arXiv Research Digest

Billion-parameter Vision-Language-Action (VLA) policies have recently shown impressive performance in robotic manipulation . VLA-AD augments teacher-provided 7-DoF action targets with high-level semantic guidance . At test time, the student policy runs independently with neither the VLA teacher nor the VLM required .

Surrogate Neural Architecture Codesign Package is an open-source AutoML framework for hardware-aware neural architecture codesign and end-to-end FPGA deployment . SNAC-Pack runs a multi-objective global search with Optuna and NSGA-II, loading trials to a shared SQLite store . A hardware surrogate model outputs per-trial resource and latency estimates, avoiding the synthesis cost that would otherwise dominate the search loop .

AgriMind is an ensemble of ResNet50, EfficientNet-B0, and DenseNet121 trained on 20,638 PlantVillage images across 15 pepper, potato, and tomato disease classes . Individual models hit 96--97% on the held-out test set, but averaging their softmax output pushes the ensemble to 99.23% .

Timeseries regression models often struggle to leverage large volumes of labeled multimodal data . This is common in domains like healthcare and predictive maintenance, where data are collected from unreliable sources . Transformer-based large language models have proven effective on structured data such as text .

Memory systems often organize user-agent interactions as retrievable external memory . RecMem stores incoming interactions in a subconscious memory layer and encode them using lightweight embedding models for retrieval . LLMs are only invoked to extract episodic and semantic memory when sustained recurrence are observed for semantically similar interactions .

Reinforcement learning (RL) allows vision-language-action policies to generalize beyond their training distribution by optimizing directly for task success . Gradient cost dominates because much of this computation is spent on phases that contribute little to learning . Probabilistic Chunk Masking (PCM) is a drop-in modification to GRPO that allocates gradient computation to a small, probabilistically selected subset of chunks per trajectory .

CT-AGD (Curvature-Tuned Accelerated Gradient Descent) is a general boosting procedure that accelerates first-order methods by explicitly capturing the local curvature using finite-difference quotients . It has a comparable storage and computational overhead as adaptive gradient methods such as Adam .

Replacement asks whether one layer's map can substitute for another's in place . Interchange asks whether two layers approximately commute when their positions are swapped . On a Pythia training trajectory (410M and 1.4B), replacement-interchange gap grows from initialization to convergence .

FORGE (Failure-Optimized Reflective Graduation and Evolution) is a staged, population-based protocol that evolves prompt-injected natural-language memory for hierarchical ReAct agents . FORGE improves average evaluation return by 1.7-7.7$\times$ over zero-shot and by 29-72% over Reflexion .

Asteria is a runtime system designed to remove this bottleneck by separating second-order optimization logic from the critical GPU training path . Asteria uses training hooks to prepare shadow states in advance, allowing expensive inverse-root computations to proceed asynchronously on the host while GPU computations continue . It uses a bounded-staleness protocol that limits synchronization frequency while preserving optimizer effectiveness through topology-aware coordination .

Large Language Models (LLMs) have demonstrated strong capabilities across diverse NLP applications, such as translation, text generation, and question answering . SGR builds query-specific subgraphs from external knowledge bases and uses their semantic structure to support multi-step inference .

Large language models (LLMs) are prone to prejudiced responses involving race, gender, and age . DebiasRAG leverages self-diagnosed bias contexts relevant to the query through regular retrieval, where the bias contexts are prepared offline . Given the query-specific bias contexts, DebiasAG reversely produces debiasing contexts, which are provided as additional fairness constraints .

Machine-generated texts (MGTs) pose risks such as disinformation and phishing, underscoring the need for reliable detection . Metric-based methods extract statistically distinguishable features of MGTs are often more practical than complex models that are prone to overfitting .

The rise of AI-generated images poses growing challenges for digital authenticity . ReAlign is a novel framework that distills reasoning texts generated by a GRPO-optimized LLM into a lightweight AIGI detector via contrastive learning .

Reconstructing coherent 3D geometry and appearance from unposed multi-view images is a fundamental yet challenging problem in computer vision . We propose IVGT, an Implicit Visual Geometry Transformer that implicitly models continuous and coherent geometry from pose-free images . IVGT learns a continuous neural scene representation in a canonical coordinate system .

The accelerating convergence of smart metering, artificial intelligence, and quantum-inspired combinatorial optimisation is reshaping how energy utilities manage physical infrastructure, customer engagement, and environmental accountability .

