Peer Revenue Past citations are no guarantee of future returns

Papers

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

Computer Science

cs Watch Computer Science (sign in to use)

Friday 14 August

Numerical Analysis · Cross-listed from physics.optics

Full-field and Bloch-periodic-factor discretizations: Accuracy and phantom modes

One can distinguish four major formulations of the Bloch-mode problem in linear periodic media. First, there are two common complementary ways of slicing the Bloch variety (Bloch wave vectors paired with the corresponding frequencies): frequency vs. wave vector or, conversely, wave vector vs. frequency. This paper deals exclusively with the latter option, directly applicable to lossy media, evanescent waves, and complex band structures. A separate crossroads is the full field (FF) vs. the lattice-periodic Bloch factor (PF) formulations. These are fully equivalent on the continuous level, but common discretization techniques may break this equivalence not just approximately but qualitatively. In the FF problem, the primary unknown eigenvalue is the Bloch phase factor (not the wavenumber), which enters only through opposite-boundary coupling. The PF formulation, on the other hand, injects the Bloch wavenumber into the differential operator and leads to a volume quadratic pencil. Consequently, standard PF discretizations of the type considered here, in contrast with the corresponding FF discretizations, violate the reciprocal-lattice (Brillouin-zone-shift) covariance – which, as the theory and numerical examples in the paper show, may lead to inaccurate or even nonphysical modes. From the mathematical perspective, the paper highlights the role of structure-preserving algorithms. The practical importance of the results and recommendations stems from the wide adoption of PF formulations in optics, photonics, and beyond – such as the computation of propagating and evanescent Bloch modes in photonic and acoustic structures, topological photonics, effective-mass and topological insulator band theories.
Read the full abstract

Artificial Intelligence

Clearing the Fog: Towards Installing and Refining Proactive Exploration Capabilities in LLM Agents

We study proactive exploration in LLM agents, i.e., the ability to explore an environment to acquire information that improves future decision-making. In this regard, we first identify two fundamental bottlenecks that hinder this capability and then propose \ours, a novel method designed to instill and refine proactive exploration. Specifically, \ours\ consists of two components: (1) Exploratory Data Construction, which synthesizes exploration-rich trajectories to mitigate the hindsight bias of standard demonstrations; and (2) RL Optimization with Contrastive Signal Guidance, which leverages contrastive trajectory pairs to distinguish productive exploration from redundant wandering. Extensive experiments demonstrate the effectiveness of \ours\ and provide insights into the characteristics of proactive exploration. Our code is available at: https://github.com/GuanZhizhao/SAFARI.
Read the full abstract

All authors Zhizhao Guan, Chen Huang, Ziming Liu, Hongru Liang, Wenqiang Lei, See-Kiong Ng, Tat-Seng Chua, Anthony G Cohn

Systems and Control · Cross-listed from eess.SP

Multi-Agent Reinforcement Learning for Joint Handover Management and Power Allocation in Multi-Orbit Satellite Networks

Future sixth-generation non-terrestrial networks are expected to combine low Earth orbit (LEO), medium Earth orbit (MEO), and geostationary Earth orbit (GEO) satellites, whose complementary layers must be coordinated through joint user association, power allocation, and handover management under fast LEO dynamics. This paper studies this problem by formulating it as a mixed-integer nonlinear program and decomposing it into a multi-agent reinforcement learning (MARL) policy that selects the associations and a convex power-allocation subproblem solved exactly at each time slot that defines the reward of the MARL part. The association policy is trained with multi-agent proximal policy optimization (MAPPO) and the targeted multi-agent communication (TarMAC) mechanism, and is made aware of the orbital layer through a state that encodes layer-dependent handover penalties. Evaluated on a realistic multi-constellation scenario built from real two-line element data over Nairobi, Kenya, the proposed policy reaches 92% of the throughput of a greedy signal-to-noise ratio (SNR) maximizing scheme while triggering more than four times fewer handovers, and improves throughput by roughly 14% over a conservative stay heuristic. Compared to an LEO-only learned policy of identical architecture, it attains slightly higher throughput with fewer handovers by offloading a fraction of the users to the MEO and GEO layers, an emergent multi-orbit behavior that drives its favorable throughput and handover trade-off.
Read the full abstract

All authors Yassine Afif, Ashutosh Balakrishnan, Philippe Martins, Mohammed Almekhlafi, Antoine Lesage-Landry, Gunes Karabulut Kurt

Hardware Architecture

Beyond Capacity: Scalable MoE LLM Inference via High-Bandwidth Flash with Direct GPU and HBM Paths

