Peer Revenue Now accepting positions

Papers

Fresh from arXiv, typeset properly. Highlighted words trade on the market.

All of arXiv

Saturday 7 November

Artificial Intelligence

BunshinRL: Whole-Workspace Forking for Agentic Reinforcement Learning Rollouts

Agentic reinforcement learning (RL) rollouts interleave model reasoning with tool calls in persistent workspaces. Scaling these rollouts incurs substantial memory costs for concurrent environments, while slow tool calls delay entire rollout batches. We present BunshinRL, an agentic RL rollout system that exploits shared environment state and recurring tool calls through three techniques: memory sharing across environments, tool-prefix caching, and speculative tool execution. These techniques build on BunshinFork, a novel live whole-workspace forking mechanism for containers. BunshinFork reconstructs process trees, shares memory copy-on-write, and preserves file, communication, and GUI state, enabling cached-state reuse and rollback after incorrect speculation. We implement BunshinRL and BunshinFork through Linux kernel module and CRIU modifications. BunshinRL reduces peak memory by up to 43%, total memory consumption by up to 51%, and end-to-end rollout latency by 23%.
Read the full abstract

All authors Yutong Huang, Vikranth Srivatsa, Alex Asch, Yiying Zhang

Nuclear Theory

Role of the scission-line structure in the formation of the fragment-mass distribution in $^{180}$Hg fission

The fission of $^{180}\mathrm{Hg}$ is investigated by studying the stochastic evolution on the potential energy surface calculated within the multidimensionally constrained relativistic Hartree–Bogoliubov (MDC-RHB) model. The MDC-RHB model provides the zero-temperature potential-energy surface in the $(\beta_{20},\beta_{30})$ plane and self-consistent density distributions, from which scission configurations are identified using a geometrical neck criterion. The evolution toward scission is described by a master equation on the same deformation mesh, with transition rates determined by deformation-dependent microscopic level densities. The level densities are obtained in a finite-temperature BCS treatment using the MDC-RHB single-particle spectra and zero-temperature pairing gaps as input. The effective barrier-top region is found to lie close to the scission line, making the detailed energetic structure of the scission boundary particularly important for the fragment distribution. The potential-energy landscape and scission-line structure strongly favor reflection-asymmetric configurations with pronounced prefragment deformations and lead to fragment-mass distributions centered close to the experimentally observed $A_{\mathrm L}/A_{\mathrm H}\simeq80/100$ division. An additional constrained $\beta_{40}$ analysis indicates that fluctuations beyond the $(\beta_{20},\beta_{30})$ collective space can generate an additional spread of the fragment masses. With increasing excitation energy over the range $E^*=34.0$–$65.5$ MeV, the symmetric component is enhanced, while the asymmetric fission mode remains dominant.
Read the full abstract

All authors Jin-Tian Liu, A. Rahmatinejad, Jie Zhao, T. M. Shneidman, N. V. Antonenko, Shan-Gui Zhou

Robotics

Grip, Trace, Write: Force-Aware Imitation and Reinforcement Learning for High-DoF Robotic Handwriting

Most robotic handwriting rigidly mounts the pen and reduces writing to trajectory tracking. A freely held pen instead couples grasp stability, tip motion, and contact force. We present Grip, Trace, Write, a force-aware simulation environment and learning framework for high-DoF handwriting. A two-branch actor separates pen retention from arm-wrist writing while preserving their closed-loop coupling. Training learns a grasp controller, initializes writing from inverse-kinematics demonstrations by behavior cloning, and improves writing through reinforcement learning with rewards on ink, force, time, and retention. Across diverse tasks, the policy improves completion, precision, force regulation, and efficiency over behavior cloning and inverse kinematics, and generalizes without fine-tuning to unseen English and Chinese content. These results establish force-aware handwriting as a testbed for dexterous tool use.
Read the full abstract

All authors Rongqian Chen, Yiding Chen, Jijie Zhou, Tom Z. Jiahao

Computation and Language

The Truth Beneath the Tokens: Prompt-Space Recovery of Knowledge Edits in Language Models

