Cryptography and Security
Audience-Bound Persistent Memory: Authorization Across the Memory Lifecycle
Sibo Liu
A personal
language agent that acts for its owner across private and shared conversations can learn a fact from one audience and later place it in the context it assembles for another. We study authorization before context across the whole
memory lifecycle. Each
memory item carries the audience present when it was recorded; derived items are partitioned by audience, receive the intersection of their sources' audiences, or are suppressed; an audience widens only by an explicit, object-specific grant; and an item enters a model attempt only when every current viewer belongs to one of its authorized audiences, with unresolved viewers failing closed to public-only. Under explicit identity, provenance and complete-mediation assumptions, this admission is sound and policy-complete on the exact assembled context, enforced by exclusion rather than by model behavior. We realize it in two independently persisted reference architectures, a flat store and a relationship graph, and, descriptively, in a native
agent-
memory runtime. In a prospectively frozen confirmation over 10,000 multi-party histories, no forbidden item entered any architecture's context, whereas unscoped
retrieval exposed forbidden items in 82% of its contexts. Entitled recall matched policy-equivalent baselines exactly and exceeded unscoped
retrieval by 0.30 Recall@5, with a Holm-confirmed advantage that grows with distractors. No architecture produced a wrong-principal substitution, but unscoped substitutions were too rare to establish the prespecified joint decision.
Read the full abstract ▾
Cryptography and Security
TTMark: Pairwise Distortion-Free Watermarking Beyond Single-Token Entropy
Ruibo Chen, Zhengmian Hu, Donghang Lu et al. (+5 more)
Distortion-free watermarking enables reliable attribution of machine-generated text while preserving output distribution. However, existing methods operate independently on each generated
token, making their detection capability fundamentally constrained by the
entropy of the next-
token distribution. We present Tandem
Token WaterMark (TTMARK), a general pairwise watermarking framework that extends distortion-free watermarking from individual
tokens to adjacent
token pairs. By watermarking the joint distribution of consecutive
tokens, TTMARK enlarges the effective watermarking alphabet from V to $V^2$, allowing the detector to exploit both
token entropy and conditional
entropy while preserving distortion-freeness over the joint distribution. We further introduce a branch-isolating concatenated tandem generation
algorithm that efficiently constructs the joint distribution in a single forward pass. Theoretically, we show that pairwise watermarking achieves better expected detection strength in low-
entropy regimes. Extensive experiments across multiple
language models,
datasets, and three representative distortion-free watermarking schemes demonstrate that TTMARK consistently improves detectability without degrading generation quality, while also improving robustness to edits and substantially enhancing localized watermark detection.
Read the full abstract ▾
All authors Ruibo Chen, Zhengmian Hu, Donghang Lu, Xuehao Cui, Georgios Milis, Yihan Wu, Jian Du, Heng Huang
Software Engineering
LatentSift: Policy-State Filtering for Token-Efficient Verification of Software Engineering Agents
Yuning Han, Yangchenchen Jin, Tyler Jandreau et al. (+1 more)
Test-time scaling improves software engineering
agents by generating multiple candidate trajectories and selecting the best one. Verifying and selecting among these long interactions can consume as many
tokens as generation itself. Existing hybrid workflows first apply an LLM-based execution-free (EF) verifier to filter candidates before running tests, which adds another model pass over every trajectory. We introduce LatentSift, a
token-free and execution-free filter that replaces this first stage with hidden states the policy already produces while generating the candidates. It represents each candidate through its
reasoning, observation, and function-call states, compares them with positive and negative banks of such states collected from successful and unsuccessful trajectories during policy training, and fuses the resulting distance scores with a learned linear score to retain promising candidates for the execution-based stages. On SWE-bench Verified, across three
agents and two policy sizes, LatentSift cuts EF-verifier
tokens by 66.6–81.0% and total verification
tokens, which include test generation, by 49.1–62.1% at K=16, while hybrid Best@16 matches or improves on each
agent's reference workflow, rising from 59.26% to 60.06% on DeepSWE-Preview.
Read the full abstract ▾
All authors Yuning Han, Yangchenchen Jin, Tyler Jandreau, Jingwei Sun
Quantum Algebra
Meromorphic open-string vertex algebras and twisted modules from Courant algebroids
Qixuan Fang and Fei Qi
We construct a sheaf of $\frac{1}{2}\mathbb{Z}$-graded meromorphic open-string vertex
algebras from a transitive Courant algebroid $E$ equipped with a generalized metric and a chosen generalized Levi-Civita connection on a smooth
manifold. Using a holonomy principle for transitive anchored bundles, we realize these
algebras as parallel sections of a tensor
algebra bundle built from bosonic-fermionic affinizations of $E$. For even-rank $E$ with suitable Clifford module bundle data, we construct a sheaf of canonically twisted modules containing the canonical weighted spinor bundle in weight zero. The associative
algebra of parallel tensors acts on spinor sections through iterates of an $E$-connection combining covariant differentiation and Clifford multiplication, with mixed terms incorporating the action of the bosonic connection on fermionic tensors. Distinguished states built from the inverse Courant pairing and inverse generalized metric give vertex-operator components, acting on embedded spinor sections as the canonical Dirac generating operator, its formal adjoint, and their anticommutator, the generalized Hodge
Laplacian. In the exact Courant case with three-form $H$ and the exterior
algebra Clifford module, the operators specialize to $d-H\wedge$, its adjoint, and the twisted Hodge
Laplacian, reducing to the ordinary de Rham and Hodge operators when $H=0$. This extends the vertex-operator realization of
geometric differential operators from functions to weighted spinors and differential forms, within the program of studying two-dimensional supersymmetric nonlinear sigma models.