Modern mixture-of-experts (MoE) language models increasingly strain the capacity and cost efficiency of high-bandwidth memory (HBM), as rapidly growing expert weights must be provisioned close to GPUs. High-bandwidth flash (HBF) offers substantially greater capacity, but conventional designs typically deliver HBF-resident expert weights to the GPU through HBM, leaving an additional direct GPU-HBF connection underutilized. We explore an HBF organization that simultaneously exploits two independent expert-delivery routes: a direct path that transfers expert weights from HBF to the GPU and a relay path that transfers them from HBF through the HBM base die to the GPU. Whole experts are assigned to one of the two routes, and transfers over both routes proceed concurrently, increasing aggregate expert-delivery bandwidth without replicating expert weights or introducing a shared relay bottleneck. Early expert determination identifies upcoming experts ahead of their conventional execution point, allowing HBF read latency to overlap with preceding computation, while separate management of immutable expert weights and mutable KV-cache data reduces interference between the two traffic classes. We evaluate the architecture using an event-driven continuous-batching LLM serving simulator with empirically measured GPU compute latencies. Across representative MoE workloads, concurrently utilizing the direct GPU-HBF and HBF-HBM-GPU routes consistently improves expert-delivery efficiency over designs restricted to either route alone. For a representative workload, the proposed architecture can achieve 1.94$\times$ higher throughput and 1.90$\times$ end-to-end speedup over a design that delivers all HBF-resident expert weights to the GPU through the HBM base die.
Read the full abstract

Robotics

CORAL: Curriculum-Optimized Reward Adaptation for LiDAR-Based Goal-Directed Urban Driving

Reinforcement learning is promising for autonomous urban driving, but long-horizon goal-directed navigation asks a policy to acquire several competing behaviors at once–reaching a distant goal, tracking a route, avoiding obstacles, obeying signals–and a fixed objective gives no order in which to learn them. This paper presents CORAL, which advances two schedules together: a five-stage curriculum that progressively lengthens routes and tightens behavioral constraints, and a stage-aware reward whose component weights shift emphasis from mission progress toward route following, safety, smoothness, and rule compliance as the task hardens. The policy is a multi-stream actor-critic network trained with Proximal Policy Optimization (PPO) in CARLA on a compact 99-dimensional state pairing a polar LiDAR histogram with vehicle telemetry, ego-frame route geometry, and traffic-rule indicators–no point-cloud encoder, no bird's-eye-view rasterization. Against two PPO baselines under an identical protocol, CORAL reaches the goal in all twenty evaluation episodes on the longest routes under the full set of behavioral constraints, where the baselines reach 5% and 10%; a factorial ablation shows that neither schedule alone matches their combination: removing either lowers both success and route completion, and disabling both drops success to 55%. Trained in one town, the policy transfers zero-shot to seven unseen towns, succeeding in 68-98% of episodes on routes of the same 100-150 m length, with mean lateral deviation below 0.35 m.
Read the full abstract

Cryptography and Security · Cross-listed from quant-ph

Equivalence Between Average-Case Hardness of Learning and Cryptography for Mixed Quantum States

The relationship between cryptography and learning theory has long been a central theme in the foundations of theoretical computer science: cryptographic primitives can imply hardness of learning, while hardness of learning can in turn be used to construct cryptographic schemes. Recent works have begun exploring analogous connections in the quantum setting, relating the average-case hardness of learning quantum states (AHL) to cryptographic primitives such as one-way state generators (OWSG). Despite recent progress exploring this for pure states, the relationship for mixed states has remained an open question. In this work, we prove that the existence of AHL for mixed quantum states is equivalent to the existence of inefficiently verifiable one-way state generators (IV-OWSGs). As a consequence, this relates mixed-state AHL to EFI pairs. Moreover, as a corollary of existing results, we obtain a separation between IV-OWSGs and OWSGs relative to the SWAP oracle.
Read the full abstract

Artificial Intelligence

Program-space Diffusion for Morphology-to-Transcriptomics Prediction

Spatial transcriptomics (ST) enables genome-wide gene expression profiling while preserving tissue architecture, but its cost and limited scalability remain major bottlenecks. This has motivated models that predict spatial expression directly from routine histology. Despite promising results, most existing approaches operate at the gene level without leveraging established transcriptomic modeling practices and rely on heterogeneous gene selection strategies, which complicates fair comparison across methods. We propose to reformulate morphology-to-transcriptomics prediction as conditional generation in transcriptional program space, thereby exploiting coordinated transcriptional variation instead of predicting genes independently. Using consensus non-negative matrix factorization (cNMF), we extract a low-dimensional set of transcriptional programs capturing coordinated expression variation in the training data, and train a conditional diffusion model to generate program activations from histology. This formulation exploits coordinated transcriptional variation and substantially lowers the dimensionality of the conditional generative task.
Read the full abstract

Cryptography and Security

A Four-Axis Trustworthiness Benchmark for LLM-as-Judge in Principle-Based Regulation