Model editing provides an efficient way to update factual knowledge in transformer models. Recent approaches reverse edits by modifying, masking, or decomposing model weights, but make recovery dependent on weight-space interventions. We instead show that pre-edit knowledge can be recovered through input-space interventions alone, without modifying the edited model's weights. We introduce Memento, a prompt-space mechanism for auditing residual knowledge that learns a reusable recovery signal to steer an edited model toward its pre-edit response. Memento supports both continuous recovery prefixes and discrete textual prompts for standard text interfaces. Across three LLMs and five editing techniques, Memento recovers 80.07% of held-out edited facts, achieving a 9.58% relative improvement over baselines, suggesting that editing may suppress the facts without fully eliminating their recoverability. We further investigate whether prompt-space recovery extends beyond model editing. On pristine, unedited LLMs, Memento improves average factual recall by 53.9% relative to standard inference. These results demonstrate the potential of prompt-space recovery for auditing residual knowledge in edited models and investigating the gap between factual knowledge and its accessibility during inference. We will release our source code.
Read the full abstract

All authors Manit Baser, Rachel Yeo Hui Min, Aditya Nawal, Dinil Mon Divakaran, Mohan Gurusamy

Mesoscale and Nanoscale Physics

Extended No-Go Theorem and Its Verification in Spin-Selective Transport through Chiral Cyclodextrin Assemblies

The classical two-terminal single-channel no-go theorem relies on time-reversal symmetry (TRS) and scattering-matrix unitarity. In open systems, omitted channels or environment leakage break unitarity, so TRS and the symplectic reciprocity symmetry TRS\(^{\dagger}\) are no longer equivalent, invalidating the classical premises. We prove that if the two-terminal single-channel scattering matrix satisfies TRS\(^{\dagger}\) plus central inversion symmetry, transmission remains no spin selectivity; this result provides an important complement to the classical no-go theorem in nonunitary regime. When these conditions are violated or the single-channel constraint is lifted, spin-selective transport occurs. In charge transport through devices based on chiral \(\gamma\)-cyclodextrin molecular assemblies, we verify separately the classical no-go theorem in unitary case and its supplementary version together with its multichannel corollary in nonunitary case: when the conditions for the corresponding no-go theorem hold, the chirality-induced spin selectivity (CISS) of the assembly disappears, whereas when they are violated, CISS emerges. These results extend the no-go theorem to nonunitary scattering systems and establish an effective theoretical framework for studying CISS in real molecular assemblies.
Read the full abstract

All authors Kai-Yuan Zhang, Shu-Zheng Zhou, Xi Sun, Hua-Hua Fu

Computer Vision

ESTRA: Enabling Extreme Attention Sparsity for Training-Free Video Diffusion Acceleration

Video diffusion transformers generate high-quality videos, but attention over long spatiotemporal sequences makes inference computationally expensive. Sparse attention reduces this cost by computing only a subset of query-key interactions exactly, and recent training-free methods further approximate the remaining interactions. These methods, however, degrade quickly at high sparsity, which limits the achievable speedup. We present ESTRA (Extreme Sparsity Through Refined Attention), a training-free acceleration method that preserves generation quality even at extreme sparsity. ESTRA concentrates exact attention computation on a small set of important key-value tokens and approximates the remaining contributions with cluster representatives. To identify these tokens efficiently, it groups similar queries and uses their representatives to score individual keys. A variance-based correction further refines the attention weights of the cluster representatives, improving the accuracy of the approximation. Experiments on Wan2.1-14B and HunyuanVideo-13B show that ESTRA establishes a new quality-efficiency frontier for training-free sparse attention, achieving $3.26\times$ and $4.71\times$ end-to-end speedups at 5% attention density while surpassing Sol-Attn and SVG-EAR in fidelity to dense generation and on most VBench dimensions. Even at 1% density, ESTRA remains more faithful to dense generation than Sol-Attn at 12.5%, while reaching up to $5.39\times$ speedup.
Read the full abstract

All authors Enhuai Liu, Yunke Wang, Yutong Wang, Changming Sun

Econometrics

Good Looks and the Academic Job Market: Measuring Attractiveness with Vision Embeddings

We introduce a new approach to measure physical attractiveness—the relative vision embedding (RVE) score. The score maps a face to a vector in the embedding space of a pre-trained deep vision model and measures its proximity to a reference set of celebrity faces. It requires no additional training or fine-tuning, no hand-crafted features, and no human raters; it is continuous, inexpensive, scalable, and fully reproducible. Using publicly available photographs with human attractiveness ratings, we show that the RVE scores correlate strongly with human ratings and that rankings are consistent across several off-the-shelf vision models. We apply the measure to administrative hiring data from a research-oriented economics department, complemented by hand-collected photographs of the job applicants. Higher RVE scores are associated with better job-market outcomes, though the association is not robust across all specifications. In contrast, the association with reference-letter length and sentiment is strong and robust: candidates with higher RVE scores receive longer letters written in more favorable language, and this sentiment gradient is steeper for female candidates. The pattern is consistent with appearance affecting academic hiring in part through letters of recommendation.
Read the full abstract