Read the full abstract ▾
High Energy Physics (Phenomenology)
$t\bar{t}$ spin correlations from toponium production scenarios to non-relativistic resummation beyond the standard perturbative baseline
A. Cota Rodríguez, Jesús A. V. Corral, A. Paredes Sotelo et al. (+1 more)
$t\bar{t}$
spin correlations originating from theoretical models assuming on one side toponium bound state formation in
proton-
proton collisions at 13 TeV are compared against predictions from non-relativistic QCD ressumation effects via Green's function formalism that go beyond the perturbative $t\bar{t}$ baseline with no bound state but could justify extra yield of events in the $t\bar{t}$ mass spectrum recently observed by ATLAS and CMS Collaborations, still awaiting an explanation. Toponium production comprises two scenarios, a simplified Effective Field Theory (EFT) with $\eta_t\rightarrow t\bar{t}$ decays and an additional model with scalar and vector toponium states with $\eta_t\rightarrow HZ$ and $J_t\rightarrow ZW^+W^-$ decays. Experimental measurements on $t\bar{t}$ system are added to the comparisons. Fitted diagonal
spin correlation $C_{kk}$, $C_{rr}$, $C_{nn}$
matrix terms show a relative increase in value ranging from 200%-2330% for toponium $t\bar{t}$ decays with respect the perturbative NLO baseline. Relative contribution from resummation effect ranges from 0.6% to 10.4% and its combination with perturbative baseline improves agreement with Data distributions, with $\chi^2/N_\mathrm{dof}$ value reduced by $\sim$4.5%-60.0% . Equivalent fitted cross-section for resummation effect $\sigma_{res}$, ranges from 7.3 to 9.1 pb in consistency with excess yield reported by LHC experiments. $\sigma_{res}$ $>$ 50 pb range is excluded at 95% C.L. A classifier performing at detector level achieves a ROC curve area of $0.91$ leading to toponium signal to background significances over 55. These results indicate that efficient isolation of observed $t\bar{t}$ excess events from background is feasible and a characterisation via
spin correlation measurements would reveal their consistency with either toponium production or non-relativistic effects beyond the perturbative baseline.
Read the full abstract ▾
All authors A. Cota Rodríguez, Jesús A. V. Corral, A. Paredes Sotelo, J. A. Murillo Quijada
Machine Learning
AdaKerNet: Neural Kernel Decoding for Task-Adaptive Prediction with Multimodal Large Models
Konstantinos D. Polyzos, Eleni Oikonomou and Tara Javidi
Large foundation models have been introduced with the promise of efficient adaptation to downstream tasks. Yet, under limited supervision, MLLMs, an important class of large foundation models, remain challenging to adapt to various downstream tasks. Adaptation typically relies either on MLLM parameter fine-tuning or on training neural-based decoders. Both approaches struggle under limited supervision, while fine-tuning additionally requires access to model parameters, which is often unavailable for closed-source models. We introduce AdaKerNet, a novel learnable task-adaptive neural kernel decoder. AdaKerNet is fully agnostic to the parameters of the underlying MLLM and operates solely on its (frozen) rich representations obtained from the diverse available modalities. AdaKerNet relies on (i) a set of learnable, Lipschitz-controlled
multimodal features derived from these MLLM representations; (ii) a reference kernel that provides a soft structural prior on those features; and (iii) a lightweight nonlinear neural predictor that adaptively deforms that structure. Learning the kernel representation and the neural predictor jointly within a unified optimization framework allows AdaKerNet to capture features and
geometric relationships relevant to the downstream task. Numerical tests across four MLLMs: BLIP-2, LLaVA-1.5, Qwen2.5-VL, and Gemini
Embedding 2, and
multimodal inputs spanning text, audio, images, and tabular measurements demonstrate significant and consistent improvements over direct MLP,
attention-, autoencoder- and kernel-based decoders, across a range of scarce-label budgets, with average error reduction of up to 41% across baselines. These results establish AdaKerNet as an effective approach for prediction from frozen
multimodal representations in the scarce label regime. Additional structural ablations highlight the complementary contributions of AdaKerNet's components.
Read the full abstract ▾
Medical Physics
Cardiac segmentation for ventricular tachycardia radioablation and radiotherapy patients with cardiac implants
Nicholas Summerfield, Dustin Jacqmin, Chase Ruff et al. (+8 more)
Deep-learning (DL)
segmentation enables cardiac substructure (CS) delineation for advanced cardiac-sparing radiation therapy (RT). Patients presenting for RT may have implantable devices (cardioverter-defibrillators (ICDs)) and metal artifacts. Stereotactic body RT (SBRT) is used to treat ventricular tachycardia (VT), a population frequently with ICDs. We apply a DL model for RT patients with ICDs or undergoing VT-SBRT, enabling
segmentation and CS-level dose analysis. CT simulation (CT-SIM) of 19 patients with ICDs and received VT-SBRT were evaluated. A validated
segmentation model was retrained using a highly-curated CT-SIM cohort, supplemented with ICD images for training and validation, with a hold-out ICD test set (n=10), to predict 20 CS. Additional CT-SIM
datasets included cancer-RT patients with (n=9) and without (n=10) ICDs to evaluate generalizability. The retrained ICD-supplemented model was compared against a non-ICD model using Dice similarity coefficient (DSC), 95% Hausdorff distance (HD95), and Wilcoxon signed-rank test (p<0.05). Dosimetric evaluation was performed for VT-SBRT CS. For VT-SBRT, the ICD-supplemented model achieved average DSC/HD95 of 0.71(0.19)/10.7(22.5)mm respectively, outperforming (p<0.05) the non-ICD model. On CT-SIM with ICDs, average DSC/HD95 were 0.70(0.22)/7.9(5.1)mm respectively, outperforming (p<0.05) the non-ICD model. On CT-SIM without ICDs, the average DSC was 0.74(0.17) (Non-ICD, DSC, 0.75(0.16), p<0.05) and average HD95 was 5.5(3.1)mm (Non-ICD, HD95, 5.3(2.8mm)). For VT-SBRT, CS dose depended on target location, with higher relative dose delivered to the ventricles, tricuspid valve, atrioventricular node, and left coronary arteries than other CS. ICD-supplemented training improved CS
segmentation despite metal artifacts without meaningfully degrading non-ICD performance, supporting
segmentation and CS-level dosimetry in ICD populations.
