist folding cycles. The shared taxel layout supports human-
robot co-training with a common tactile encoder and no learned cross-sensor mapping. Across three contact-rich manipulation tasks, tactile observations improve in-distribution success over
vision-only policies. With a fixed
robot demonstration budget, adding human demonstrations more than doubles mean success across eight evaluation conditions, from 22.8% to 45.9%, improving success in all five out-of-distribution conditions. We plan to open-source the resources needed to fabricate and operate these skins.
Read the full abstract ▾
All authors Jingyun Yang, Baiyu Shi, Timothy Yu, Haitian Liu, Alberta Longhini, Weichen Wang, Rika Antonova, Zhenan Bao, Jeannette Bohg
Geophysics
Multi-Modal Surface Wave Inversion using Physics Guided Variational Autoencoders
Sujith Swaminadhan, Andrey Bakulin, Sergey Fomel et al. (+1 more)
The growth in seismic acquisition techniques has led to large-scale surveys, making traditional manual dispersion analysis logistically and computationally prohibitive. While deep learning techniques can offer an automated alternative, standard convolutional neural networks (CNNs) often fail to account for the non-uniqueness of the inverse problem and can produce geologically unrealistic, smooth velocity models. To address these challenges, we propose a physics-guided multi-modal variational autoencoder (PG-VAE) for scalable near-surface characterization. Our framework introduces a multi-stream fusion strategy that maps independent latent features from both raw shot gathers and spectral-dispersion images by leveraging the physics of Rayleigh-wave dispersion. We implement a specialized decoder architecture with learned transposed convolutions and residual refinement to enforce a structural bias towards layered earth models. Physics guidance is incorporated via a differentiable surrogate network that serves as a forward-consistency regularizer, penalizing predictions that violate the physics of Rayleigh-wave dispersion. The network is trained on diverse synthetic data generated via elastic modeling, and then validated on unseen synthetic data, and subsequently evaluated on an ultra-high-density field
dataset from the Devine test site in Texas. We also employ gradient-weighted class activation mapping (Grad-CAM) to examine the input features contributing to the network predictions. The proposed method could serve as a scalable tool for near-surface characterization and as an initial model-building tool for high-resolution elastic full-waveform inversion.
Read the full abstract ▾
All authors Sujith Swaminadhan, Andrey Bakulin, Sergey Fomel, Yangkang Chen
Nuclear Theory
Vicente Vento
In this reflection, I want to look back over more than thirty years, from 1978 to 2008, of my professional and personal relationship with Mannque Rho. I will briefly describe some of the original ideas that emerged from our collaboration: the Little Bag, the Cheshire Cat principle, and Skyrmion matter. I do not intend to be exhaustive, neither in content nor in references, which can be found in our published works. Instead, I want to retrace the path we took together through physics, accompanied by many good friends.
Read the full abstract ▾
Dynamical Systems
A Lean~4 Framework for the Radii Polynomial Method
Fengyang Wang, Konstantin Mischaikow and Marcio Gameiro
Understanding nonlinear dynamics is central to many fields of study, yet most nonlinear systems are beyond the reach of closed-form solutions. Numerical simulation is then used to reveal dynamical structures before their existence can be established. This alone, however, leaves the validity of the computed solution unanswered. Rigorous numerical methods address this gap by combining computation with explicit analytic estimates. Among them, the radii
polynomial approach, a Newton–Kantorovich type argument, turns a numerical approximation into a proof of existence and uniqueness of a true solution near that approximation. In previous practice, the verification of these inequalities relies on interval-arithmetic libraries and hand-derived algebraic bounds. In this paper, we present a framework for the radii
polynomial method in the Lean~4, which turns numerical certificates into fully type-checked
theorems about the true solution. A central challenge in this process is to manage the complexity of the formalization. Our approach begins with a family of problems and their common structure. The spaces in which the analysis takes place are developed through universal properties in suitable categories, with extension properties that characterize maps by their action on generators. These properties guide the design of a reusable interface such that problem-specific arguments concentrate on the numerical data and the algebraic bounds that distinguish one equation from another. We formalize a general radii
polynomial theorem, develop weighted coefficient
algebras and reusable operator estimates, connect finite computational checks to the hypotheses of the
theorem, and establish the realization
theorems that transport certified coefficient space zeros to analytic solutions of the original equations.
