arXiv Research Digest

Multi-aspect Knowledge Distillation (MaKD) mimics self-attention and feed-forward modules in greater depth to capture rich language knowledge information at different aspects . MaKD can achieve competitive performance compared with strong baselines with the same storage parameter budget .

Video Anomaly Detection (VAD) is a fundamental challenge in computer vision . We argue that the bottleneck in VAD is not necessarily model capacity, but rather the static nature of inquiry . We propose QVAD, a question-centric agentic framework that treats VLM-LLM interaction as a dynamic dialogue .

Prompt compression has established itself as a cost-effective and low-latency method for accelerating inference in large language models . We present the first systematic, large-scale study of this trade-off . We show effective compression can reduce memory usage enough to offload workloads from data center .

A human validated multi judge LLM evaluation further reveals calibration bias in single judge pipelines . Logit level KD provides the most reliable gains, while complex distillation improves semantic similarity for short summaries but degrades longer outputs .

Graph domain adaptation (GDA) aims to transfer knowledge from a labeled source graph to an unlabeled target graph under distribution shifts . Existing methods are largely feature-centric and overlook structural discrepancies . Dual-Aligned Structural Basis Distillation (DSBD) for GDA is a novel framework that explicitly models and adapts cross-domain structural variation . DSBD constructs a differentiable structural basis by synthesizing continuous probabilistic prototype graphs, enabling gradient-based optimization over

For multimodal large language models, visual information is relatively sparse compared with text . MI-Pruner is simple, efficient and non-intrusive, requiring no access to internal attention maps or architectural modifications .

Traditional methods struggle to effectively separate objects from intricate backgrounds and fail to fully leverage the rich multi-scale information contained within images . We have developed a synergistic feature fusion network (SFFNet) with dual-domain edge enhancement specifically tailored for object detection in UAV images .

JoyAI-LLM Flash is an efficient Mixture-of-Experts (MoE) language model designed to redefine the trade-off between strong performance and token efficiency in the sub-50B parameter regime . The model comprises 48B total parameters while activating only 2.7B parameters per forward pass, achieving a substantially higher sparsity ratio than contemporary industry leading models .

Autoregressive (AR) video diffusion models enable long-form video generation but remain expensive due to repeated multi-step denoising . Existing training-free acceleration methods rely on binary cache-or-recompute decisions, overlooking intermediate cases where direct reuse is too coarse yet full recomputation is unnecessary . SCOPE introduces a tri-modal scheduler over cache, predict, and recompute .

This paper reviews the NTIRE 2026 challenge on efficient single-image super-resolution . The aim of this challenge is to devise a network that reduces one or several aspects, such as runtime, parameters, and FLOPs . The challenge had 95 registered participants, and 15 teams made valid submissions .

Scaling vision-language-Action models by upgrading the vision encoder is expected to improve downstream manipulation performance . We show that this expectation fails when actions are represented as discrete tokens . In any visuomotor pipeline, scaling behavior is governed by the location of the tightest information bottleneck .

SkillRT is a compilation and runtime system designed for portable and efficient skill execution . SkillRT achieves up to 3.2x speedup with enhanced parallelism, and 19-50x latency reduction through code solidification .

Most of the recent generative image super-resolution (SR) methods rely on adapting large text-to-image (T2I) diffusion models . While effective, this paradigm starts from a generic T2I generator . We propose VOSR, a Vision-Only generative framework for SR .

Large language models (LLMs) encode vast world knowledge in their parameters, yet they remain limited by static knowledge, finite context windows, and weakly structured causal reasoning . This survey provides a unified account of augmentation strategies along a single axis: the degree of structured context supplied at inference time .

EffiMiniVLM is a compact dual-encoder vision-language regression framework that integrates an EfficientNet-B0 image encoder and a MiniLM-based text encoder with a lightweight regression head . The proposed model contains 27.7M parameters and requires 6.8 GFLOPs, yet achieves a CES score of 0.40 .

Distilling video generation models to extremely low inference budgets (e.g., 2--4 NFEs) is crucial for real-time deployment . Trajectory-style consistency distillation often becomes conservative under complex video dynamics, yielding an over-smoothed appearance and weak motion . We propose Self-Consistent Distribution Matching Distillation (SC-DMD) which explicitly regularizes the endpoint-consistent composition of consecutive denoising updates .