Principle-based regulation, with evaluative standards such as "fair, clear, and not misleading" or "deliver good outcomes", cannot be reduced to binary predicates, and LLM-as-judge is increasingly used as the substitute. Our position is that any such judge must be evaluated on four axes: accuracy, paraphrase robustness, adversarial robustness, and calibration. We release Principle-Bench, 168 cryptoasset financial-promotion scenarios mapped to two UK FCA principles, with paraphrase, adversarial keyword-stuffing, and boundary perturbations authored under a pre-registered rubric; the first benchmark covering all four axes for principle-based regulation. We also introduce Ceca (Calibrated Exemplar-Cluster Assessment): a calibrated, auditable assessor that emits exact per-exemplar counterfactual attributions. Across keyword counting, three sentence-transformer embedders, an open-weight LLM-judge, and a calibrated cascade, no method dominates all four axes. A 120B LLM-judge, strongest on benign inputs, loses 47 accuracy points (0.74 to 0.27) on keyword-stuffed Consumer Duty inputs: "compliance theatre." A second judge from a different model family agrees only at Cohen's kappa = 0.16 on that split, localising the failure to the model rather than the corpus. Any deployment-grade LLM-judge for principle-based regulation must report per-principle adversarial deception and post-hoc calibration alongside aggregate accuracy.
Read the full abstract

Numerical Analysis

Reduced Order Modeling and Applications to Inverse State Estimation

Solving parametric partial differential equations (PDEs) repeatedly for many parameter values,as needed in optimal control, uncertainty quantification, and inverse problems, is often prohibitive with classical discretizations. This chapter introduces Reduced Order Modeling (ROM), which builds compressed yet accurate representations of parametric solution sets for fast online evaluation. We formulate the approximation of parametric PDEs as a supervised learning task and review linear approximation, very effective for elliptic and parabolic problems, before turning to nonlinear methods needed when solutions exhibit discontinuities or steep gradients. We then show how to use reduced-order models to efficiently recover the state of a physical system from limited measurements, which is an inverse problem known as state estimation. Throughout, we favor results with theoretical guarantees while also discussing practical methods, and provide a companion Jupyter notebook implementing the main algorithms.
Read the full abstract

Numerical Analysis · Cross-listed from math.AP

Analytic inverse problems with finitely many random measurements

While infinite-dimensional inverse problems are traditionally analyzed assuming continuous data, practical applications rely on finitely many discrete measurements. Recent deterministic approaches establish that unknowns belonging to a $d$-dimensional model class can be stably recovered from finitely many measurements. However, for severely ill-posed problems, such as the Calderón problem and inverse scattering, the known constructions may require a number of measurements that is exponential in $d$. We show that random sampling reduces this count dramatically if one asks only for exact identifiability. By exploiting the analytic geometry of the forward maps, we prove that, whenever the infinite-data problem is injective on the model class, $2d+1$ random scalar measurements determine the unknown uniquely, almost surely. Applications are given to the Calderón problem, with both infinite- and finite-dimensional boundary sampling, and to inverse medium scattering from randomly sampled far-field values.
Read the full abstract

All authors Giovanni S. Alberti, Damiano Poletti, Simone Sanna, Matteo Santacesaria

Logic in Computer Science

Solving QBF with Counterexample Guided Refinement

We propose two novel approaches for using Counterexample-Guided Abstraction Refinement (CEGAR) in Quantified Boolean Formula (QBF) solvers. The first approach develops a recursive algorithm whose search is driven by CEGAR (rather than by DPLL). The second approach employs CEGAR as an additional learning technique in an existing DPLL-based QBF solver. Experimental evaluation of the implemented prototypes shows that the CEGAR-driven solver outperforms existing solvers on a number of families in the QBF-LIB and that the DPLL solver benefits from the additional type of learning. Thus this article opens two promising avenues in QBF: CEGAR-driven solvers as an alternative to existing approaches and a novel type of learning in DPLL.
Read the full abstract

All authors Mikoláš Janota, William Klieber, Joao Marques-Silva, Edmund Clarke

Computer Vision

TRIAGE: Risk-Controlled Pseudo-Label Admission for Annotation-Efficient Semi-Supervised Retinal OCT Classification

The advanced retinal disease diagnosing imaging modality, optical coherence tomography (OCT), encounters a lack of automation because of the high expenses for annotations performed by specialists. The use of SSL solves the problem of insufficient annotations using unlabeled B-scans; however, most of the current techniques for generating pseudo-labels are based on prediction confidence without considering the asymmetry between different types of errors. This paper proposes TRIAGE, a risk-controlled semi-supervised framework for OCT scans classification, which uses the concept of a patient-level conformal risk controller with an asymmetric cost matrix. TRIAGE unites three crucial modules: a hierarchical classifier that is capable of working with partially abnormal supervision of the disease subtypes, a patient-grouped conformal risk controller with primal-dual coverage control, and a context-aware Transformer teacher for cross-slice verification. On the dataset from Noor Eye Hospital (16,822 B-scans, 161 patients, and 554 volumes) with a test set of unseen patients, TRIAGE demonstrates 89.66% scan-level accuracy, 0.8805 macro-F1, 0.9641 macro-AUC, and an 8.34% under-grading rate when using only 20% of the labeled data. With only 5% of the labeled data, TRIAGE keeps 76.88% accuracy and a 0.1656 under-grading rate. Compared with the other six state-of-the-art semi-supervised methods, TRIAGE significantly outperforms them with ablation study demonstrating the contribution of each module in the overall framework performance (by 42.7% in terms of under-grading rate comparing to fixed threshold methods). TRIAGE demonstrates 98.00% accuracy for 3-class classification with 1% labeled data and 95.94% accuracy for 8-class classification with 10% labeled data on the OCT-C8 dataset.
Read the full abstract