Plasma Physics

Self-organization of magnetic islands and zonal magnetic fields driven by slab microtearing modes in the absence of magnetic shear

Self-organization in collisional microtearing turbulence is investigated using gyrokinetic simulations in a slab geometry without equilibrium magnetic shear. Unlike magnetically driven conventional tearing instabilities, microtearing instabilities are driven by the electron temperature gradient. Electron temperature fluctuations are found to contain most of the gyrokinetic free energy. Although magnetic perturbations contribute little to the free energy, the magnetic field structure plays a crucial role in the nonlinear evolution of microtearing turbulence. Zonal magnetic fields emerge spontaneously through the nonlinear interaction between magnetic perturbations and electron temperature fluctuations. These zonal fields provide an effective magnetic shear that enables magnetic islands to form despite the absence of equilibrium magnetic shear. The electron temperature profile is flattened within the islands. This flattening reduces the instability drive and is expected to promote turbulence saturation.
Read the full abstract

Artificial Intelligence

DiSARM: Diverse and Surprising Association Rule Mining with Large Language Models

Exploratory Data Analysis (EDA) aims to discover informative patterns in large datasets. For transactional data, Association Rule Mining (ARM) is an established approach that facilitates EDA by efficiently generating candidate patterns. Beyond efficient mining, identifying informative rules has also been extensively studied through both objective and subjective interestingness measures. Recent advances in Large Language Models (LLMs) offer new opportunities for assessing subjective interestingness using world knowledge. Nevertheless, relying solely on LLMs raises concerns about scalability and statistical reliability. Building on prior ARM research, we propose a hybrid framework, DiSARM, that leverages the world knowledge of LLMs to assess the subjective interestingness of rules, while maintaining statistical reliability and promoting diversity among the selected rules. Across three transactional datasets, experiments demonstrate that DiSARM effectively combines statistical reliability with surprise and diversity, offering a promising approach to ADE from transactional data.
Read the full abstract

Differential Geometry

Uniqueness and stability of Appleton's solitons

We prove that Appleton's noncollapsed steady gradient Ricci solitons on complex line bundles over positive Kähler–Einstein manifolds are the unique such solitons arising from his metric ansatz. We then show that the four-dimensional examples on $O_{\mathbb{CP}^1}(-k)$, $k \geq 3$, are strictly linearly stable. By a result of Deruelle, this implies dynamical stability under the Ricci flow.
Read the full abstract

Robotics

Spatial Representation Distillation and Knowledge Routing for Vision-Language-Action Models

Understanding three-dimensional physical structure has become essential for vision-language-action models (VLAs). However, existing approaches either rely on explicit geometric inputs at inference time or adopt a single visual pathway, potentially compromising the visual-semantic knowledge inherited from pre-trained models. We present SPARK-VLA, Spatial Representation distillation and Knowledge routing for VLAs, a preserve-specialize-route framework for spatially adapting VLAs without requiring external 3D inputs or experts at inference. SPARK-VLA consists of two key components: (1) spatial representation distillation, which transfers 3D understanding from a frozen 3D expert to a LoRA-adapted visual branch by aligning attention responses while preserving the original visual pathway, and (2) a lightweight knowledge router, which selectively forwards action-relevant visual tokens to the language model using routing supervision derived from action-loss gradients. Across LIBERO, LIBERO-Plus, CALVIN, and real-world manipulation, SPARK-VLA achieves strong performance with only a 0.5B language backbone and approximately 53M trainable parameters, providing an effective and parameter-efficient approach to 3D-aware VLA learning. Code is available at https://github.com/park-jungin/spark-vla.
Read the full abstract

All authors Jungin Park, Chaoran Zhu, Changjae Oh, Kwanghoon Sohn

Superconductivity

Transverse quasiparticle heat transport in p-wave magnetic Josephson junctions

We study transverse heat currents carried by Bogoliubov quasiparticles in $p$-wave magnetic Josephson junctions under a longitudinal temperature bias. We find that such a junction supports a transverse spin heat current while the net transverse heat current vanishes. This effect stems from the $p$-wave magnetism, which shifts the two spin sectors oppositely in transverse momentum space and thereby causes them to carry opposite heat currents. A weak Zeeman field lifts the compensation between the two spin-sector contributions, thereby generating a finite net transverse heat current. The directions of both currents can be switched by reversing the sign of the $p$-wave magnetic splitting or tuning the superconducting phase difference, while both currents vanish in the absence of $p$-wave magnetism. These findings highlight $p$-wave magnetic Josephson junctions as a versatile platform for switchable transverse heat transport and controllable spin caloritronics.
Read the full abstract