Video Large Language Models (Video-LLMs) remain prone to spatiotemporal hallucinations . STEAR identifies high-risk decoding steps and selects token-conditioned visual evidence from grounding-sensitive middle layers . It uses this shared evidence for two coupled purposes: restoring missing local grounding in middle layers, and constructing temporal counterfactuals to falsify inconsistent reasoning during late-layer decoding .

Large Reasoning Models (LRMs) have recently achieved remarkable success in complex reasoning tasks . However, closer scrutiny reveals persistent failure modes compromising performance and cost . NeuReasoner is an explainable, controllable, and unified reasoning framework driven by MoN .

HyperCT is a framework that dynamically adapts a Vision Transformer backbone via a Hypernetwork . Validated on a large-scale dataset of radiological and cardiological tasks, it outperforms various strong baselines .

Model predictive control (MPC) with learned world models has emerged as a promising paradigm for embodied control . We demonstrate that this hierarchical approach enables zero-shot control on real-world non-greedy robotic tasks, achieving a 70% success rate on pick-&-place using only a final goal specification .

Learned Transfer MIA (LT-MIA) trains a membership inference classifier exclusively on transformer-based models . It transfers zero-shot to Mamba (state-space), RWKV-4 (linear attention), and RecurrentGemma (gated recurrence) all exceed performance on held-out transformers .

MV-VDP is a multi-view video diffusion policy that jointly models the 3D spatio-temporal state of the environment . The core idea is to simultaneously predict multiple-view heatmap videos and RGB videos . It also aligns the representation format of video pretraining with action finetuning, and specify what actions the robot should take .

PRISM is a structured topic modeling framework combining rich representations captured by LLMs with the low cost and interpretability of latent semantic clustering methods . PRISM fine-tunes a sentence encoding model using a sparse set of LLM- provided labels on samples drawn from some corpus of interest . We segment this embedding space with thresholded clustering, yielding clusters that separate closely related topics within some narrow domain .

Evaluating the factuality of long-form output generated by large language models remains challenging . We propose a comprehensive factuality evaluation framework that jointly measures precision and recall . Our method leverages external knowledge sources to construct reference facts and determine whether they are captured in generated text .

Few-Shot Medical Image Segmentation (FSMIS) aims to segment novel object classes in medical images using only minimal annotated examples . Diffusion Models (DM) excel in visual tasks but their potential for FSMIS remains largely unexplored . We propose that the rich visual priors learned by large-scale DMs offer a powerful foundation for a more robust and data-efficient segmentation approach .

This paper explores the use of server learning for enhancing the robustness of federated learning against malicious attacks . We propose a heuristic algorithm that uses server learning and client update filtering in combination with geometric median aggregation .

This work establishes a rigorous bridge between infinite-dimensional delay dynamics and finite-dimensional Koopman learning, with explicit and interpretable error guarantees .

Agent Skills is an emerging open standard that defines a modular, filesystem-based packaging format enabling LLM-based agents to acquire domain-specific expertise on demand . Despite rapid adoption across multiple agentic platforms, the security properties of Agent Skills have not been studied .

Pay-for-use Wi-Fi hotspots are emerging as a scalable solution to provide affordable Internet access in underserved and rural regions . Despite their growing adoption, their security properties remain largely unexplored .

Federated learning (FL) enables collaborative training across organizations without sharing raw data . FedSQ freezes a structural copy of the pretrained model to induce fixed binary gating masks during fine-tuning . Fixing the gating reduces learning to within-regime affine refinements, which stabilizes aggregation under heterogeneous partitions .

This paper presents the first machine learning (ML) analysis of NASA Open Science Data Repository (OSDR) dataset OSD-970, derived from the Rodent Research-1 (RR-1) mission . Using RT-qPCR data from 89 adipogenesis and thermogenesis pathway genes in gonadal WAT of 16 female C57BL/6J mice .

Accurate crowd simulation is crucial for public safety management, emergency evacuation planning, and intelligent transportation systems . The proposed STDDN method has demonstrated significantly superior simulation performance compared to state-of-the-art methods on long-term tasks .