Argus is an agentic system in which a Searcher and a Navigator cooperate to treat deep research as assembling a jigsaw from complementary evidence pieces . The Searcher collects evidence traces for a given sub-query through ReAct-style interaction . The Navigator maintains a shared evidence graph, verifying which pieces are still missing, dispatching Searchers to gather them, and reasoning over the completed graph to produce a source-traced final answer . Argus gains 5.5 points

Clinical decision support systems (CDSS) require scrutable, auditable pipelines that enable rigorous, reproducible validation . Most "open" models are open-weight only, withholding data provenance, curation procedures, and generation pipelines that determine model behavior . Fully Open (FO) models, which expose the complete training stack end-to-end, do not currently exist in medicine .

QSurv is a scalable deep learning framework that enables nonparametric continuous-time modeling without relying on time discretization or restrictive distributional assumptions . We propose a training objective based on Gauss-Legendre numerical quadrature, which approximates the cumulative hazard with high-order accuracy . QSurv achieves competitive predictive performance with advantages in instantaneous hazard function estimation .

Effective tutoring requires distinguishing optimal, valid but suboptimal, and incorrect student solutions . We present benchmark of seven LLM feedback agents in propositional logic using knowledge-graph-derived ground truth across 10,836 solution--feedback pairs and three feedback conditions .

CybORG CAGE-2 is a cyber defense environment modeled as a Partially Observable Markov Decision Process . Programmatic state abstraction delivers the largest returns per token spent . Distributing deliberation tools across a hierarchy degrades performance relative to hierarchy alone for all five model families .

A case study for how AI coding systems can be used to generate novel scientific hypotheses . We combine a generic coding agent (Google's AntiGravity) with an LLM-driven tree search algorithm to autonomously generate high-efficiency three-dimensional photovoltaic structures .

Agricultural landscape segmentation in the Global South is challenging as it is characterized by fragmented plots, high intra-class variance, and a scarcity of labeled training data . MAgSeg is an architecturally efficient approach that enables standard MLLMs to perform segmentation of complex smallholder agricultural landscapes from high-resolution satellite imagery, without requiring auxiliary vision decoders .

Autoregressive next-token training offers a unified formulation for image generation and text understanding . But it also creates strong modality competition that destabilizes optimization and limits large-batch scaling . We propose a second-order optimization framework with Multi-Level Variance Correction .

Global AI weather models can produce skillful medium-range forecasts, but their native 0.25 degrees resolution limits direct use for local hazard applications . We introduce SwAIther-Precip, a lead-time-aware downscaling framework that converts coarse-resolution AIFS forecasts into probabilistic km-scale precipitation fields over Switzerland .

Differential privacy changes the effective sample size governing CVaR learning . For tail mass $τ$, the privacy-relevant sample size is not $n$, but $nτ$ For convex Lipschitz learning, modular upper and lower reductions show that CVAR-specific privacy term necessarily scales as $1/2 .

Federated Embedding Distribution Authentication (FedEDAuth) is a lightweight, embedding level client authentication framework that detects and filters malicious participants before model aggregation . FedEDAuth leverages reference embedding distributions derived from a golden dataset and evaluates clients using outlier analysis, mean shift measurements, and micro-cluster behavior .

Federated Learning (FL) has emerged as a promising paradigm for collaborative model training across distributed edge devices while preserving data privacy . Storing this amount of data centrally is challenging due to issues like limited communication, privacy, and regulations . Choosing the right FL architecture depends on the application's needs .

In this paper, we make a case that endogenous tokens such as cryptoassets are not money . We define and classify tokens found on public, permissionless ledgers . We then discuss the work of Kahn et al in Money is Privacy on cash versus simplified credit .

Federated Learning (FL) enables collaborative training across decentralized clients, but most methods assume aligned feature schemas . In these heterogeneous feature spaces, parameter-averaging methods transfer little information across weakly overlapping or disjoint feature groups . We propose a federated imputation framework that separates structural feature unavailability from conventional missingness . This enables indirect cross-client knowledge transfer, even when features are never jointly observed locally .

Federated learning (FL) is vulnerable to data poisoning attacks due to its distributed nature . In this paper, we propose a diffusion-based data poisoning framework against FL systems . We conduct the most systematic and broad experimental evaluation for FL poisoning attacks against various defenses .

Robust aggregation is the core operation in Byzantine-tolerant federated learning . To ensure the quality of aggregation independently of data distribution or attacks, validity conditions are needed . The widespread convex validity requires the output to lie in the convex hull of the honest vectors .