Computer Vision

Split and Bridge: Multimodal Generation via Diffusion Bridging

Visual data are often expressed through multiple complementary modalities (e.g., images and segmentation masks), each capturing different aspects of the same underlying structure. Most existing multimodal methods adopt a top-down paradigm, learning a shared latent variable through joint training from scratch, which limits scalability. We propose Split and Bridge (SnB), a bottom-up framework for multimodal generation that couples pretrained unimodal diffusion models at sampling time. Rather than learning a unified latent space, SnB generates coherent multimodal samples by following a shared diffusion trajectory up to a split point and then branching into modality-specific trajectories guided by a bridge module. This design enables plug-and-play reuse of powerful pretrained models without joint training, making it scalable and easily adaptable. Across standard benchmarks (PolyMNIST, CelebAMask-HQ) and more complex settings (PIE-Bench++, ROCO), SnB achieves a strong balance between sample quality and coherence. We show that SnB naturally extends to downstream zero-shot tasks such as interleaved co-generation, image editing, and style coupling/decoupling. Our results demonstrate that bottom-up generation provides a practical framework that complements existing top-down multimodal approaches. Code is available at https://github.com/AhmadArrabi/SnB
Read the full abstract

All authors Ahmad Arrabi, Xiaohan Zhang, Xingyu Li, Safwan Wshah

Data Structures and Algorithms

Lower Bounds Above $\sqrt{3}$ for Online Makespan Minimization on Four Identical Machines

For deterministic online makespan minimization, $m = 4$ identical machines is the smallest case whose optimal competitive ratio is open: it is known only to lie between $\sqrt{3} \approx 1.73205$ and $26/15 \approx 1.73333$. We prove a lower bound of $14383/8304 \approx 1.73206$ using a single branching layer, and refine the construction with a second, mirrored branching layer to obtain $14350/8283 \approx 1.73246$, closing about a third of the gap. Our constructions are the first branching adversaries for $m = 4$ machines, and the first explicit lower bounds for online makespan minimization in which the adversary presents jobs in non-monotone order of size.
Read the full abstract

Computation and Language

Who Remembers What? Multi-Granularity Memory Extraction for Multi-User, Multi-Agent Interaction

Self-evolving agents improve their own skills, memory, and harness code from experience without retraining the base LLM, and group-evolving agents go further by sharing these lessons across agents. Yet most such systems overlook that sharing across users and heterogeneous agents requires deciding who can benefit from each lesson and what information must remain private. We present MixUA, a memory system that distills independent user-agent trajectories into lessons and assigns each lesson to a user-agent pair, user, agent, or global scope according to whom it benefits and what it may disclose. For evaluation, we introduce CrossLearnBench, a mixed-user-agent benchmark that pairs established agent tasks with user preferences and privacy probes that test whether private facts leak into memory other users can read. With five users and seven agents, MixUA raises pass rates by up to 16 points over no memory and 5 over a single global memory, even on tasks previously tried only by other users on other agents. It retains up to 90% of previously stated preferences, and at most 0.02% of privacy probes find a planted private fact in a forbidden scope. We hope MixUA offers a scalable path toward a global system where users and agents co-evolve.
Read the full abstract

All authors Yunzhe Wang, Mesut Erhan Unal, Sheng Zhang, Shabnam Ghadar, Peng Tang

Dynamical Systems

An introduction to discrete dynamics in condensed sets: orbits, periodic loci, and arithmetic examples

We formulate discrete dynamics internally in the topos Cond of condensed sets. For an endomorphism $f\colon X\to X$, we construct orbit quotients, backward histories, preimage trees, and periodic loci as objects evaluated on profinite parameter spaces. The framework recovers global-section dynamics, fully faithfully embeds compact Hausdorff systems, characterizes the invertible extension, and is compatible with inverse limits. It organizes $p$-adic congruence towers and shows, for a Cantor-type non-Archimedean Julia set, that classical periodic points can be dense although the internal periodic locus is not isomorphic to the compact condensed Julia set.
Read the full abstract