Read the full abstract ▾
All authors Nicholas Summerfield, Dustin Jacqmin, Chase Ruff, Andrew M. Baschnagel, Adam Burr, Michael Bassetti, Ryan Kipp, Matthew Kalscheur, Patrick M. Hill, Ming Dong, Carri Glide-Hurst
Neurons and Cognition
Cross-attention encoding models reveal dynamic spatiotemporal routing across human higher visual cortex
Iishaan Inabathini and Margaret M. Henderson
Understanding how the
brain parses actions and events from time-varying natural inputs is a central challenge in neuroscience. Recent work has used deep neural network (DNN) models to build stimulus-computable fMRI encoding models that predict single-voxel responses to complex natural
videos. However, the majority of
video-computable encoding models predict responses using simple linear mappings from model
tokens, overlooking the spatiotemporal structure shared by
video representations and neural responses. Recent cross-
attention encoding models address this limitation for static images, enabling flexible stimulus-dependent weighting of image content across space. Here, we extend this framework to naturalistic
video, using per-parcel cross-
attention to dynamically route features from a self-supervised
video model (V-JEPA-2) across both space and time, fitting this model to fMRI responses to short
video clips. We compare joint spatiotemporal
attention with factorized and selectively constrained alternatives, and find that joint routing improves predictions of
brain responses to held-out
videos across higher visual regions, most consistently in lateral and dorsal visual areas associated with dynamic motion perception. Moreover, our method provides interpretable, stimulus-specific
attention maps that dynamically follow moving objects, revealing which locations and temporal moments contribute to each neural response. We further show that
attention maps from parcels in different category-selective networks (face-, body-, scene-selective) differentially weight content in accordance with expected semantic selectivity. Together, this work provides a new computational framework for understanding how visual information is adaptively weighted by cortical populations during dynamic visual perception.
Read the full abstract ▾
Artificial Intelligence
Engineering Simplicity: Simple Mechanism Interfaces Steer LLM Agents
Kehang Zhu, Anand Shah and David Parkes
Can interaction formats and textual scaffolds help large
language model (LLM)
agents make better decisions, and do better decisions come with better explanations? We study these questions in auctions and matching, multi-
agent environments with explicit rules and known optimal strategies. These settings let us vary how a decision problem is presented while retaining a
benchmark for evaluating behavior. Drawing on human-motivated theories of simplicity, we compare interfaces that elicit a complete bid or ranking with sequential interfaces that make safe choices easier to identify. We then hold the interaction format fixed and vary
reasoning scaffolds and rule descriptions. Across four model families, the ascending auction interface substantially reduces bid deviations. The matching comparison also shows why sequential responses require different error accounting from complete rankings. Laying out payoff contingencies and explaining why truth-telling is safe also improve choices, whereas
prompts to plan through matching rounds or form beliefs about opponents worsen play overall. In auctions, these behavioral gains are not accompanied by corresponding improvements in measured verbal indicators of strategic understanding in the
agents' short stated plans. Other
prompts change those indicators without improving bids. Our findings suggest that human-motivated theories of simplicity can inform the design of decision environments for artificial
agents. They also show why scaffolds should be evaluated through realized choices as well as explanations: improvements in one need not appear in the other.
Read the full abstract ▾
Computer Vision
Compress to Remember: Learning Compact Memory via On-Policy Distillation for Long Video Generation
Xiaoyu Wu, Weihang Guo, Yifei Wang et al. (+3 more)
Standard
video generators do not natively compact historical context into reusable
memory tokens. As generation continues, the growing history makes it increasingly difficult to retain information from earlier frames due to long-context degradation. Key-frame-based approaches address this challenge by retaining selected past frames, but can discard information needed for future generation. Rather than relying on frame selection alone, we study whether a frozen
video generator can supply the supervision needed to learn a compact representation of the history. We propose Prediction-Aligned Context Compaction (PACC), which uses a learned compressor to aggregate information across past frames into compact
memory tokens. We train the compressor through on-policy
distillation, using the same frozen generator both as a student when conditioned on compressed
memory and as a teacher when conditioned on the full history. The student generates continuations, while the teacher provides targets for the same noisy inputs at each denoising step. Only the compressor is updated to align the student's predictions with these targets. We evaluate PACC on MBench, which jointly measures
memory-event coverage and consistency. PACC outperforms the strongest baseline by 6.63 points on Causal-rCM and 3.19 points on Causal Forcing. Evaluation on VBench-Long using MovieGen
prompts further shows that PACC produces minute-long
videos with generation quality competitive with baselines. Together, these results show that learning to compact historical context can improve long-
video memory without modifying the underlying generator.