Read the full abstract ▾
Information Theory
Joint Movement and Compression Ratio Design for Mobile Embodied AI Networks (MEAN)
Yahao Ding, Jiaxiang Wang, Zhouxiang Zhao et al. (+3 more)
Mobile
embodied AI networks (MEAN) enable
embodied agents to perceive, reason, communicate, and act in wireless environments. In such networks,
agent mobility can improve channel conditions, while semantic compression can reduce transmission payloads. However, movement consumes energy, and stronger compression incurs additional computational cost. This paper studies joint movement, semantic compression, and transmit power design for an uplink MEAN system. We formulate a max-min energy efficiency (EE) problem by jointly optimizing transmit power, movement distance, and semantic compression ratio under controllable power constraints. The problem is non-convex due to the coupled signal-to-interference-plus-noise ratio (SINR), mobility-dependent channel gains, and fractional EE objective. To solve it, we propose an alternating optimization (AO)-Dinkelbach
algorithm, where the fractional objective is handled by the Dinkelbach transformation, transmit power is updated via successive convex approximation (SCA), and movement distance is updated by coordinate-wise grid search. Simulation results show that the proposed scheme outperforms no-mobility and no-compression baselines, demonstrating the benefit of jointly exploiting mobility control, semantic compression, and power allocation in MEAN.
Read the full abstract ▾
All authors Yahao Ding, Jiaxiang Wang, Zhouxiang Zhao, Zhaohui Yang, Mingzhe Chen, Mohammad Shikh-Bahaei
Emerging Technologies
Duplication-Aware Retiming and Cell Interface Redesign for Superconductor Circuit Minimization
Panagiotis Papanikolaou, Alex Vanasse, Haoran Jin et al. (+2 more)
Superconductor electronics have increasingly shifted away from RSFQ and its variants toward logic families that eliminate explicit gate-level clocking. While this transition enables simpler
circuits and more efficient architectures, it also introduces an implicit reliance on dual-rail codes, resulting in inherent gate duplication. This work presents a duplication-aware retiming methodology for Josephson junction (JJ) count minimization, co-optimizing register placement and polarity assignment. The approach applies beyond SFQ to any monotonic
circuit. We further identify
cell interfaces—specifically, interconnect drivers, receivers, and fanout (FO) elements—as dominant contributors to JJ count in each
cell. A new amplifier design is introduced to reduce these costs, integrated within existing SFQ
cells, experimentally verified, and characterized to form a new SFQ
cell library. Our results demonstrate a 63-71% JJ count reduction in single-cycle implementations and 41-66% reduction in multi-cycle implementations compared to the prior state-of-the-art. The latter establishes a new Pareto frontier, achieving both shorter critical paths and lower JJ counts than the best-to-date single-cycle implementations.
Read the full abstract ▾
All authors Panagiotis Papanikolaou, Alex Vanasse, Haoran Jin, Georgios Tzimpragos, Jennifer Volk
High Energy Astrophysical Phenomena
Neutrino flavor instabilities intermediate between slow and fast
Nicolas Viaux and Lucas Johns
The
neutrino flavor composition in
supernovae and
neutron star mergers is shaped by
flavor instabilities, which are commonly classified into three types: slow (SFI), fast (FFI), and collisional (CFI). The focus here is on mixed-type instabilities combining attributes of the standard varieties. We examine instability growth rates in two-dimensional parameter spaces (“phase diagrams”) whose axes correspond to instabilities of pure type. We find critical curves connecting the pure-type instability onset points. We also exhibit qualitative trends in the growth-rate spectra $\gamma_{\bm{k}}$, emphasizing the occurrence of crossovers and phase transitions in the fastest-growing
Fourier mode $k_{\textrm{max}}$ and drawing
attention to the distinction between spatial-mean ($\bm{k} = 0$) and
flavor-wave ($\bm{k} \neq 0$) instabilities. Our paper is a step toward understanding
flavor instabilities as they actually occur in nature rather than in idealized limits.