Non-Exemplar Continual Graph Learning (NECGL) seeks to eliminate the privacy risks intrinsic to rehearsal-based paradigms . We propose Analytic Drift Resister (ADR), a novel and theoretically grounded NECGL framework . ADR exploits iterative backpropagation to break free from the frozen pre-trained constraint, adapting to evolving task graph distributions and fortifying model plasticity .

Environment-injected Trajectory-based Agent Memory Poisoning (eTAMP) is the first attack to achieve cross-session, cross-site compromise without requiring direct memory access . eTAMP achieves substantial attack success rates: up to 32.5% on GPT-5.2, 23.4% .

The Internet of Medical Things (IoMT) is transforming healthcare by enabling seamless connectivity among medical devices, systems, and services . This paper proposes a novel Tsetlin Machine (TM)-based Intrusion Detection System . The TM is a rule-based and interpretable machine learning (ML) approach that models attack patterns using propositional logic .

PR3DICTR: Platform for Research in 3D Image Classification and sTandardised tRaining . Built using community-standard distributions (PyTorch and MONAI) Built using modular design principles and standardization .

Cardiovascular modeling has rapidly advanced over the past few decades due to the rising needs for health tracking and early detection of cardiovascular diseases . Certain hemodynamic parameters like the terminal resistance/compliance, are difficult to clinically estimate and often yield non-physiological hemodynamics when sampled naively . We present a systematic framework for training machine learning (ML) models, capable of instantaneous hemodynamic prediction and parameter estimation .

Transformer attention computes a single softmax-weighted average over values -- a one-pass estimate that cannot correct its own errors . We show that a single Hopfield-style update erases all query information orthogonal to the stored-pattern subspace . Further iteration under local contraction can collapse distinct queries in the same region to the same fixed point .

Reflective Context Learning (RCL) is a unified framework for agents that learn through repeated interaction, reflection on behavior and failure modes, and iterative updates to context . In RCL, reflection converts trajectories and current context into a directional update signal analogous to gradients .

Recent success of reinforcement learning (RL) in large reasoning models has inspired the growing adoption of RL for post-training Multimodal Large Language Models to enhance their visual reasoning capabilities . Hallucination-as-Cue Framework introduces hallucination-inductive, modality-specific corruptions that remove or replace essential information required to derive correct answers .

Proton magnetic resonance spectroscopic imaging ($^1$H MRSI) enables mapping of whole-brain metabolites concentrations in-vivo . Metabolite maps of human subjects acquired at 3T and 7T with isotropic resolutions of 10 mm, 3.4 mm and 2 mm were quantified with HyperFitS and compared to conventional LCModel fitting . Results show a substantial agreement between the new and gold-standard methods, with significantly faster fitting times by HyperFit

We study high-dimensional convex empirical risk minimization (ERM) under general non-Gaussian data designs . By extending Convex Gaussian Min-Max Theorem, we derive an asymptotic min-max characterization of key statistics . This result clarifies the scope and limits of Gaussian universality for ERMs .

Tool-augmented AI agents substantially extend the practical capabilities of large language models, but they also introduce security risks that cannot be identified through model-only evaluation . We present a systematic security assessment of six representative OpenClaw-series agent frameworks .

On-policy distillation (OPD) has become a popular training paradigm in the LLM community . This paper demonstrates that learning signals solely derived from the privileged teacher result in severe information leakage and unstable long-term training . We identify the optimal niche for self-distillation and propose RLSD (RLSD)

Kimi K2.5 is an open-weight LLM that rivals closed models across coding, multimodal, and agentic benchmarks, but was released without an accompanying safety evaluation . We evaluate the model for CBRNE misuse risk, cybersecurity risk, misalignment, political censorship, bias, and harmlessness, in both agentic and non-agentic settings .

Cyber-attacks continue to grow in scale and sophistication, yet existing network intrusion detection approaches lack the semantic depth required for path reasoning over attacker-victim interactions . We address this by first modelling network alerts as a knowledge graph, then formulating hyper-relational alert prediction .

Large language model (LLM) agents learn by interacting with environments . Long-horizon training remains bottlenecked by sparse and delayed rewards . We propose Self-Guide, a self-generated internal reward for language agents that supports both inference-time guidance and training-time supervision .