Combinatorics

Bulgarian solitaire with distinguishable cells

Classical Bulgarian solitaire tracks only the sizes of the stacks. We ask how the lengths of its periodic orbits change when individual cards are distinguished. To study this question, we introduce four actions $T_1,T_2,T_3,T_4$ on partitions with labeled cells. The actions differ in whether the top or bottom cell is removed from each column and in the order in which the removed cells are arranged to form a new column. We determine the exact sizes of the periodic orbits for all four actions. The main work lies in the analysis of $T_2$, for which we develop a coloring algorithm on Ferrers diagrams. The coloring encodes the cycle structure of the permutation of the labels induced by one full period of the underlying partition.
Read the full abstract

Numerical Analysis

High-order long-time stable methods for the forced incompressible Navier–Stokes equations

We present in this paper two scalar auxiliary variable (SAV) correction steps for constructing long-time stable schemes for the forced incompressible Navier–Stokes equations with no-slip boundary conditions. A main feature is that the same correction steps apply to any linear predictor satisfying the order and spatial regularity requirements in Assumption~P. Their construction and stability analysis are independent of the particular form of the predictor and require no energy estimate for it. The first correction uses a damped, shifted kinetic-energy variable to obtain a uniform-in-time $L^2$ bound in two and three dimensions, together with a weighted dissipation estimate. In two dimensions, the second correction combines a higher-energy auxiliary variable with a scalar relaxation to yield a uniform-in-time $H^1$ bound. Under the stated assumptions, these estimates hold on arbitrary positive time grids, without restrictions on step sizes or adjacent step ratios. We also derive eventual absorbing bounds; for the two-SAV scheme, these apply to the admissible initial class associated with fixed parameters. The generalized BDF schemes require one constant-coefficient Helmholtz solve per velocity component and one pressure Poisson solve per step; the second correction adds no field solve. Numerical experiments exhibit second-, third-, and fourth-order convergence for the tested BDF schemes and illustrate the long-time stability and efficiency of adaptive time stepping.
Read the full abstract

Robotics

Multi-Objective Adaptation After Deployment: A Position on World Models for Remote Robot Teams

Autonomous robot teams deployed in remote, unstructured environments may face changes in the environment, hardware, and team composition that degrade previously effective policies. Shifting mission objectives can be handled by selecting a different solution from an existing Pareto set, but shifting deployment conditions can reorder the set itself, making previously dominated policies competitive or new policies necessary. However, evaluating these policies in new contexts while minimizing risk to the hardware remains an open problem. We argue that, although largely untested in this domain, world models merit investigation for post-deployment multi-objective adaptation. By predicting policy outcomes under updated estimates of current conditions, world models could enable teams to re-evaluate existing policies without physically executing each candidate, provided predicted performance changes reflect those realized on deployed systems. We identify research priorities for model-based policy re-evaluation, validation of predicted against realized performance, and incorporating shared information into world models to support coordinated team adaptation.
Read the full abstract

Plasma Physics

Magnetosonic Solitons Driven by a Moving Charged Source: Two-Fluid Modeling and PIC Validation

We develop and validate a one-dimensional electron-ion two-fluid model, self-consistently coupled to Maxwell equations, for magnetosonic structures generated by a moving charged source. The model is independently benchmarked against kinetic particle-in-cell (PIC) simulations to assess its ability to reproduce electrostatic and electromagnetic responses. In the zero-magnetic-field limit, the electrostatic precursor structures are validated using the Korteweg–de Vries amplitude–width invariant. For a finite magnetic field, the frequency and propagation speed of the far wake are consistent with the low-frequency fast-magnetosonic branch, providing a linear benchmark for the electromagnetic response. The model reproduces the transition from a wake-dominated regime to localized forward precursors as the source velocity exceeds the magnetosonic speed, and to source-locked pinned structures at higher velocities. Independent PIC simulations recover the principal features of these regimes, supporting the validity of the two-fluid description for the parameters considered. By retaining separate electron and ion dynamics, electron inertia, and charge-separation effects, the validated model provides a framework for investigating source-driven plasma-wave structures across a broader range of plasma conditions and source-magnetic-field geometries, with potential applications to interpreting plasma-wave signatures associated with moving charged objects in space plasmas.
Read the full abstract

Machine Learning

HexIt: Learning Hexahedral Block Decomposition with Expert Iteration on Combinatorial Maps

Decomposing a polyhedral solid into a few well-shaped hexahedral blocks is a long-standing open problem in mesh generation. We cast it as a sequential decision problem on a 3-dimensional generalised map: an agent splits edges, faces and volumes with three insert operations until every cell is a hexahedron and the mesh passes a quality threshold, and a policy that convolves over the map's involutions chooses each insert. A successful decomposition takes over a hundred decisions, and a uniformly random policy never finds one. We clone a policy from constructed decompositions and improve it by expert iteration. On held-out chamfered solids that share no boundary topology with the cloning data, the final policy solves an expected $94.4\%$ in $20$ samples, and expert iteration raises the best-of-20 solve rate on large chamfered solids from $0.203$ to $0.355$. At matched compute, PPO, GRPO and VinePPO updates all make the cloned policy worse.
Read the full abstract