Read the full abstract ▾
Artificial Intelligence
World Editing: Intervening on Executable Worlds at Increasing Depth
Max Ku, Nok-Kan Law, Yu-Chien Tang et al. (+15 more)
Interactive world models are increasingly capable of generating environments and acting within them, yet deliberately editing an existing executable world remains underexplored. We formulate world editing as intervening on an existing world while preserving properties that should remain unchanged, and introduce intervention depth as an axis describing how strongly an edit couples world entities, dynamics, and systems. We instantiate this capability through industry-grade game modding and introduce IGMWorld, together with IGMBench, a
benchmark of 110 tasks and over 1.1K executable state and behavioral criteria across Minecraft and Terraria. The tasks span property, entity, dynamics, and system interventions and are evaluated through deterministic executability, behavioral, preservation, and visual checks. Frontier coding
agents already exhibit substantial world-editing capability: the strongest configuration solves 78.2% of tasks under a strict task-level criterion, while criterion-level performance reaches 94.8%. Reliability generally decreases with intervention depth, and this pattern persists even among tasks with similar numbers of evaluation criteria. Most failed edits still build and load successfully, suggesting that the main difficulty is making the edited world behave as requested. Visual consistency remains a separate weakness, with all evaluated configurations below 50% joint visual pass rate. These results show that world editing is a distinct capability from world generation and interaction, and that executable games provide a practical testbed for studying it.
Read the full abstract ▾
All authors Max Ku, Nok-Kan Law, Yu-Chien Tang, Shih-Ying Yeh, Ping Nie, Andy Zheng, Tat Hei Lai, Fei-Yueh Chen, Nikko Yu, Wei-Chieh Sun, Suzy Huang, Chiao-Wei Hsu, Chih-Chuan Huang, Chak-Wing Mak, Ho Yin Sam Ng, Edisy Kin Wai Chan, Min-Hung Chen, Ho Kei Cheng
Artificial Intelligence
Choosing Before Acting: Comparative Value Estimation for Long-Horizon Tool-Use Agents
Yu Li, Zheng Zhang, Xin Liu et al. (+3 more)
Large
language models (LLMs) rely on long-horizon tool invocation sequences for complex tasks, where each invocation can alter the task state and condition subsequent decisions. In long-horizon tool use, final-outcome rewards provide weak credit assignment over long interaction traces. Step-level rewards can offer more targeted feedback, but obtaining reliable step supervision often requires human or LLM judgment, or additional rollouts to estimate the downstream effect of an intermediate decision. In this paper, we argue that effective tool-use
agents should estimate the long-horizon value of a possible next tool invocation before executing it. This objective requires comparative supervision over alternative invocations under the same context, while logged trajectories only contain the invocation that was actually taken. Therefore, we propose Comparative Inference for Tool-use
Agents (CITA). CITA trains a Comparative Inference Model (CIM) from paired signals that combine observed tool behavior, scalable supervision from a
Bayesian tool-graph simulator, and semantic judgments from LLM-based comparison. The resulting CIM learns to estimate how likely a possible next tool invocation is to support final task success under the current context. Across three tool-use
benchmarks and multiple backbone LLMs, CITA consistently improves Tool F1 and task success. Additional analysis shows that CIM learns accurate step-level value estimates for comparative tool choices.
Read the full abstract ▾
All authors Yu Li, Zheng Zhang, Xin Liu, Shengtian Yang, Guangfeng Cai, Lei Feng
Combinatorics
The Boolean lattice $\mathcal{B}_6$ is not a sphere order
David A. Meyer
Using the equivalence of sphere order representations of posets with Minkowski space
embeddings, we show that the Boolean
lattice $\mathcal{B}_6$ is not a sphere order. This is the first explicit finite example of a poset that is not a sphere order.