Read the full abstract ▾
All authors Xiaoyu Wu, Weihang Guo, Yifei Wang, Xinze Feng, Lydia E. Kavraki, Zhiwei Steven Wu
High Energy Physics (Phenomenology)
Effective kinematic variables for searching for dark matter mediators in single top quark production processes
E.E. Boos, V.E. Bunichev and L.V. Dudko
Effective kinematic variables are proposed in the rest frame of the common cluster of the top
quark and mediator decay products for searching for and determining the type of dark matter mediator particles and measuring their masses in single top
quark production processes. The efficiency of the variables is confirmed by full Monte Carlo simulation taking into account the response of the LHC detectors.
Read the full abstract ▾
Software Engineering
Strategies for Deploying AI Agents in Production at Scientific User Facilities
Ming Du, Xiangyu Yin, Michael Prince et al. (+40 more)
Agentic artificial intelligence (AI) is moving beyond research demonstrations toward production use at scientific user facilities, including light sources,
neutron sources, nanoscience centers, and autonomous laboratories. Its scientific value extends beyond increasing throughput.
Agents can perform repeatable tasks in calibration, measurement execution, and quality control, as well as initial analyses that turn data into reviewable evidence, allowing scientists to focus on hypotheses, unexpected observations, and interpretation. Drawing on deployments of LLM-driven
agents at the APS, this perspective distills practical strategies with an emphasis on elements that can be reused across instruments and facilities. We discuss
agent harnesses for beamline control, facility knowledge
retrieval, and data analysis while keeping the underlying design principles independent of any specific implementation. These principles cover inference endpoints, tool-server architectures, non-text data, computationally intensive services, reusable skills, and governed learning throughout an instrument's lifecycle. We also consider how network and Linux operations, governed shared
memory, and deterministic orchestration can extend these patterns across facility services. Because LLM capabilities continue to evolve, these recommendations represent a snapshot of the technology as of the date on the cover.
Read the full abstract ▾
All authors Ming Du, Xiangyu Yin, Michael Prince, Yi Jiang, Rajat Sainju, Tekin Bicer, Yanqi Luo, Eric Codrea, Peco Myint, Nina Andrejevic, Juanjuan Huang, Trupti Mohanty, Pawan Tripathi, Dishant Beniwal, Hemant Sharma, Doga Gursoy, Aileen Luo, Tao Zhou, Chenran Xu, Jan Ilavsky, Matthew T. Dearing, Ryan Chard, Hoon Seo, Dariusz Jarosz, Elaine Chandler, Thomas Fors, Hairong Shang, Pete Jemian, Xuli Wu, Madeline Miller, Ryan Aydelott, Nicholas Schwarz, Francesco de Carlo, Byeongdu Lee, Yine Sun, Moises Smart, Philippe Piot, Joe Sullivan, Mathew J. Cherukara, Alec Sandy, Stefan Vogt, Jonathan Lang, Laurent C. Chapon
Probability
Distance flexibility in spatial matching: the value of concentration
Taha Ameen and Sophie H. Yu
In spatial matching
markets, a supply unit's flexibility is measured by its service radius, the maximum distance at which it can serve demand. In dimensions $k \geq 2$, we study how a platform should allocate service radii among the supply nodes subject to a budget on their sum. The platform makes this choice before observing supply and demand locations, with the objective of maximizing the expected fulfilled demand. We show that the shape of a preferred allocation depends on the total budget: under suitable conditions, large budgets favor allocations that are more uniform in the sense of majorization, while small budgets favor concentration. We also characterize a non-uniform allocation that is asymptotically optimal for a very-sparse regime, and show that the uniform allocation is suboptimal in this regime. Our results provide theoretical explanations for the radius allocation questions raised by the numerical experiments in [ASY26b].
Read the full abstract ▾
Analysis of PDEs
A Fractional Logistic-Type Elliptic Problem with Global Interactions: Existence, Local Uniqueness, and the Fractional-to-Local Limit
Alireza Khatib, Somayeh Mousavinasr and Bashir Zeimarani
We study a fractional logistic-type
elliptic problem on a bounded domain with homogeneous exterior conditions and linear and nonlinear integral interactions. The interaction kernels may be nonsymmetric and sign-changing, so the problem need not admit a variational formulation. In the symmetric competitive regime, coercivity yields a weak solution by global minimization. For arbitrary fixed values of the interaction parameter, explicit smallness conditions allow us to apply Schauder's fixed-point
theorem without
symmetry or sign restrictions on the kernels. Additional sign assumptions give nonnegative solutions. When the nonlinear exponent satisfies $q\geq2$, a contraction condition yields uniqueness in an invariant ball and convergence of the Picard iteration. Finally, as the fractional order approaches one, we prove subsequential convergence to a solution of the local Dirichlet problem, together with convergence of the fractional energies to the Dirichlet energy. Uniqueness of the local solution in the limiting ball gives convergence of the entire family.