Artificial Intelligence

D-REAL: Acceptance-Length Optimization with Unbiased Gradients for Speculative Decoding

Speculative decoding efficiency depends on the number of consecutively accepted draft tokens, yet token-level training objectives do not directly optimize this quantity. We introduce D-REAL (Drafter Reinforcement of Expected Acceptance Length), a post-training method that directly optimizes expected acceptance length for a fixed decoding context. D-REAL weights position-wise draft-target distributional overlap by prefix survival probabilities along a draft-sampled prefix, faithfully capturing the sequential dependence of verification on earlier proposals. By accounting for the parameter dependence of the proposal distribution, we derive an unbiased gradient estimator for this single-round objective under stochastic verification. We evaluate D-REAL on DSpark and DFlash2 with Qwen3-8B as the target model across ten benchmarks. Starting from baselines whose mean acceptance length has plateaued, one epoch of D-REAL post-training consistently improves the benchmark-averaged mean acceptance length by up to 0.23 tokens across both architectures under greedy (T = 0.0) and stochastic (T = 1.0) sampling, translating into up to a 9.4% end-to-end serving throughput improvement.
Read the full abstract

All authors Meng Wei, Yefei Chen, Antong Li, Yanan Gao, Xingliang Shi, Xin Gao, Zhiling Ye, Yun Yue, Chunxiao Guo, Peng Wei

High Energy Astrophysical Phenomena

Faint Thermonuclear Transients from the Double Detonations of Common White Dwarfs

The double detonation model is a viable channel to explain many observed Type Ia supernovae (SNe Ia), with thin helium shelled white dwarf (WD) progenitors of 0.9–1.1 M$_\odot$ reproducing normal SNe Ia observables. Recent work has also shown that all C/O WDs $\leq$ 1.0 M$_\odot$ can sustain a helium shell detonation that triggers a detonation of the underlying core. As WDs of lower masses (0.5–0.7 M$_\odot$) far outnumber those at higher mass, it is important to explore their double detonation outcomes. To that end, we conducted multidimensional double detonation simulations of low mass C/O cores with very thin helium shells, finding explosion outcomes that differ non-trivially from those at higher masses. Not surprisingly, their cores are incompletely burned with low yields of radioactive elements. Though still formally unbound in isolation, some of the ejecta could remain bound to a nearby companion. In addition, the offset initiation of the core detonation wave and its failure to burn through the entire core lead to a very asymmetric distribution of $^{56}$Ni that creates dramatically viewing angle-dependent light curves and spectra. We find that the peak bolometric luminosities for the same explosion vary up to nearly 2 magnitudes over the sky, and that the spectra and light curves from these models are within the range of faint SNe Iax. In particular, our 0.50 M$_\odot$ models agree well with the photometry of SNe Iax 2008ha and 2024vjm.
Read the full abstract

Robotics

Deploying Physically and Epistemically Compliant Agents by Distilling Belief-Space Task and Motion Planners

In this work, we train agents to act under uncertainty according to specified, belief-dependent rules. Deep reinforcement learning (RL) can in theory train agents to follow rules if encoded into well-designed rewards, but in practice, RL fails to perfectly optimize these rewards, leading to ineffective or noncompliant agent behavior. Belief-space planners by construction produce behavior in line with developer specification, but require hand-programming online planners and belief-state estimators, adding significant engineering complexity at scale. We combine the two: we plan offline to generate a dataset of sample trajectories and then distill the plans into a policy that acts on raw observations. We propose two changes to planner distillation that address problems specific to history-based policies: Dual Averaging, a loss that combines per-timestep predictive loss and average loss on task changes, and $\rho$-Optimal Reinforcement, a feedback mechanism that reinforces the policy's own successful rollouts when these are near-optimal. We provide a sufficient condition under which static search problems with noiseless observation models is complete, and show that maximum-likelihood-observation planning implies this condition in static search with continuous beliefs. We create a challenging new domain, Scavenger, where an agent needs to gain information in order to succeed. We evaluate our distilled policy in two versions of Scavenger, one where information is provided visually and one where it is provided as a command in natural language. In either case, the learned policy attains a success rate above 90% in Scavenger, compared with at most 5% for RL. We also find that distilled planners transfer with greater than 60% success to following unseen and intentionally vague commands and greater than 75% success to following floorplans larger than those seen in training.
Read the full abstract