LLM-based coding agents extend their capabilities via third-party agent skills distributed through open marketplaces without mandatory security review . Unlike traditional packages, these skills are executed as operational directives with system-level privileges .

Third-party skills extend LLM agents with powerful capabilities but often handle sensitive credentials in privileged environments . We present the first large-scale empirical study of this problem, analyzing 17,022 skills (sampled from 170,226 on SkillsMP) We identify 520 vulnerable skills with 1,708 issues and derive a taxonomy of 10 leakage patterns .

HL7 FHIR allows concurrent access to a set of shared patient resources, i.e., EHR systems, pharmacy systems, lab systems, and devices . The FRAG model is implemented as a three-pass graph traversal detection algorithm and tested against a time window-based baseline on 1,500 synthetic transaction logs .

MECO includes 42 participants and provides approximately 38 hours of multimodal signals, yielding 30,592 synchronized samples . The modalities cover video, audio, electroencephalography (EEG), and electrocardiography (ECG) In addition, the dataset offers comprehensive annotations of emotional and cognitive states .

Transferability analysis finds transferability rates vary depending on the task . Some models, in particular on deepfake detection, have different transferability behavior . We call these models `flat-earther' models .

We present a method to identify a valence-arousal (VA) subspace within large language model representations . Projections along our recovered VA subspace correlate with human-crowdsourced VA ratings across 44k lexical items . Steering along these directions induces near-monotonic bidirectional control over refusal and sycophancy .

AI systems increasingly take on instructional roles - providing feedback, guiding practice, evaluating work . Does it matter to learners who they believe is on the other side? We investigated this using a three-condition experiment (N=148) in which participants completed a creative coding tutorial and received feedback generated by the same large language model attributed to either an AI system or a human teaching assistant .

Multi-modal recommendation (MMR) enriches item representations by introducing item content, e.g., visual and textual descriptions . Success hinges on aligning these content modalities with user preferences derived from interaction data .

Partial deepfake speech detection requires identifying manipulated regions that may occur within short temporal portions of otherwise bona fide utterances . We propose a split-and-conquer framework that decomposes the problem into two stages: boundary detection and segment-level classification . This formulation simplifies the learning objective by separating temporal localization from authenticity assessment .

This paper reports a survey of generative AI use among 83 MBA thesis students in Japan . 95.2% reported at least some use and 77.1% heavy use . Students engaged AI across the full research-writing workflow - literature review, drafting, and consultation when stuck .

Chart-RL is a novel reinforcement learning framework that enhances VLMs chart understanding through feedback-driven policy optimization of visual perception and logical inference . The RL fine-tuned Qwen3-VL-4B-Instruct model achieved an answer accuracy of 0.634 .

We study visual explanation in geometry education as a Referring Image Segmentation problem . We present a fully automated procedural data engine that generates over 200,000 synthetic geometry diagrams with pixel-perfect segmentation masks and linguistically diverse referring expressions . We propose domain-specific fine-tuning of vision-language models (VLMs)

DuoChart combines synthesized charts with real-world charts to construct diverse, high-quality chart training data . CharTool-7B outperforms the base model by **+8.0%** on CharXiv (Reasoning) and **+9.78% on ChartQAPro .

Squirrel ecology offers a sharp comparative case because arboreal locomotion, scatter-hoarding, and audience-sensitive caching couple all three demands in one organism . We introduce a minimal hierarchical partially observed control model with latent dynamics and structured episodic memory .

Industrial software development across chip design, GPU optimization, and embedded systems lacks expert reasoning traces showing how engineers reason about hardware constraints and timing semantics . We propose InCoder-32B-Thinking, trained on data from the Error-driven Chain-of-Thought (ECoT) synthesis framework with an industrial code world model (ICWM) to generate reasoning traces .

Multi-speaker scenarios remain challenging due to overlapping speech, backchannels, rapid turn-taking, and context window constraints . We propose Speaker-Reasoner, an end-to-end Speech LLM with agentic multi-turn temporal reasoning .