Astrophysics of Galaxies
Inferring magnetic field strengths in high-redshift lensing galaxies from Faraday rotation measure observations
Stefan Reissl, Ralf S. Klessen, Annalisa Pillepich et al. (+3 more)
Observations of Faraday rotation in the
lensing galaxy of CLASS B1152+199 at z = 0.439 suggest that efficient dynamo mechanisms had already amplified
galactic magnetic fields by this epoch. We assess the reliability of this method for deriving
magnetic field strengths and the likelihood of observing similarly large Faraday rotation differences. We select 64
star-forming disk
galaxies from the TNG50 cosmological magnetohydrodynamical simulation, matched to CLASS B1152+199 in mass and
redshift, and forward model them using polarized radiative transfer with POLARIS and subsequent RM synthesis. TNG50 predicts characteristic large-scale field strengths of B$_0$ = 1.9-15.6 $\mu$G for individual
galaxies, while the combined sample yields B$_0$ = 10.3 $\mu$G. Field strengths B$_{obs}$ inferred from the synthetic observations generally recover B$_0$ within a factor of three. B$_{obs}$ depends only weakly on the assumed
magnetic field geometry, whereas
galaxy inclination and beam depolarization have stronger effects. The differential rotation measure recovered from RM synthesis remains correlated with the directly integrated value, although beam depolarization systematically reduces its magnitude. Only 1.41% of directly integrated sightline pairs reproduce the observed $\Delta$RM = 1040 $\pm$ 60 rad m$^{-2}$ of CLASS B1152+199, increasing to 3.16% for sightlines at galactocentric distances comparable to the observed 2.6 and 6.5 kpc. Differential Faraday rotation of gravitationally lensed sources therefore provides a robust probe of large-scale
magnetic fields, while the large $\Delta$RM of CLASS B1152+199 is relatively rare among the considered TNG50 analog
galaxies.
Read the full abstract ▾
All authors Stefan Reissl, Ralf S. Klessen, Annalisa Pillepich, Sui Ann Mao, Rüdiger Pakmor, Dylan Nelson
High Energy Physics (Phenomenology)
Precise QCD Predictions for Photon-Associated Hadron Production at the LHC
Gabriele Fiore, Aude Gehrmann-De Ridder, Alexander Huss et al. (+2 more)
Prompt-
photon production with recoiling QCD radiation is crucial for new-physics searches, determinations of parton distribution and fragmentation functions, and jet-quenching studies in heavy-ion collisions. We present the first fully differential NNLO QCD predictions for
photon-associated identified-
hadron production at the LHC. We compare predictions obtained with different fragmentation function (FF) sets to ALICE and ATLAS data at $\sqrt{s}=5.02\,{\rm TeV}$. Our calculation provides an important theoretical ingredient for FF global fits and offers a precision
proton–
proton baseline for
photon-tagged heavy-ion studies.
Read the full abstract ▾
All authors Gabriele Fiore, Aude Gehrmann-De Ridder, Alexander Huss, Francesco Merlotti, Giovanni Stagnitto
Mesoscale and Nanoscale Physics
Fluctuation-induced magnetoresistance in graphene Hall bars at charge neutrality
Anna Grigoreva, Alex Levchenko, Eleftherios Kirkinis et al. (+1 more)
In the temperature interval $T\sim 50 - 100 K$, the rate of electron-electron collisions in
graphene devices may exceed the momentum relaxation rate due to disorder and electron-
phonon scattering. In this regime, the motion of the electron
liquid may be described by the hydrodynamic equations. In the hydrodynamic approximation, at the charge neutrality point the electric current is decoupled from the hydrodynamic flow, and the conductivity of the system is equal to the intrinsic conductivity of the electron
liquid, $\sigma_0$. It was recently shown that the coupling of charge transport and hydrodynamic flow induced by the thermal fluctuations enhances the macroscopic conductivity of the system and results in very strong positive magnetoresistance (MR) at relatively weak fields. Here we develop a quantitative theory of the fluctuation mechanism of MR at charge neutrality and obtain analytic expressions for MR in
graphene Hall bars. The functional dependence of MR on the
magnetic field $H$ is sensitive to the
Hall bar geometry and orientation relative to the current. We also account for the effect of momentum relaxation on MR.