Read the full abstract ▾
Artificial Intelligence
Better Nearest Neighbor Graph Indices via (Efficient) LLM-Guided Pruning
Fangzhou Wu, Haike Xu and Sandeep Silwal
Graph-based approximate nearest neighbor search (ANNS) is widely used for large-scale semantic search. Its indices are constructed primarily based on
geometric relationships among
embeddings of an input
dataset (e.g., documents or images), rather than explicitly optimizing for semantic relevance. However, when using these indices for downstream query
retrieval, performance is evaluated based on the semantic relevance of the retrieved results to the query. This creates a fundamental "geometry-semantic" mismatch between how the indices are constructed and how their
retrieval results are evaluated. While existing LLM-based reranking methods can partially mitigate this mismatch at query time, they leave this underlying structural problem in the graph unresolved. We therefore propose LLM-Guided Graph
Pruning (LGP), a general framework that addresses this mismatch directly by leveraging LLM
reasoning to refine an existing ANN graph index itself. LGP identifies structurally "low-value" neighbors of nodes and replaces them with LLM-selected alternatives that provide useful semantic information while retaining desired
geometric structures of the original graph, including sparsity and efficient navigability. Experiments on representative semantic
retrieval benchmarks show that LGP consistently improves end-to-end
retrieval performance over both vanilla greedy graph search and LLM-based reranking across widely used graph-based ANN indices such as DiskANN and HNSW.
Read the full abstract ▾
High Energy Physics (Phenomenology)
Vector Meson Dominance and Vector Meson Production
Craig D. Roberts
Against a background supported by the three pillars of emergent mass, this contribution challenges the fidelity of vector
meson dominance as an instrument for relating the electromagnetic vector-
meson (V) production reaction $e + p \to e^\prime + V + p$ to the hadronic process $V + p \to V + p$. It also describes a viable reaction model for exclusive photoproduction of light and heavy vector
mesons from the
proton, which exposes the $q\bar q$ content of the
photon. That model's successes reveal that it is premature to link extant $V + p \to V + p$ data with, for instance, in-
proton gluon distributions, the QCD trace anomaly, or pentaquark production. Improved reaction theory and more precise data are required before the validity of such links can objectively be assessed.
Read the full abstract ▾
Artificial Intelligence
HyperZip: Efficient Data Compression through Personalized Diffusion LLMs with Hypernetworks
Thai Nguyen, Khang Tran and NhatHai Phan
Large
language models (LLMs) have shown strong potential for lossless data compression, but existing approaches are constrained by the high computational cost and low throughput of autoregressive decoding. We propose HyperZip, an efficient and scalable LLM-based compression framework that leverages
diffusion-based LLMs (dLLMs) with Multi-
Token Prediction (MTP) to accelerate LLM-based data compression processes. We identify a trade-off in
diffusion-based compression, where increasing decoding throughput degrades the compression rate. To mitigate this trade-off, HyperZip employs a hypernetwork to generate data-specific updates from a context representation, adapting the dLLM to the target data without costly fine-tuning, resulting in a low compression rate and high throughput. Extensive experiments show that HyperZip achieves a superior trade-off between compression rate and speed compared with state-of-the-art baselines.
Read the full abstract ▾
Strongly Correlated Electrons
Artificial versus Natural Atoms: The uncanny capability of the many-body Schrödinger equation to produce emergent behavior
Constantine Yannouleas
The paper reviews the theoretical and experimental progress achieved in the last 25 years in understanding the novel physics of artificial atoms and molecules as arising from the formation of Wigner molecules (WMs) of localized (to a stronger or lesser extent) fermionic or bosonic particles, which are finite
quantum analogs of the more familiar bulk Wigner
crystal. The term artificial atoms, as used here, encompasses a broad range of recently fabricated
quantum nanodevices and experimental apparatuses consisting of a finite number of mutually repelling confined particles, including two-dimensional
semiconductor and moiré transition metal dichalcogenide
quantum dots, as well as trapped ultracold neutral atoms or ions. These nano-sized or micro-sized artificial devices and apparatuses (in single well or multi-well of variable-shape arrangements) hold a great promise for technological applications in the field of
quantum information and
quantum computers, as well as for advances in fundamental many-body physics. Prominent
quantum effects of Wigner molecularization are the strong quenching of the spectral energy gaps, the appearance of rovibational spectra (in analogy with natural molecules),
entanglement, and pinning due to an external perturbation. In high
magnetic fields or at rapid rotation, WMs provide an alternative theory to the fractional
quantum Hall effect. The physics of Wigner molecules is shown to derive from the solutions of the many-body Schrödinger equation (MBSE) in the regime of strong interparticle correlations arising from the dominance of the potential over the kinetic energy, or from a high
magnetic field, as well as from a rapid rotation... (continues in the paper).
Read the full abstract ▾
Cosmology and Nongalactic Astrophysics
Probing Baryons with the Kinematic Sunyae Zel'dovich Effect and Machine Learning Derived Peculiar Velocities using DESI DR2 and ACT DR6
Yulin Gong, Rachel Bean, Patricio A. Gallardo et al. (+54 more)
We present novel empirical constraints on average optical depths and gas density profiles from combining kinematic Sunyaev Zel'dovich effect (kSZ) measurements and nonlinear peculiar velocities, reconstructed using machine learning, for massive
halos traced by DESI DR2 luminous red
galaxies (LRG). The pairwise kSZ is measured using two ACT DR6
cosmic microwave background (CMB) temperature maps (ILC and 150 GHz maps) for seven luminosity-selected LRG samples. We find consistent results between the two maps and significant detections across the samples. We reconstruct LRG line-of-sight peculiar velocities in two
cosmic epochs, centered on $z=$ 0.55 and 0.8, using a simulation-trained
Transformer machine learning model to extend beyond the linear approximation. Average optical depths are inferred by comparing the kSZ pairwise correlation to the measured pairwise velocity correlation from the reconstructed velocity field and, separately, from a theoretical prediction using the best fit Planck cosmology, with the highest significance reaching SNR = 14.2 and 15.1 respectively. We also obtain velocity-weighted AP-filtered kSZ-stacked cluster density profiles, which show evidence of extended ionized gas around massive LRG groups and no strong
redshift evolution within current measurement uncertainties. These results demonstrate the power of combining spectroscopic
galaxy and CMB data, and employing machine learning methods, to probe diffuse
baryons and coherent large-scale velocity fields. They open up new avenues to leverage upcoming
galaxy survey data from Euclid, Roman and Rubin LSST, in tandem with multi-frequency CMB/sub-mm data from CCAT and Simons Observatory.