All authors Md Ashraful Hossen Akash, Shyla Afroge, Abdullah Al Mamun, Md. Kishor Morol, Tze Hui Liew

Artificial Intelligence

AnchorBench: A Multi-Pathway Benchmark for the Anchoring Effect in LLMs

The anchoring effect is a cognitive bias in which an initial reference value shifts a later judgment toward itself. This effect is well established in human judgment and decision-making, and recent work suggests that large language models (LLMs) exhibit similar behavior. However, existing work on anchoring in LLMs typically evaluates only a narrow set of anchor pathways and rarely distinguishes irrelevant from plausible anchors. We introduce AnchorBench, a benchmark for the anchoring effect in LLMs that evaluates multiple anchor pathways under an explicit anchor relevance axis. Across fourteen models, including ten open-weight models and four frontier API models, and a large set of controlled prompts, we find that (1) anchoring is strongly pathway-dependent, (2) plausible anchors usually induce larger shifts than irrelevant ones when introduced through stronger pathways, (3) anchor influence generally weakens as the anchor moves farther from the evidence-supported answer, most clearly on External and RAG, and (4) high task accuracy on the anchor-free control condition (Acc$_{10}$: answers within 10 points of gold) does not guarantee robustness: even frontier API models above 95% control accuracy remain susceptible to plausible anchors.
Read the full abstract

All authors Yiderigun Borjigin, Alexander Hermann, Christian Cyron, Roland Aydin

Machine Learning

Quantum Multi-Armed Bandits and Linear Bandits: Lower Bounds and Algorithms

We study quantum multi-armed bandits (QMAB) and quantum linear bandits (QLB), where the learner queries each arm or action through a quantum reward oracle or its inverse. Prior work gives algorithms over horizon $T$ with regret $O(K\log T)$ for QMAB with $K$ arms and $O(d^2\operatorname{polylog} T)$ for $d$-dimensional QLB. This leaves open the optimal dependence on $K$ and $T$ and whether the dependence on $d$ can be further improved. In this work, we prove the first tight minimax regret bound of $\Theta(K\log(1+T/K))$ for QMAB and the first lower bound of $\Omega(d\log(1+T/d))$ for finite-action QLB, ruling out regret independent of $T$. Our lower bounds rely on a high-confidence single-arm quantum testing lower bound for distinguishing a fixed reward mean from an interval of alternatives. A bandit-to-testing reduction then lifts it to the QMAB lower bound, while a linear embedding gives the finite-action QLB lower bound. The matching QMAB upper bound is obtained using a tail bound for the Quantum Monte Carlo (QMC) estimator. For finite-action QLB, we propose a phased elimination algorithm that combines a low-bias low-variance quantum mean estimator with a small-support $G$-optimal design through a query allocation matched to the design weights. When the action set has size $\operatorname{poly}(d)$, its regret is nearly linear in $d$ and matches our lower bound up to polylogarithmic factors.
Read the full abstract

Networking

Robust Constraint-Aware Bayesian Tuning of BBRv2 for QUIC under Tactile Internet Constraints

Tactile Internet applications place strict require- ments on latency, jitter, loss, and responsiveness, which makes transport configuration a critical design factor. Although BBRv2 offers a model-based congestion control framework with strong throughput potential, its default behavior may not be well aligned with delay-sensitive interactive scenarios. This paper presents a robust and constraint-aware tuning framework for BBRv2 in QUIC, where parameter selection is formulated as an expensive black-box optimization problem over multiple emulated network conditions. The tuning process uses Bayesian optimization with the Tree Structured Parzen Estimator to efficiently explore a bounded parameter space under noisy experimental measure- ments. The objective is designed to preserve throughput while enforcing limits on tail latency and loss, while delay instability is evaluated separately through the jitter metric. Experimental results across low, medium, and high impairment scenarios show that the tuned configuration improves tail latency, jitter behavior, and loss performance while maintaining competitive goodput relative to standard QUIC congestion control baselines. These results support robust black-box tuning as a practical method for adapting QUIC transport behavior to tactile Internet style requirements.
Read the full abstract