Problem solving plays an essential role in science education . Generative AI (GAI) chatbots have emerged as a promising tool for supporting students' science problem solving . However, general-purpose chatbots (e.g., ChatGPT) often provide direct, ready-made answers, may lead to cognitive offloading .

R2-Write is an automated framework that synthesizes high-quality thinking trajectories enriched with explicit reflection and revision patterns . To prevent redundant reflections, we design a process reward mechanism that supervises reflection quality during reinforcement learning .

Learning methods using synthetic data have attracted attention as an effective approach for increasing the diversity of training data while reducing collection costs . Many existing methods improve robustness only indirectly through diversification of training samples and do not explicitly teach the model which regions in the input space truly contribute to discrimination .

Council Mode is a novel multi-agent consensus framework . It dispatches queries to multiple heterogeneous frontier LLMs in parallel and synthesizes their outputs through a dedicated consensus model . Council pipeline operates in three phases: (1) an intelligent triage classifier that routes queries based on complexity, (2) parallel expert generation across architecturally diverse models, and (3) structured consensus synthesis .

We present SentiAvatar, a framework for building expressive interactive 3D digital humans, and use it to create SuSu, a virtual character that speaks, gestures, and emotes in real time . The source code, model, and dataset are available at https://sentiavatar.io .

The ocean interior regulates Earth's climate but remains sparsely observed due to limited in situ measurements . We present a depth-aware generative framework for reconstructing high-resolution ocean states from extremely sparse surface data . The framework accurately reconstructs subsurface temperature, salinity, and velocity fields across multiple depths .

Longitudinal health agents must reason across multi-source trajectories that combine continuous device streams, sparse clinical exams, and episodic life events . We present ESL-Bench, an event-driven synthesis framework and benchmark providing 100 synthetic users . Users paired with 100 evaluation queries across five dimensions - Lookup, Trend, Comparison, Anomaly, Explanation - stratified into Easy, Medium, and Hard tiers .

NavCrafter explores 3D scenes from a single image by synthesizing novel-view video sequences with camera controllability and temporal-spatial consistency . The framework leverages video diffusion models to capture rich 3D priors and adopts a geometry-aware expansion strategy to progressively extend scene coverage . We further propose a collision-aware camera trajectory planner and enhanced 3D Gaussian Splatting pipeline .

ChatSVA is an end-to-end SVA generation system built upon a multi-agent framework . The AgentBridge platform enables this approach by systematically generating high-purity datasets, overcoming the data scarcity inherent to few-shot scenarios . The online service has been publicly released at an online service .

Generalized reciprocity -- the tendency to help others after receiving help oneself -- is widely theorized as a mechanism sustaining cooperation on online knowledge-sharing platforms . Yet robust empirical evidence from field settings remains surprisingly scarce . Using Cox proportional hazards models on over 21 million questions, we find that receiving an answer significantly increases a user's propensity to help other users . This effect is concentrated among newcomers and declines with platform experience .

We present a domain-adapted RAG pipeline for tutoring move annotation . We adapt retrieval by fine-tuning a lightweight embedding model on tutoring corpora and indexing dialogues at the utterance level to retrieve labeled demonstrations . Retrieval corrects systematic label biases present in zero-shot prompting .

Synthetic Aperture Radar (SAR) is the operational standard because of its all-weather capability, but it remains challenging to distinguish morphologically similar ice classes under severe class imbalance . This paper establishes a trustworthy SAR only baseline that future fusion work can build upon .

Multimodal large language models (MLLMs) enable powerful cross-modal reasoning capabilities but impose substantial computational and latency burdens . We propose MSAO, an adaptive modality sparsity-aware offloading framework with edge-cloud collaboration for efficient MLLM Inference .

Advancements in extended reality (XR) are driving the development of the metaverse, which demands efficient real-time transformation of 2D scenes into 3D objects . We propose a digital twin (DT)-based in-network computing (INC)-assisted multi-access edge computing (MEC) framework . In this framework, a network operator manages wireless and computational resources for XR user devices . XUDs autonomously offload tasks to maximize their utilities .

This paper investigates three approaches for spacecraft telemetry anomaly detection -- forecasting & threshold, direct classification, and image classification -- and optimizes them for edge deployment using multi-objective neural architecture optimization . Analysis of deployment viability shows our optimized models require just 0.36-6.25% of CubeSat RAM .