Read the full abstract ▾
All authors Yulin Gong, Rachel Bean, Patricio A. Gallardo, Boryana Hadzhiyska, Yun-Hsin Hsu, Jenna Moore, Eve M. Vavagiakis, Nicholas Battaglia, J. Aguilar, S. Ahlen, A. Aviles, F. Beutler, D. Bianchi, D. Brooks, A. Carnero Rosell, T. Claybaugh, A. de la Macorra, Arjun Dey, Biprateep Dey, P. Doel, A. Font-Ribera, J. E. Forero-Romero, E. Gaztañaga, G. Gutierrez, K. Honscheid, S. Juneau, T. Karim, R. Kehoe, D. Kirkby, A. Kremin, O. Lahav, M. Landriau, L. Le Guillou, M. Manera, A. Meisner, R. Miquel, S. Nadathur, J. A. Newman, H. E. Noriega, E. Paillas, N. Palanque-Delabrouille, W. J. Percival, F. Prada, I. Pérez-Ràfols, C. Ravoux, G. Rossi, L. Samushia, E. Sanchez, C. Saulder, D. Schlegel, M. Schubnell, H. Seo, J. Silber, M. Siudek, G. Tarlé, B. A. Weaver, R. Zhou
Machine Learning
Explainability from Training with Applications to TCR-Epitope Prediction
Jiarui Li, Zixiang Yin, Samuel Landry et al. (+2 more)
Deep learning models have achieved strong performance in artificial intelligence for science, yet their black-box nature limits our understanding of how they learn scientific tasks. Existing methods for interpretability provide limited insight into how models organize evidence and evolve during learning. We introduce explainability from training (EFT), a model-agnostic paradigm that traces model interpretation during training to explain why models rely on specific features and how they organize these features as predictive evidence. We apply EFT to four state-of-the-art T
cell receptor (TCR)-epitope prediction models, TCR-SRIM, TULIP, MixTCRpred, and NetTCR-2.2, spanning post-hoc and interpret-by-design approaches as well as
transformers and CNNs. To investigate how structural information affects model explanations, we introduce a
benchmark, TCR-XAI2, containing 388 unique experimentally resolved TCR-epitope structures, complemented by structures predicted using AlphaFold3, Boltz-2, TCRModel2, tFold-TCR, and OpenFold3. Using EFT with TCR-XAI2, we demonstrate that (1) CNN and
transformer models exhibit distinct learning trajectories; (2) TCR $\alpha$ and $\beta$ evidence can conflict during learning, limiting the benefits of jointly modeling both chains, while MHC information mitigates this; and (3) real versus predicted structural data for TCR-epitope prediction exhibits distinct TCR and peptide feature preferences as well as differing trajectories of model certainty.
Read the full abstract ▾
All authors Jiarui Li, Zixiang Yin, Samuel Landry, Zhengming Ding, Ramgopal Mettu
Strongly Correlated Electrons
Quantum metric induced nonlinear transport in the hidden loop-current phase of kagome metal RbV$_3$Sb$_5$
Jia-Chen Shi, Shupeng Xu, Utkarsh Khandelwal et al. (+4 more)
A hidden low-temperature phase with possible loop-current order has been proposed in the kagome metals AV$_3$Sb$_5$ (A=K, Rb or Cs), but its experimental signatures remain subtle and indirect. Here, we use third-order nonlinear transport to probe this hidden phase in RbV$_3$Sb$_5$. At ~35 K, the longitudinal cubic response develops a strong kink and pronounced directional anisotropy with a strong departure from common relaxation time scaling, while the transverse cubic response acquires a
magnetic field-odd component at the same temperature. Their coincident onset identifies the third-harmonic response as a sensitive marker of the low-temperature electronic reconstruction with time-reversal broken
symmetry. Crucially, we find that the
quantum metric quadrupole contributes directly to the longitudinal third-order response. In a loop-current charge-density-wave model, its direction selective enhancement captures the observed angular reconstruction. Our work shows how higher-order nonlinear transport translates subtle changes in electronic
symmetry and
quantum metric into measurable electrical signatures in
quantum materials that is important for studying complex electronic phases of matter.
Read the full abstract ▾
All authors Jia-Chen Shi, Shupeng Xu, Utkarsh Khandelwal, Nashra Pistawala, Luminita Harnagea, Steven J. May, Ritesh Agarwal
Computer Vision
StructRL: Online Structured Reinforcement Learning for Long-Horizon Vision-Language-Action Tasks
Ziyi Yin, Sangmin Woo, Kang Zhou et al. (+4 more)
Vision-
language-action (VLA) models perform well on shorter-horizon manipulation tasks but still struggle with long-horizon tasks that require multiple dependent manipulations from a single command. Online reinforcement learning (RL) can improve these policies through environment interaction, yet many existing methods provide reward only after the complete task succeeds. However, such terminal supervision is sparse and does not distinguish early failures from rollouts that make substantial partial progress. We propose StructRL, an online RL framework that constructs structured intermediate supervision from verifiable subtask completions. StructRL decomposes each task into verifiable subtasks, grants intermediate rewards only after the prerequisite subtasks have been completed, and scales each reward according to completion pace. Across RoboCasa365 and LIBERO-Long with GR00T-N1.5 and pi 0.5, StructRL consistently outperforms evaluated online RL baselines. These results show that verifiable, structured intermediate rewards improve long-horizon VLA post-training. Code is available at https://github.com/amazon-science/StructRL.