All authors Muhammad Hanif Lashari, Shakil Ahmed, Wafa Batayneh, Ashfaq Khokhar

Computer Vision

Intelligent Detection of Mechanical, Electrical, and Plumbing (MEP) Metrics Based on 2D Floor Plans

This research developed a neural network-based model to extract various information from 2D floor plans. We detect lighting symbols, identify the appropriate type of light, and extract the associated texts with lights. The study aims to enable efficient floor designing and determining the number and type of lights needed per floor, i.e., allow efficient design and estimate the power requirement of the floor plan. The model was developed using Mask RCNN as the base. The images were annotated and converted into a Coco data format for training the model. The model achieved bbox\_mAP and segm\_mAP values of 0.7596 and 0.7111, respectively. It also performed well at different IoU thresholds, i.e., with bbox\_mAP 50 and segm\_mAP 75 values of 0.9850 and 0.9219, respectively. The developed model will help various industries, such as architecture and construction, to improve design time and create efficient workflows by automatically detecting Mechanical, Electrical, and Plumbing (MEP) objects from floor plans, and it is the first step towards building tools that will help energy-efficient building design.
Read the full abstract

Software Engineering

Breaking Models to Test the Judge: A Mutation Testing Approach for Semantic Evaluators of Domain Class Diagrams

In software engineering, many semantic modeling tasks lack a unique ground truth, as human judgments are both costly and subjective. This paper explores mutation testing as a scalable alternative for evaluating semantic judges (e.g., LLM-based) of models. We propose a mutation testing approach in which controlled semantic defects are injected into domain class diagrams. Starting from pairs of PlantUML class diagrams and textual system descriptions, we apply mutation operators (e.g., removing a class) to generate faulty variants. A candidate judge is then evaluated based on its ability to detect the injected defects. We define 11 mutation operators for the task of comparing a domain class diagram against a textual description and evaluate the proposed approach against a conventional manual assessment of judgment validity. Across six judge configurations (three LLMs and two prompt variants), the automated mutation testing approach is largely consistent with the manual assessment in identifying the better-performing configurations. The results suggest that mutation testing may serve as a scalable proxy for analyzing semantic judges.
Read the full abstract

All authors Kevin Delcourt, Meriem Ben Chaaben, Abdelhamid Rouatbi, Luciano Marchezan, Houari Sahraoui

Computation and Language

Envs-FORGE: Frontier-Optimized Reward-Grounded Environment Synthesis for Agent RL

Reinforcement learning (RL) for terminal agents needs executable training environments with reliable rewards and useful difficulty. Fixed recipes such as few-shot, Self-Instruct, and Evol-Instruct apply the same prompting policy to every seed, even when the current policy would benefit from a harder, easier, or simply different task. We present Envs-FORGE, a prompting policy that converts verifier rewards into per-seed environment-synthesis actions. Envs-FORGE estimates seed pass rates, scores six projection–direction actions around a target learning frontier, and solves a per-seed mixed-integer linear program (MILP) to choose the action that conditions generation. The selected action drives synchronized rewriting of the instruction, fixtures, oracle solution, tests, and Docker environment; only gold-verified bundles enter RL training. The indexed MILP form also supports optional soft skill coverage for portfolio planning. On Qwen 3.5 35B, Envs-FORGE improves Pass@1 over Base by 9.2 percentage points on tb-core (40.0% to 49.2%) and 6.4 points on tb-2.0 (23.0% to 29.4%), exceeding the strongest fixed-recipe baseline by 2.4 and 2.1 points. It reaches 77.1% on SWE-bench Verified versus 73.4% for Base, and improves tb-core by 6.8–9.2 points across the evaluated 4B–35B models. All synthesis methods export 100 verified environments and use 2.27M–2.88M synthesis tokens, placing the comparison at the same downstream training-set size and the same operational scale. The source code is available at https://github.com/DataArcTech/DataArc-SynData-Toolkit/.
Read the full abstract

All authors Xiaojun Wu, Cehao Yang, Honghao Liu, Xueyuan Lin, Zhichao Shi, Hao Zhou, Xuhui Jiang, Chengjin Xu, Jia Li, Jian Guo

Numerical Analysis

Coupling-Aware Vanka Smoothing for Multigrid Preconditioning of the Implicit Immersed Boundary Equations