An on-device epsilon-net filter retains only semantically novel frames, building a denoised embedding index . A single-pass streaming filter outperforms offline alternatives (k-means, farthest-point, uniform, random) across eight vision-language models (8M-632M) on two egocentric datasets (AEA, EPIC-KITCHENS) Combined, architecture reaches 45.6% Hit@5 on held-out

The fast pace of modern AI is rapidly transforming traditional industrial systems into vast, intelligent and potentially unmanned autonomous operational environments driven by AI-based solutions . These solutions leverage various forms of machine learning, reinforcement learning, and generative AI . These smart capabilities have pushed the envelope in multiple industrial domains, enabling predictive maintenance, optimized performance, and streamlined workflows .

UET-RVMCU is a lightweight RISC-V microcontroller derived from the UETRV-PCore . The project demonstrates the feasibility of transforming an application-class SoC into a feature-rich microcontroller suitable for embedded systems .

EEspice is an open-source circuit simulation framework whose modular architecture decouples device model evaluation into independently replaceable kernels . It partitions MOSFET instances into independent color groups, which can be processed in parallel .

Softmax can become a computational bottleneck in the Transformer model's Multi-Head Attention (MHA) block . Head-Calibrated Clipped-Linear Softmax (HCCS) is a bounded, monotone surrogate to the exponential softmax function . HCCS uses a clipped linear mapping of the max centered attention logits .

Efficient utilization of GPU resources and power has become critical with the growing demand for GPUs in high-performance computing . VASP is a widely used materials science application on Perlmutter at NERSC, an HPE Cray EX system .

Multi-FPGA systems (MFS) are widely adopted for VLSI emulation and rapid prototyping . In an MFS, FPGAs connect only to a limited number of neighbors through bandwidth-constrained links . RePart reduces total hop distance by 52.3% on average over state-of-the-art hypergraph partitioners .

Resource sharing in multi-tenant cloud environments enables cost efficiency but introduces the Noisy Neighbor problem . Despite extensive research on detecting such effects, there are no explainable methodologies for quantifying the severity of impact and establishing causal relationships among tenants .

Reasoning about consistency models for replicated data systems is a challenging task that requires a deep understanding of both the consistency models themselves and a large part of human inputs in mechanized verification approaches . We introduce HistMSO, a monadic second-order logic (MSO) for histories and abstract executions .

We develop a sketch-based factor reduction and a Nesterov-accelerated projected gradient algorithm (NPGA) with GPU acceleration . The method combines randomized subspace embedding, spectral truncation, and ridge stabilization to construct an effective factor $L_{eff}$. It then solves the resulting constrained problem with a structured projection computed by scalar dual search and matrix-vector kernels .

ScaleGNN is a 4D parallel framework for scalable mini-batch GNN training . It combines communication-free distributed sampling, 3D parallel matrix multiplication (PMM), and data parallelism . On Perlmutter, scaleGNN achieves 3.5x end-to-end training speedup over the SOTA baseline on ogbn-products .

Large language models (LLMs) have grown beyond the memory capacity of single GPU devices . While NF4 (4-bit NormalFloat) quantization enables 4$\times$ memory reduction, inference on current NVIDIA GPUs (e.g., Ampere A100) requires expensive dequantization back to FP16 format .

Cloud computing allows scalable resource provisioning, but dynamic workload changes often lead to higher costs due to over-provisioning . Machine learning (ML) approaches are effective for predicting workload patterns, but they can introduce delays during sudden traffic spikes . In contrast, mathematical heuristics like Game Theory provide fast and reliable scheduling decisions, but do not account for future workload changes .

Cloud infrastructure supports efficient operation of data pipelines regarding requirements like cost, speed, and resource utilization . Study contributes theory of optimization goals like minimizing cost, reducing execution time, and cost-makespan trade-offs .

Parameter-efficient fine-tuning (PEFT) has emerged as a crucial paradigm for adapting large language models under constrained computational budgets . But standard PEFT methods often struggle in multi-task fine-tuneing settings, where diverse optimization objectives induce task interference and limited parameter budgets lead to representational deficiency . We propose FourierMoE, which integrates the MoE architecture with the inverse discrete Fourier transform (IDFT) for frequency-aware adaptation .