All authors Phillip J.K. Christoffersen, Rachel Ma, Patrick Barragán, Dylan Hadfield-Menell

Machine Learning

Applied bifidelity material response model construction and uncertainty quantification with a sampling hypernetwork

Multifidelity modeling offers a practical route to accurate prediction and uncertainty quantification when high-fidelity data are scarce but lower-fidelity data are comparatively abundant. This work develops a bifidelity Langevin sampling hypernetwork framework that constructs probabilistic neu- ral response models by learning a posterior distribution over model parameters rather than a single deterministic fit. The method combines a flexible shared-latent data representation for low- and high-fidelity responses, enabling parameter sharing and transfer learning between the fidelities. We introduce several methodological enhancements to improve the robustness and efficiency of this ap- proach, including stability-promoting drift models, staged and conditional training strategies, adap- tive initialization of Langevin trajectories, and training procedures designed to better exploit corre- lations between fidelity levels. These additions improve convergence behavior, reduce the sampling effort required to obtain representative posterior ensembles, and provide more reliable predictive uncertainty in data-limited regimes. The resulting framework makes more effective use of available multifidelity data while preserving interpretability through the learned posterior structure, yielding improved predictive performance, better uncertainty characterization, and a more stable workflow for constructing uncertainty-aware surrogate models. The method is demonstrated on experimental, high-fidelity simulation and illustrative synthetic data.
Read the full abstract

All authors Reese E. Jones, Cosmin Safta, Dan Bolintineau, Craig M. Hamel, Gianluca Geraci, Sharlotte LB Kramer

High Energy Astrophysical Phenomena

Current Closure and Gamma-Ray Signatures of Spark-Gap and Current-Sheet Activities in Radiative Black Hole Magnetospheres

We investigate the evolution of an axisymmetric black hole magnetosphere using general relativistic particle-in-cell simulations, focusing on its global current structure, variability, and radiative effects. We include inverse-Compton (IC) scattering of ambient photons and subsequent two-photon pair production, and compare simulations with and without synchrotron-curvature (SC) radiation reaction. Starting from an axisymmetric vertical magnetic field, the system develops a polar spark gap and a variable equatorial current sheet that together sustain the global poloidal current. For the parameters explored in this study, we observe that the spark gap oscillates upon the onset of SC cooling, whereas the current sheet is less sensitive to radiative cooling. Time-dependent electric fields in the return-current layer trigger magnetic-field rearrangement followed by magnetic reconnection. Cross-correlation analysis finds no significant correlation between the two regions' variability, suggesting dynamical independence. We further perform post-processing ray-tracing calculations to evaluate the radiative efficiencies of the IC and SC processes. High-energy particles dominate the bolometric luminosity, which reaches up to 1 percent of the Blandford-Znajek luminosity for SC emission and 0.1 percent for IC emission. The relative IC luminosity to SC is higher at lower viewing angles because particle momenta are more anisotropic near the pole than near the equator, due to efficient acceleration along the local magnetic field lines. Together with energetic estimates, these results suggest that both regions can contribute to high-energy emission in active galactic nuclei, with spark-gap photons reaching the sub-TeV regime.
Read the full abstract

All authors Koki Kin, Shota Kisaka, Kenji Toma, John M. Mehlhaff, Benoît Cerutti

Materials Science

Multifractality in sputtered crystalline and amorphous transition metal nitride thin films

In this work, we present evidence for multifractal growth in Titanium and Niobium nitride thin films on different substrates. Self-similar multifractal structures such as dendrites and multi-branched snowflake-like structures are reported. The fractal dimensions are controlled by a combination of processing parameters and conditions at the film-substrate interface. Significantly, fractal growth is observed not only in crystalline films on crystalline and amorphous substrates but also on amorphous films on crystalline substrates. The 2D Multifractal Detrended Fluctuation Analysis (2D MFDFA) carried out on the scanning electron micrographs, clearly reveals the multifractal nature of the nitride thin films. The statistical moments were varied between -5 to +5. The Hausdorff dimensions for the TiN films are in the range of 1.65 to 1.79 while it is 1.64 to 1.74 for the NbN thin films. It is proposed that the observed multifractality is due to a competition between morphological instabilities and Diffusion-Limited Aggregation
Read the full abstract

All authors K. Anantha Padmanabhan, Kuraganti Vasu, P. Manimaran, M. Ghanashyam Krishna