The immersed boundary (IB) method models fluid–structure interaction using the natural Lagrangian and Eulerian formulations of structural mechanics and fluid dynamics, respectively, but explicit time discretization of the IB force imposes a stiffness-dependent upper bound on the time-step size. Treating these forces implicitly removes this restriction, but the resulting coupled linear systems become increasingly difficult to solve as the structural stiffness increases. Algebraically eliminating the Lagrangian degrees of freedom yields a reduced Eulerian velocity–pressure IB system. Here we introduce a coupling-aware Vanka (CAV) smoothing strategy to enable effective multigrid preconditioning of this system. CAV patches are built as unions of standard pressure-centered Vanka patches, with the graph of the Eulerian elasticity matrix determining which patches are combined. Under grid refinement, CAV patch sizes remain bounded, and the computational cost of each multigrid cycle thereby grows linearly with the number of Eulerian degrees of freedom. Tests using target-point, membrane, and beam force laws show that CAV-preconditioned FGMRES reduces the relative residual by ten orders of magnitude in $9--15$ iterations, with little growth under grid refinement. In a nonlinear benchmark modeling flow past a flexible fiber, the average number of FGMRES iterations per Newton solve increases only from $8.6$ to $9.5$ as the Eulerian grid is refined from $32\times32$ to $256\times256$ cells. To our knowledge, CAV provides the first robust multigrid strategy for time-dependent implicit IB formulations.
Read the full abstract

Computer Vision

Spatial Message Passing in Language Space for Pathology Image Interpretation

Multimodal Large Language Models (MLLMs) can generate pathological descriptions from histological images, but gigapixel Whole Slide Images (WSIs) exceed their visual context limits. The standard tiling workaround makes WSIs tractable yet severs the tissue neighborhoods that define tumor-stroma interfaces and morphology. We introduce Spatial Language Message Passing (SLMP), a framework that performs spatial reasoning entirely in language space, human-readable by construction. SLMP represents a WSI region as a spatial text graph: tiles are nodes initialized with MLLM descriptions, and edges encode spatial adjacency. For each tile, an LLM refines its description by integrating language messages from adjacent tiles under a shared aggregation policy that, on the tile grid, acts as an adaptive local kernel operating on text rather than learned embeddings. This policy is an inspectable prompt that can be refined from model-observed tissue phenotypes via textual gradients, enabling automatic semantic optimization from local cellular context to broader tissue morphology without fine-tuning MLLM weights. On representative HER2 and CAMELYON16 regions, SLMP improves tile-level tumor description accuracy in settings spanning general-purpose and pathology-specialized backbones, with gains of +3.3 to +19.6 percentage points. Random-neighbor ablations confirm that these gains stem from spatial context rather than additional text alone, and inspecting the optimized policies reveals interpretable, tissue-specific decision rules. Besides, without any weight updates or fine-tuning the backbone MLLM, SLMP substantially improves general-purpose MLLMs and narrows its gap to pathology-specialized counterparts, offering a transparent and flexible mechanism for incorporating spatial reasoning into MLLM-based pathology analysis.
Read the full abstract

All authors Jing-Cheng Yang, Hao-Jung Wang, Jinhao Du, Yang Hu, Ming-shan Tsai, Jens Rittscher, Bin Li

Machine Learning · Cross-listed from hep-th

Information Spreading in Diffusion Models from Effective Field Theory

We study score-matching diffusion models with a convolutional architecture. We argue that the inductive bias of locality means that the machinery of effective field theory from physics can be usefully applied to describe the denoising dynamics. We apply this formalism first to a simple toy example which permits an analytical description, and thereafter to MNIST, and show that in both cases, the mutual information between two points grows in a manner predicted by a simple effective field theory of Brownian motion.
Read the full abstract

Artificial Intelligence

Sensor-Driven Mission Synthesis for UAV/UGV Swarms: A TB-CSPN Coordination Architecture with Hardware-Enforced Safety

This paper presents a coordination architecture for heterogeneous UAV/UGV swarms that synthesises mission actions from uncertain, multi-modal sensor evidence while preserving hardware-enforced safety at the actuation boundary. The approach combines radar, RF, acoustic, and visual observations with Topic-Based Communication Space Petri Net (TB-CSPN) orchestration to support incremental mission formation under partial and evolving information. Consultant agents transform sensor outputs into temporally bounded semantic tokens, while supervisor agents provide authorisation and policy-governed release of mission transitions. This separation between interpretation, coordination, and execution yields auditable decision paths, constrains non-determinism within the coordination layer through guards and synchronisation, and enables bounded-time integration of heterogeneous evidence. To improve resilience in contested environments, including cyber compromise, spoofing, jamming, and communication loss, the digital coordination layer is complemented by independent analogue safety envelopes that clamp or veto unsafe actuator commands issued to individual vehicles. A coastal-surveillance case study illustrates how the proposed architecture enables dependable, governed, and physically safe swarm coordination under operational uncertainty.
Read the full abstract

Computational Complexity

One-Sided-Error Parameterized Reductions for the Minimum Distance and Shortest Vector Problems