Large-scale distributed training has been a research hot spot in machine learning systems for industry and academia in recent years . But conducting experiments without physical machines and corresponding resources is difficult . One solution is to leverage distributed training simulators, but current ones do not support importing real-world developed models .

Neutral atom quantum computers (NAQCs) are among the most promising computational platforms for quantum computing . Controlling and measuring individual atoms and their states is typically the most time-consuming task during computation . To resolve this challenge, we propose a highly-parallel atom-detection accelerator for tweezer-based NAQCs .

We obtain better algorithms for computing more balanced orientations and degree splits in LOCAL . Important to our result is a connection to the hypergraph sinkless orientation problem [BMNSU, SODA'25]

The classical-quantum system heterogeneity renders existing distributed communication mechanisms (e.g. MPI, NCCL etc.) inadequate . This bottleneck severely impairs operational synergy and programming efficiency . To address these challenges, this paper proposes a message-passing library tailored for large-scale quantum heterogeneous distributed computing .

When MPI-parallel simulations run on shared Kubernetes clusters, conventional CPU scheduling leaves the vast majority of provisioned cycles idle . This paper presents a multiplexing framework that reclaims this idle capacity by co-locating multiple simulations .

Modern GPU clusters are built upon a complex hierarchy of heterogeneous interconnects, ranging from multi-rail RDMA to proprietary fabrics such as Multi-Node NVLink and Ascend UB . Orchestrating these diverse links effectively remains a critical challenge in disaggregated LLM serving . We present TENT, a data-movement engine that decouples transfer intent from physical execution . In LLM inference with SGLang HiCache, TENT achieves up to 1.36

The Production and Distributed Analysis (PanDA) system was originally developed for the ATLAS experiment at the CERN Large Hadron Collider . PanDA supports AI/ML-driven workflows through a scalable and flexible workflow engine . We present an AI-assisted framework for detector design optimization .

Modern deep learning training often occurs at the Edge of Stability (EoS) where the sharpness of the loss exceeds the threshold below which classical convergence analysis applies . In this work, we introduce and study a structural property of loss functions that we term product-stability . We show that for losses with product-stable minima, gradient descent applied to objectives of the form $(x,y) \mapsto l(xy)$ can provably converge to the local minimum even

Reinforcement Learning (RL) has been shown to substantially improve the reasoning capability of small and large language models . We propose an RL framework that uses rewards from an LLM that acts as a judge evaluating model outputs over large amounts of unlabeled data .

Routing algorithms must be able to react to traffic bursts within milliseconds . LogGIA is a scalable graph neural routing algorithm that predicts log-space link weights from attributed topology-and-telemetry graphs . It uses data-driven pre-training stage followed by on-policy Reinforcement Learning .

Class-incremental learning (CIL) is typically evaluated under predefined schedules with equal-sized tasks . We propose a model-agnostic framework for robust CIL learning under free-flow arrivals . We constrain distillation to replayed data, normalize the scale of contrastive and knowledge transfer losses .

Diffusion models often degrade when trained in latent spaces (e.g., VAEs) yet formal causes remain poorly understood . We quantify latent-space diffusability through the rate of change of the Minimum Mean Squared Error (MMSE) along the diffusion trajectory . We demonstrate that while global isometry ensures FI alignment, FIR is governed by the encoder's local geometric properties .

AutoVerifier is an LLM-based agentic framework that automates end-to-end verification of technical claims without requiring domain expertise . It decomposes every technical assertion into structured claim triples of the form (Subject, Predicate, Object) and builds knowledge graphs that enable structured reasoning across six progressively enriching layers .

Graph neural networks (GNNs) are a well-regarded tool for learned control of networked dynamical systems due to their ability to be deployed in a distributed manner . Current distributed GNNs assume that all nodes in the network collect geometric observations in compatible bases . This paper presents a GNN parametrization that is globally invariant to choice of local basis .

Function vectors (FVs) can steer large language model behavior when added to the residual stream . We hypothesized that FV steering failures reflect an absence of task-relevant information: logit lens would fail alongside steering . We found that FVs succeed even when the logit lenses cannot decode the correct answer at any layer . This steerability-without-decodability pattern is universal .