Read the full abstract ▾
All authors Ziyi Yin, Sangmin Woo, Kang Zhou, Sungyeon Kim, Aosong Feng, Haibo Ding, Jun Huan
Sound
Trigger Sound Suppression for Misophonia
Vaishnavi Vidyasagar, Jasmine Zhang, Mahima Uliyar et al. (+4 more)
Misophonia, a disorder of decreased tolerance to specific sounds, affects 5-20% of the population, yet sufferers have no good options: therapy helps a minority, and earplugs or noise cancellation silence everything. We present a study for neural trigger sound suppression for misophonia, selectively removing trigger sounds. We curate a
dataset covering the 10 most common trigger classes. Using streaming dual-path networks operating on 6 ms audio chunks, we explore both one-hot and multi-hot-conditioned models that suppress 1-3 triggers from the acoustic scene. We validate our model outputs in a listening study with 30 adults with clinically elevated misophonia impairment. Participants reported significantly lower distress and arousal, and improved valence, for suppressed audio.
Read the full abstract ▾
All authors Vaishnavi Vidyasagar, Jasmine Zhang, Mahima Uliyar, Seunghyun Oh, Emily Catherine Gates, Mark Zachary Rosenthal, Shyamnath Gollakota
Optics
Asymmetric imaging: Reciprocity bounds and manifestations
Romil Audhkhasi, Anna Wirth-Singh, Rose Johnson et al. (+6 more)
Recent demonstrations of unidirectional imaging using diffractive multilayers have brought a concept akin to optical isolation to imaging, yet the fundamental limits of such asymmetry have remained unknown. Here we develop a general theory of asymmetric imaging in linear, passive, reciprocal optical systems and derive a fundamental bound on the achievable unidirectionality. Reciprocity requires the forward and reverse transmission operators to have identical singular values, a wave-optical analogue of étendue equality, with a simple consequence: perfect asymmetry is attainable only for sets of $M \leq N/2$ orthogonal object patterns, where $N$ is the number of available object/image modes; for $M > N/2$, the average power asymmetry is bounded by $(N-M)/M$. Guided by this result, we present two complementary realizations: refractive imaging systems formed by bulk lenses and stops that attain the half-occupation limit, and a 1 cm-diameter all-silicon metasurface doublet that exhibits asymmetric focusing at 4 $\mu$m while fully complying with Lorentz reciprocity. We fabricate the doublet by direct
laser writing and experimentally demonstrate its focusing asymmetry. By establishing mechanism-independent limits, our results provide a
benchmark and design principle for asymmetric imaging across platforms, with implications for defense, surveillance and secure communications.
Read the full abstract ▾
All authors Romil Audhkhasi, Anna Wirth-Singh, Rose Johnson, Maksym Zhelyeznyakov, Vladimir Yarmolik, Rafael Piestun, Andrea Alù, Arka Majumdar, Owen Miller
General Economics
Ranking the wrong places: vulnerability assessment, flood losses and risk sharing in Italy and Europe
Stefano Blando
Before a flood, governments decide where to invest in protection; after it, how much of the loss to compensate. The European Union informs the first decision with a composite index that ranks every subnational unit by vulnerability. This paper tests the ranking by comparing Italian provinces struck by the same flood between 2005 and 2024. The published index does not order provinces by the losses they suffer, with or without controls for hazard, exposure and income ($\rho$ between -0.06 and 0.09). The failure lies in what it weights. As published, the index mostly measures economic development, while losses follow water: they arrive in recurring groups of provinces that share a river basin, and more than half of the ranking error lies between those groups. In a pre-registered test that rebuilds it and eight rivals at every forecast origin from 2010 to 2020, the index comes last on both questions an allocator asks. On where damage will land, its top ten provinces capture 16 per cent of later damage, less than a random draw; Italy's public flood hazard map, blind to losses, captures 44. On who loses more from the same water, its rank correlation with later losses is 0.03; its own components, reweighted on past losses, reach 0.42. Neither repair needs new data. The misranked loss is paid locally. Of EUR 12.3 billion recorded over 2013-2022, 85 per cent was not covered by national transfers, and prevention funding, scored the same way, does no better than chance. Because correlated losses cluster within borders, national pools in Europe need 39 per cent more reserve than cross-border pools of the same size. The index fails the same way across 1,142 European units, where the repairs are weaker.
Read the full abstract ▾
Computer Vision
Xiaoyu Wu, Yifei Wang and Chen Wei
Generative and representation learning remain asymmetrically connected: semantic representations are used to improve
diffusion generation, whereas the models' own representations are often treated as a by-product of synthesis. We ask whether
diffusion models can instead be trained to learn substantially stronger semantic representations without sacrificing generation quality. SelfFlow takes a step in this direction by introducing self-supervised patch
alignment into flow matching, but its main gains remain in faster convergence and improved generation. Inspired by DINO and iBOT, we extend this framework with cross-view class-
token alignment to further strengthen semantic representations. Specifically, we form two independently noised, dual-timestep observations of each image and align each student class-
token representation with the stop-gradient EMA-teacher target from the other observation. This objective is optimized jointly with the inherited flow-matching and local patch objectives. Notably, although the additional objective acts only on the class
token, it strengthens both class-
token and patch representations. Compared with a matched two-view baseline, ImageNet linear-probing accuracy improves by 9.4% using the class
token and 10.1% using mean-pooled patch
tokens, while frozen-backbone VOC2012
segmentation improves by 3.6 mIoU. These representation gains are achieved while maintaining comparable ImageNet generation FID. In text-to-image training, the same objective also improves generation FID, reducing it from 2.52 to 2.37 at matched checkpoints. Our results show that representation need not remain a by-product of generation or merely a tool for improving it: it can be directly optimized as a first-class capability of
diffusion pretraining alongside generation.