Algebraic Geometry

Extendability of general prime $K3$ surfaces without Gaussian maps and classification of prime Fano threefolds

In arXiv:2409.03960 an approach to the question of extendability of projective varieties via degeneration to ribbons was introduced, and in arXiv:2503.12661 it was used to give a new proof of optimal results on the extendability of general non-prime $K3$ surfaces and the classification of non-prime Fano threefolds. In this article we treat the prime case. We give a new proof of the optimal results on the extendability of general prime $K3$ surfaces of every genus, of the classification of prime Fano threefolds of Picard rank one due to Fano, Iskovskikh and Mukai, and of the irreducibility of their Hilbert schemes. This involves degenerations of $K3$ surfaces to ribbons on embedded rational normal scrolls, called $K3$ carpets. We compute exactly the cohomology of the twisted normal bundle of these $K3$ carpets, instead of computing coranks of Gaussian maps of the canonical curve sections. The key challenge is to analyze the coboundary map of the sequence relating the normal bundle of the $K3$ carpet to the normal bundle of the scroll. We give a deformation theoretic interpretation of this map. Given such a $K3$ carpet and its hyperplane section, which is a canonical ribbon, we show that the coboundary map corresponds to the obstruction to lifting a first order deformation of the underlying scroll fixing the underlying rational normal curve to a first order deformation of the $K3$ carpet fixing the canonical ribbon. To compute the obstruction, we use a result on extensions of non-split ribbons on any curve to a unique $K$-trivial carpet in a canonical way.
Read the full abstract

Cryptography and Security

Governed Self-Improvement for a Local-First Security Analyzer: Baseline Measurements and a Preregistered Evaluation Protocol for Tantular Cyber

Local-first security tools let organizations analyze code without sending it to third-party services, but the small language models they can run locally are weaker than hosted ones and are hard to improve without breaching the privacy guarantee that motivates them. We present Tantular Cyber, an open-source, read-only repository security analyzer that keeps deterministic scanner findings authoritative and confines a locally served language model to advisory explanations in Bahasa Indonesia. We report its measured v0.3.0 baseline on a synthetic, ground-truth benchmark: combining Semgrep with an interprocedural call-graph scanner yields precision 0.843 and recall 0.915 (Semgrep alone: 0.818 and 0.766). The advisory layer is the weak point: 57% of explanations met an evidence-grounding check, against 95% for a different 8B model under identical prompts. A forensic re-scoring of retained outputs shows that every grounding failure cites a code identifier absent from the evidence shown, and that a prompt change which appeared to raise grounding from 0.56 to 0.72 did so partly by citing nothing. We document two evaluation-validity hazards we encountered: silent model-file migration during a run and non-reproducible published figures. Building on these lessons and on constraint-aware evaluation from security operations research, we propose a governed self-improvement loop (code-free feedback, curated data, a cheapest-first intervention ladder from prompts to LoRA and weight-space edits, hard gates, and human release approval) and preregister the protocol by which its first cycle will be judged. We report no results for the loop itself.
Read the full abstract

Biomolecules

Efficient and Accurate Zero Shot Generation of Symmetric Protein Complexes

Symmetric organization is a fundamental principle of biomolecular complex assembly. The same principle underlies engineered applications, from vaccine platforms to nanomaterials, making symmetric de novo generation a core capability for computational protein design. We establish a theoretical foundation equating generation of an assembly with underlying symmetry group $G$ to generation within a quotient space over $G$. This directly allows $\mathrm{SE}(3)$-equivariant models to generate symmetric structures with minimal distribution shift. With this insight we introduce ZeUS ($\textbf{Ze}$ro-shot $\textbf{U}$nified $\textbf{S}$ymmetrization), a framework that converts any pretrained $\mathrm{SE}(3)$-equivariant backbone generative model into an efficient generator of symmetric assemblies with no retraining. ZeUS operates on a single $\textit{canonical subunit}$ and utilizes $\textit{phantom subunits}$ to preserve internal model operations, reducing memory and runtime cost by $\mathcal{O}(|G|)$ for finite groups $G$, compared to a full-complex forward pass. We implement ZeUS for IPA-based models (FrameDiff, FoldFlow) and axial-attention-based RFdiffusion, showing high generation efficiency with minimal degradation to generation quality. We then apply ZeUS to generate symmetric complexes of over $10{,}000$ residues with exact, untruncated attention, a scale previously intractable with any model.
Read the full abstract

All authors Rory Gao, Yuanzhou Chen, Prajit Rajkumar, Wei Wang