It is notoriously difficult to obtain deterministic reductions for the Minimum Distance Problem (MDP) and the Shortest Vector Problem (SVP). Under two-sided-error randomized reductions, Bennett, Cheraghchi, Guruswami, and Ribeiro (STOC 2023) proved parameterized hardness of approximation for these problems. We partially derandomize their reductions and present one-sided-error randomized reductions: MDP is W[1]-hard to approximate within an arbitrary constant factor under FPT many-one one-sided-error randomized reductions; For every $p \ge 1$, SVP in the $\ell_p$ norm is W[1]-hard to approximate within an arbitrary constant factor below $2^{1/p}$. We demonstrate the usefulness of one-sided-error randomized reductions by showing that they can be conditionally derandomized when the target problem has an OR function. Under a standard hardness-vs-randomness assumption, namely a plausible lower-bound assumption against nondeterministic circuits, we prove a general theorem formalizing this derandomization. Here, an OR function combines several instances into one instance that preserves their disjunction. We construct such OR functions for the relevant MDP and SVP gap problems, and thereby obtain deterministic W[1]-hardness for approximating MDP over every fixed finite field within every constant factor, and for approximating SVP in $\ell_p$ norms for every fixed integer $p$ within every factor below $2^{1/p}$. Applying the same framework to Micciancio's one-sided-error randomized reduction (ToC 2012) yields, under the same circuit lower-bound assumption, deterministic polynomial-time NP-hardness of approximating Euclidean SVP within every constant factor.
Read the full abstract

Machine Learning

Detecting Contaminated Code-Generation Prompt Batches via Influence Functions

Large language models (LLMs) are increasingly used for code generation, yet they remain vulnerable to prompts that elicit insecure implementations. Existing defenses typically rely on predefined threat models or known vulnerability patterns, limiting their effectiveness against novel attacks. We propose CodeSIFT, a threat-model-agnostic detection method that leverages influence functions to identify batches of prompts that induce anomalous model behavior. Rather than detecting specific vulnerabilities, CodeSIFT measures the parameter-space influence of generated code and uses a statistical test to determine whether a candidate prompt set deviates from a benign reference distribution. To evaluate our approach, we introduce two benchmark datasets covering a variety of vulnerabilities. We evaluate CodeSIFT on three open-weight code LLMs ranging from 3B to 7B parameters, achieving AUROC scores of up to 0.98 at moderate-to-high injection rates, while maintaining well-calibrated false positive rates and substantially outperforming static analysis baselines. These results suggest that influence-function-based detection is a promising direction for identifying malicious code-generation prompts without requiring prior knowledge of the underlying attack class.
Read the full abstract

All authors Francesco Quinzan, Noor Munir, Yishun Lu, Stephen Roberts

Numerical Analysis

Analysis of a Conforming Finite Element Method for Second-Harmonic Generation Scattering

In the frequency domain, nonlinear acoustic and electromagnetic wave propagation can, in certain regimes, be modeled by second-harmonic generation systems consisting of two coupled nonlinear Helmholtz equations. We analyze a conforming finite element approximation of a scalar second-harmonic generation scattering problem truncated using an exact Dirichlet-to-Neumann boundary condition. The discretization uses continuous piecewise polynomial finite elements of degree $p$. Under small-data conditions involving the incident-field and nonlinear susceptibilities, we prove existence and uniqueness of the continuous and discrete nonlinear solutions in a prescribed small ball. In the same regime, we derive quasi-optimal a priori $H^1$-error estimates by a nonlinear Céa-type argument combining linear Galerkin quasi-optimality with small-data stability estimates for the coupled nonlinearities. We also analyze the convergence of the fixed-point iteration used to compute the discrete nonlinear solution and quantify the combined effects of finite element approximation and nonlinear iteration error. A manufactured-solution experiment illustrates the predicted finite element convergence rates, while additional PML-based scattering computations demonstrate the behavior of the nonlinear fixed-point solver in two and three dimensions.
Read the full abstract

Logic in Computer Science · Cross-listed from math.LO

Multimodal Logic Programming with Full Formulas

This paper presents a first-order multimodal logic programming system called MMLP. The system accepts arbitrary formulas as both programs and queries, without restricting either side to Horn clauses or a separate goal grammar. Its declarative semantics is given independently by a Hilbert system for selected D, T, I, B, 4, and 5 modal principles. Execution uses a nested proof calculus with finite grammar certificates for modal propagation. Certificate reachability is equivalent to the associated Horn closure, certificate existence is decidable, and the calculus is cut-free complete. For quantified answer computation, we give a unification algorithm based on permission sets that controls eigenparameter scope. The algorithm always terminates and fails exactly when no admissible solution exists. A successful run returns a unifier that is itself admissible and through which all admissible solutions factor. Computed answers are declaratively correct, and each declaratively correct answer is an ordinary output instance of a computed answer. In proof search, MMLP admits syntactic focalization and a fair and complete enumeration of focused answers. It represents exactly the same answer substitutions as Nguyen's pure logical KDI4s5-MPROLOG on their common fragment, while admitting a strictly larger program and query language.
Read the full abstract

Computer Vision

Conditional Neural Optimal Transport for Predicting Cellular Phenotypes from Molecular Structure