Read the full abstract ▾
All authors Anna Grigoreva, Alex Levchenko, Eleftherios Kirkinis, Anton Andreev
Differential Geometry
Ricci flow on manifolds with positive isotropic curvature in all dimensions
Simon Brendle and Raphael Tsiamis
We extend the classification of
manifolds with positive isotropic
curvature to all dimensions $n \geq 5$. This result also completes the classification of uniformly positive isotropic
curvature on complete non-compact $\kappa$-noncollapsed ancient solutions and gradient shrinking
Ricci solitons. The proof involves a construction of pinching cones in all dimensions, refining and extending to all dimensions the results of the first author. We have used MATHEMATICA to find viable parameters for the cone families. The final presentation is self-contained and can be verified manually. No stage of this work involved LLMs.
Read the full abstract ▾
Machine Learning
Slow-Fast Multi-Teacher On-Policy Distillation for Capability Preservation
Xiaofei Yin, Tong Chu, Jiyuan Fu et al. (+3 more)
Foundation
multimodal large
language models are designed to support a broad spectrum of capabilities across diverse domains. Multi-teacher on-policy
distillation (MOPD) provides an effective framework for consolidating domain-specific expertise into a single student model. However, MOPD training gradually drives the student away from its initialization model, and general capabilities decline as the displacement grows, resulting in capability interference. A direct remedy is constraining the student toward its initialization, but this suppresses the acquisition of domain expertise as well. We propose Slow-Fast Multi-Teacher On-Policy
Distillation (SF-MOPD), which couples a fast model, the current student updated directly by each teacher, with a slow model, an exponential moving average of the student. The slow model absorbs the learning signal gradually, serving as a moving capability reference that fuses the general foundation with confirmed domain expertise. For each teacher, SF-MOPD computes the teacher-induced update in log-probability space and removes only the component that pushes the fast model further away from the slow model, while retaining aligned and orthogonal components. Experiments across multiple model scales demonstrate that SF-MOPD effectively mitigates capability interference, enhances specialized
multimodal capabilities, and reduces the average degradation on general-capability
benchmarks, consistently outperforming vanilla MOPD.
Read the full abstract ▾
All authors Xiaofei Yin, Tong Chu, Jiyuan Fu, Jun Lan, Shuheng Zhou, Huijia Zhu
Robotics
World-Calibrated Proposal-to-Action Flow for Vision-Language-Action Models
Jie He, Wei Li, Junwen Tong et al. (+3 more)
Flow-based
Vision-
Language-Action (VLA) policies generate action chunks by transporting samples from a task-agnostic isotropic Gaussian source. As this source is conditioned on neither recent execution nor predicted future
evolution, (i) it discards the local continuity established by recently executed motion. (ii) Even when predictive world representations are introduced, they often only condition the transport dynamics rather than determine where generation starts, how far it may deviate, or along which action directions it may expand. Building on this observation, we introduce ProAct, a world-calibrated proposal-to-action framework that makes the generative source itself predictable. (i) To preserve motion continuity, a lightweight Proposal Expert converts recent actions into a scene-aware hypothesis via one motion-anchored endpoint flow-matching step, initializing generation near the demonstrated action
manifold. (ii) To jointly capture intended scene
evolution and proposal-future compatibility, a prospective World Expert treats the hypothesis as a soft motion prior while predicting the task-consistent latent future. (iii) From this compatibility, the model calibrates a proposal-centered anisotropic source, where a bounded per-step extent controls the allowed deviation and a trace-normalized low-rank geometry under a condition-number budget allocates refinement over coupled translation, rotation, and gripper directions. Compared with $\pi_{0.5}$, ProAct improves performance across simulation and real-world tasks while reducing denoising steps by 50%, inference latency by up to 25.8%, and increasing throughput by up to 34.8%.