Read the full abstract ▾
Neural and Evolutionary Computing
Massively Parallel Reinforcement Learning with a Chaotic Reconfigurable Clockless Chip
Eric Oliveira-Gomes and Damien Rontani
Hardware accelerators based on physical dynamical systems offer an attractive route toward energy-efficient reinforcement learning applications. However, their scalability is challenging because it requires many statistically independent
entropy sources. Here, we introduce a quasi-analog decision-making architecture based on asynchronous Boolean networks (or
lattices) implemented on a clockless reconfigurable chip. Each node in the network consists of a single logic element that acts as an autonomous
entropy source. This architecture gives rise to distributed Boolean chaos, in which a spatially coupled network generates parallel streams of chaotic Boolean transitions with very low statistical dependence between nodes. We experimentally demonstrate parallel decision-making on a 1024-armed bandit problem, which is beyond the scale of previous hardware implementations, while significantly improving power-law scaling performance. Separately, we scale the proposed
entropy source to 5120 parallel channels, yielding an aggregate sample generation rate of 2.14 TS/s. Our solution is implemented on a commercial reconfigurable CMOS chip and offers high integration density and ease of programmability. Our results pave the way for using distributed Boolean chaos as a valuable hardware substrate for large-scale reinforcement learning and for the development of fully integrated, high-throughput decision-making accelerators.
Read the full abstract ▾
Game Theory
Solving the Arctic Product-Mix Auction
Elizabeth Baldwin, Paul Klemperer and Edwin Lock
Arctic product-mix auctions allow budget-constrained bidders to express preferences over multiple substitute goods. We give the first strongly
polynomial algorithm to find the competitive
equilibria when the auctioneer has separable convex piecewise-linear costs. It combines a strongly
polynomial reduction to the costless Arctic auction with an existing strongly
polynomial algorithm for costless Arctic auctions. We also give a
polynomial-time Turing reduction from costless Arctic auctions to linear Fisher
markets, and a strongly
polynomial many-to-one reduction to linear Arrow-Debreu exchange
markets. If a good is sold in positive quantity in two competitive
equilibria, its price is the same in both.
Read the full abstract ▾
Geophysics
From Wildfire Severity to Snow Persistence: A Multisource GeoAI Study of the 2020 Creek Fire
Parastoo Farajpoor and Mohammadreza Narimani
Accurate prediction of post-fire snow conditions does not necessarily establish how wildfire changed those conditions. We developed an explainable geospatial artificial intelligence framework combining multisource Earth observations, meteorological information, and matched before-after comparisons to examine seasonal snow persistence following the 2020 Creek Fire in California's Sierra Nevada. Harmonized Landsat Sentinel-2 observations were used to estimate the fraction of clear-sky observations containing snow during October-July for eleven water years, 2016-2026, on a 500 m analysis grid. We matched 3,778 snow-zone
cells inside the fire perimeter to comparable unburned controls using terrain and pre-fire snow conditions; absolute standardized mean differences after matching were no greater than 0.047. HLS persistence agreed closely with MODIS, with a mean annual spatial correlation of 0.947. The landscape-average before-after control-impact contrast was +0.0017, with a year-level 95% confidence interval of -0.023 to +0.027. A stronger response emerged in the highest burn-severity class, where observed persistence increased by 0.026 relative to matched controls, equivalent to 2.6 percentage points. Under 5 km spatial cross-validation, XGBoost predicted raw post-fire persistence with an out-of-fold coefficient of determination of 0.811, whereas predictive skill for the fire-adjusted anomaly reached 0.046. Elevation and temperature together accounted for 65.7% of mean absolute model attribution. These findings distinguish a severity-associated optical snow response from the terrain-
climate relationships that dominate predictive skill. Combining matched comparisons with explainable GeoAI provides a practical framework for forest monitoring that separates accurate environmental mapping from inference about disturbance effects.
Read the full abstract ▾
Computation and Language
DeepRewind: Predicting and Repairing Premature Commitments in Deep Research Agents
Amirhossein Abaskohi, Amirhossein Dabiriaghdam, Lele Wang et al. (+2 more)
Deep-research
agents conduct long-horizon investigations through iterative search, evidence evaluation, belief revision, and synthesis. However, they may commit to claims before sufficient evidence is available, causing later
reasoning to reinforce an incorrect interpretation. We introduce DeepRewind, an additive control layer for reversible deep research that represents the
agent's evolving epistemic state as a typed graph of sources, evidence, claims, hypotheses, assumptions, commitments, plans, and drafts. Before accepting an intermediate conclusion, a
prompt-based world model predicts its impact and estimates reversibility based on hypothesis narrowing, information loss, recovery cost, and contradiction-trigger coverage. A binary controller blocks risky commitments, while a consistency monitor performs dependency-aware rollback when later evidence invalidates them. Across DRBench and LiveDRBench, DeepRewind improves insight recall by 3.6 percentage points and reduces premature commitments by 59.1% relative to Open Deep Research.
Read the full abstract ▾
All authors Amirhossein Abaskohi, Amirhossein Dabiriaghdam, Lele Wang, Peter West, Giuseppe Carenini