We address nonconvex learning problems over undirected networks . We provide theoretical privacy guarantees within a differential privacy framework . We show the algorithm's superior performance on a classification task under the same privacy budget, compared with state-of-the-art methods .

In this work we evaluate the performance of three classes of methods for detecting financial anomalies . We apply these methods to the TSX-60 data to identify major financial stress events in the Canadian stock market . We show how neural network-based methods achieve the strongest performance .

We generalize the framework of Fredholm Neural Networks to learn non-expansive integral operators arising in Fredholm Integral Equations (FIEs) of the second kind in arbitrary dimensions . We also demonstrate how FREDINOs can be used to learn the solution operator of non-linear elliptic PDEs .

Given $n$ independent samples from a $d$-dimensional probability distribution, our aim is to generate diffusion-based samples . We define a plug-in estimator and show that it is minimax-optimal . We develop Wasserstein bounds between the distribution of the model and the true distribution .

We introduce a nonnegative functional on the space of line arrangements in $\mathbb{P}^2$ that vanishes precisely on free arrangements . The functional is obtained as a semicontinuous relaxation of Saito's criterion for freeness . Using this functional as a reward signal, we develop a sequential construction procedure in which lines are added one at a time so as to minimize the angular distance .

Reward models (RMs) used in reinforcement learning from human feedback are vulnerable to reward hacking . We propose SignCertified Policy Optimization (SignCert-PO) down-weighting non-robust completions in the policy gradient update . We make the assumption that reward hacking is often caused by flipped advantage signs .

Paper proposes inversion-free stochastic natural gradient method for probability distributions whose parameters lie on a Riemannian manifold . The method maintains an online approximation of the inverse FIM, which is efficiently updated at quadratic cost using score vectors sampled at successive iterates .

Exponential Weights (EW) algorithm with isotropic Gaussian prior for online logistic regression . We show that EW can be both computationally tractable and geometrically adaptive in online classification .

ExSTraQt (EXtract Suspicious TRAnsactions from Quasi-Temporal graph representation) is an advanced supervised learning approach to detect money laundering (or suspicious) transactions in financial datasets . We claim that our framework could seamlessly complement existing AML detection systems in banks .

Data assimilation is the process of estimating the time-evolving state of a dynamical system by integrating model predictions and noisy observations . Classical filters often struggle with accuracy or computational feasibility in high dimensions .

Feature maps associated with positive definite kernels play a central role in kernel methods and learning theory . They are closely related to robustness and stability guarantees . In this paper, we study the Lipschitz regularity of feature maps under differentiability assumptions . We also study continuous and shift-invariant kernels such as Gaussian, Laplace, and Matérn kernels .

AI-based surrogate models have shown strong potential in several areas of computational physics for accelerating otherwise expensive high-fidelity simulations . On the studied case, the learned surrogate produces 6-hour predictions in about $0.4\,\mathrm{s}$ on a single NVIDIA A100 GPU .

Transfer learning (TL) offers a promising avenue to mitigate this challenge by exploiting information from related but richer source domains . FT-MDN-Transformer is a mixture-density tabular Transformer architecture specifically designed for TL in RR forecasting .

Growing number of individual generating units, hybrid resources, and security constraints has significantly increased the computational burden of network-constrained unit commitment . This paper proposes a solver-compatible dimensionality reduction framework for UC that exploits structural regularities in commitment decisions . The framework identifies a sparse subset of structurally stable commitment binaries to fix prior to optimization .

Active Preference Learning (APL) seeks to optimize query efficiency in online Direct Preference Optimization . Modern LLMs inherit strong priors from web-scale pretraining, which can limit headroom of post-training data-selection strategies . APL fails to mitigate this capability collapse or reduce variance significantly better than random sampling .

This paper focuses on the state estimation problem in distributed sensor networks . Unlike existing AKF that handle missing data and measurement outliers, the proposed VB-AKF adopts a dual-mask generative model with two independent Bernoulli random variables .

MOMO uses model merge to integrate representations learned independently from three key Martian sensors (HiRISE, CTX, and THEMIS) Central to our method is our novel Equal Validation Loss (EVL) strategy .