High-content microscopy enables systematic profiling of cellular responses to chemical perturbations, but the scale of the chemical space makes exhaustive phenotypic characterization experimentally infeasible. This motivates computational models that can predict image-derived phenotypes without acquiring the corresponding treated cells. We formulate molecule-induced phenotype prediction as an inductive conditional transport problem in image representation space. Given a negative-control phenotype and the structure of a molecule, we aim to predict the phenotype induced by the corresponding molecule. We first evaluate classical optimal transport baselines and show that static couplings do not yield useful predictions on large-scale phenotypic image datasets. We then introduce a molecule-conditioned Neural Optimal Transport (NOT) model with a Monge-Gap regularization training objective that learns to transport negative-control unperturbed phenotypes toward perturbed phenotypes using molecular structure as conditioning information. NOT recovers molecule-specific phenotypic effects while reducing microscopy-associated technical variation, thereby facilitating comparisons across experimental batches. On unseen active molecules, the model outperforms baseline approaches, demonstrating that chemically conditioned transport can generalize beyond the molecules observed during training. We identified the molecular encoder as the main limitation to this generalization, while transport in a compressed representation space improves performance and scalability. These results establish NOT as a promising framework for predicting cellular phenotypes from molecular structure and negative-control phenotypes, while highlighting the development of more informative molecular representations as a key direction for improving out-of-distribution performance.
Read the full abstract

All authors Gauthier Avité, Maxime Sanchez-Renauld, Nicolas Bourriez, Auguste Genovesio

Human-Computer Interaction

Human and Artificial Intelligence - Promoting Trustworthy and Understandable Collaboration

Methods of Artificial Intelligence (AI) enable the personalization of information for individual user experiences in many domains; however, they can also conflict with established design principles, e.g., due to uncertainties regarding the real world. Building trust and understanding can serve as an approach to create a more balanced relationship between humans and AI. Building upon a pilot study, an online survey was conducted to investigate 12 individual aspects related to the topics of explainability and controllability. The results indicate that both topics, despite their different and numerous facets, are generally perceived as important by respondents; simultaneously, however, a wide dispersion of opinions is frequently observed. This could be an indication that, alongside a fundamental consensus, individual perspectives, technical knowledge and understanding, context-specific factors, or personal experiences play a role in the perception of such systems.
Read the full abstract

Artificial Intelligence

Intern-S2-Mobius: Foundation Model with Decoupled Knowledge and Reasoning

We introduce Mobius-v0, an architecture that comprises a globally shared Memory (FFN) that stores knowledge vectors and multiple Reasoners (Self-Attn) that iteratively achieve compositional reasoning. Using hidden states as cache and carrier, reasoners repeatedly query memory for required knowledge-vectors, while the knowledge is transmitted back to reasoning operators. Through this knowledge-reasoning-separation architecture, Mobius achieves better knowledge compression and reasoning efficiency. Built upon Mobius-v0 architecture: 1) Our 7B model trained-from-scratch achieves similar downstream score as a 7B Transformer baseline with 62.6% of baseline's training data. 2) Our Intern-S2-Mobius, continually-pretrained from Qwen3.5-35B, achieves similar downstream score while delivering nearly 4x end-to-end inference speedup.
Read the full abstract

All authors Kai Chen, Jifeng Ding, Ning Ding, Jiaye Ge, Lixin Gu, Yicheng Gu, Qipeng Guo, Ermo Hua, Haian Huang, Haozheng Hou, Jie Hou, Xiangyu Hong, Che Jiang, Minxi Jin, Cheng Liang, Dahua Lin, Dawei Liu, Kuikun Liu, Chengqi Lv, Haijun Lv, Han Lv, Ningsheng Ma, Biqing Qi, Jianmin Qian, Shiya Su, Youbang Sun, Huanze Tang, Zhongbo Tian, Hanjing Wang, Rui Wang, Ting Wang, Yi Wang, Baiting Wu, Jun Xu, Bowen Yang, Hui Wang, Weida Wang, Haochen Ye, Jiashuo Yu, Shan Yu, Xiaoyi Yu, Qirui Zeng, Qi Zhang, Ming Zhang, Wenwei Zhang, Bowen Zhou, Xinyu Zhou

Machine Learning

Convex losses and their applications to SVM, SVR, and Shallow Neural Networks

We propose multiple new convex losses for SVM and Neural Networks, applied to binary classification tasks. While there are practical limitations in exploiting them with the dual SVM models, we are able to use them with SVM primal formulation and Neural Networks. In detail, the primal SVM problem with the modified losses has been solved with the Particle Swarm Optimization algorithm. We prove that the proposed losses are a generalization of the standard loss, and we experiment them with several small data-sets. This preliminary study shows that using pattern correlations inside the loss function could in theory enhance the generalization performances on some data-sets. To evaluate the performance of each loss, we adopt a Nested Cross-Validation procedure. Results show that generalization measures are the same with or without the new losses.
Read the full abstract