Read the full abstract ▾
All authors Jie He, Wei Li, Junwen Tong, Rui Shao, Wei-Shi Zheng, Liqiang Nie
Computer Vision
SCION: Scene Composition with Instanced Neural Primitives
William Koch, Amogh Joshi, Cyrus Vachha et al. (+2 more)
Real-world scenes are compositional: bricks, blades of grass, pebbles, and tree leaves recur across human-built and natural environments. Existing neural scene representations model these elements independently. Most 3D Gaussian Splatting and follow-up abstraction and compression methods treat each element as unique, fitting millions of independent Gaussians per scene. Prior methods like Splat and Replace fit template objects, but they require mostly manual selection of repeated elements. As a result, these representations store redundant parameters and provide weak manipulation handles for downstream tasks. We introduce SCION, a hierarchical compositional scene representation that replaces independent Gaussians with a compact vocabulary of reusable primitives and lightweight world-space instances that place transformed copies throughout the scene. We fit this representation to multi-view captures via a joint optimization over discrete and continuous scene parameters, combining two-level densification over splats and instances with an
adversarial loss that preserves detail across shared primitives. The recovered structure yields a compact, controllable representation while maintaining high quality even at 1.2 MB. SCION achieves rate-distortion favorable to existing Gaussian compression methods, and it enables instance-level scene editing and animation without retraining. Our results show that neural scene representations need not memorize scenes as independent primitives; they can discover reusable parts. Project webpage: https://light.princeton.edu/SCION
Read the full abstract ▾
All authors William Koch, Amogh Joshi, Cyrus Vachha, Cheng Zheng, Felix Heide
Quantum Physics
Halving the cost of QROM again via dense encoding
Danial Motlagh
Quantum read only
memory (QROM) is a substantial source of non-Clifford cost in fault-tolerant
quantum algorithms. Borrowing dirty
qubits can reduce this cost, provided their unknown initial states are restored. We consider the practically relevant
qubit-constrained regime, where limited workspace restricts the table-block size and table loading dominates the cost. For $N$ entries of $b$ bits and approximately $b\lambda$ dirty
qubits, SelectSwap has leading Toffoli count $2N/\lambda$. Recent work using sequential bit packets reduces this to $(1+1/b)N/\lambda$. In this work we introduce a dense encoding construction of QROM, where two bits of classical information are stored in each dirty
qubit instead of one by utilizing both the $Z$ and $X$ bases. On its own, dense encoding provides a $2\times$ improvement over SelectSwap by reducing the leading Toffoli count to $N/\lambda$. Combining it with sequential bit packets gives $\tfrac12(1+1/b)N/\lambda$, a further halving of the sequential construction's leading cost, and an approximately $4\times$ improvement over SelectSwap in the large-$N$ limit for practical values of $b$.
Read the full abstract ▾
Computer Vision
DeskForge: Dense Supervision from Desktop Environments for Computer-Use Agents
A. Said Gurbuz, Ahmed Nassar, Sunghwan Hong et al. (+2 more)
Computer-use
agents need to reliably ground action targets in complex desktop scenes, where multiple applications, overlapping windows, and visually similar controls compete for
attention. Existing training data rarely pair such scenes with dense annotations or vary them in a controlled way. We introduce DeskForge, a controllable desktop environment that composes and explores real applications to generate large-scale supervision for computer-use
agents. It varies application states, content, window layout, appearance, and resolution, and fuses screenshots, accessibility trees, and window geometry into dense element annotations while recording the outcome of each executed action. Using this environment, we construct DeskForge-1M, a corpus of 1.2M annotated desktop observations containing 159.7M element instances. We fine-tune four
vision-
language models on 200K grounding examples drawn from DeskForge-1M. All four improve across held-out desktop conditions and on all five external GUI grounding
benchmarks; for Qwen3.5-4B, accuracy increases by 11.51 percentage points on ScreenSpot-Pro and 10.11 points on OSWorld-G. The gains also translate to long-horizon task completion: under a fixed planner, the fine-tuned action models solve more WebArena-Infinity and OpenApps tasks, with Qwen3.5-4B increasing from 31 to 50 of 119 tasks and from 3 to 15 of 100 tasks, respectively. These results show that controllable composition of real desktop environments provides a scalable source of supervision for improving both GUI grounding and long-horizon computer use. The framework code, the
dataset, and the fine-tuned model are available from the project page: https://saidgurbuz.github.io/deskforge/
Read the full abstract ▾
All authors A. Said Gurbuz, Ahmed Nassar, Sunghwan Hong, Marc Pollefeys, Peter W. J. Staar