Website Fingerprinting has traditionally focused on inferring which website a user visits from encrypted traffic metadata such as packet sizes and timing . An adversary can infer a user's persona using only packet-length and inter-arrival-time sequences . We formalize persona fingerprinting under both closed-set and open-world settings .

Transformer-based architectures have become the dominant paradigm for Continuous-Time Dynamic Graph (CTDG) learning . However, existing Transformers fail to focus on critical nodes even when they are present in the input, as temporal shift weakens attention contrast and produces overly dispersed attention distributions . This diagnosis suggests a simple fix: replace standard attention with differential attention .

Physics-informed neural networks (PINNs) are powerful surrogates for differential equations but are notoriously difficult to train due to spectral bias, stiffness, and poor accuracy . Adversarial training based on generative adversarial networks (GANs) has recently gained surprisingly strong empirical results in improving training .

Cybersecurity Knowledge Graphs unify diverse Cyber Threat Intelligence sources into structured, queryable formats . CTiKG incorporates hybrid NLP models that leverage SecureBERT+ contextual embeddings and expert knowledge from a domain ontology . Experiments on the DNRTI-AUG-STIX2 dataset demonstrate significant improvements over state-of-the-art baselines .

The system secures three distinct layers: network, host, and hypervisor . Machine learning models at each layer detect known attack patterns, while prediction confidence distinguishes reliable decisions from uncertain outcomes . The proposed system learns adaptive thresholds and reduces LLM escalation by 58.78% .

Distribution utilities now expected to deliver bills that customers can actually read attach a defensible carbon number to every kWh sold and schedule load against grid stress and emissions constraints . We propose an end-to-end framework that unifies four production-grade capabilities under one architectural roof .

Generative artificial intelligence (AI) is increasingly integrated into online platforms where humans exchange opinions . We show biases introduced by AI in human-to-human communication can be amplified through the network and shift collective opinion in their direction . In light of these findings, we investigate whether such biases are controllable by online platforms .

We introduce a dynamics-level approach to watermarking generative models . We embed the watermark directly into the learned continuous dynamics -- the velocity field of a flow matching model . We formulate this as random coding over a continuous channel: a key-dependent perturbation is added during training .

Magnetic order is a fundamental property of materials, governing collective behavior and enabling a broad range of functionalities . Yet magnetic structure remains difficult to determine: experiments are costly and specialized . First-principles methods often struggle with the noncollinear and incommensurate orders found in real materials .

Deep Neural Networks (DNNs) are high-value intellectual property (IP) yet deploying them to edge environments exposes them to unrestricted oracle access . Existing defenses fail to address this practically: passive watermarking only offers post-hoc provenance, while active defenses impose prohibitive latency or require persistent access to sensitive training data .

Aphasias reveal the functional organization of human language by providing causal links between affected brain regions and specific symptom profiles . We introduce an aphasia-inspired technique to characterize emergent functional organizations of language models (LMs) We measure the effects of this intervention against clinical aphasias symptoms as diagnosed by the Text Aphasia Battery (TAB)

Traditional scientific modeling typically begins with fixed, instance-wise effective equations and then carries out equation-specific analysis and computation . We propose an alternative paradigm by learning mesoscopic dynamics within a mathematically constrained hypothesis class .

We examine one dimension of AI governance: how to monitor and audit AI-enabled products and services throughout the AI development lifecycle . We propose techniques that enable developers to perform offline auditing and online (runtime) monitoring of product-specific (temporally extended) behavioral constraints .

Pediatric critical care is a dynamic, high-stakes process involving constant monitoring and adjustments in life-saving treatments . Modeling these interventions is crucial for effective decision support . We frame clinical decision-making as learning to act from trajectories, i.e., imitation learning .

Bayesian Amnesic Piecewise-Robust SAC unifies Bayesian Online Change Detection (BOCD) with robust ensemble RL . The BAPR operator is a combination of mode-conditional Bellman operators weighted by a frozen belief distribution .

Ransomware recovery in critical manufacturing infrastructure is not only a backup-restoration problem, but a critical-infrastructure continuity and interdependency problem . After ransomware, a plant may remain unable to schedule work, authenticate operators or release product, reconnect OT assets, or coordinate suppliers .

Entropic Autoencoders (EAEs) inherently suffer from posterior collapse, a failure mode in which latent variables are effectively ignored . EAEs mitigate posterior collapse by learning non-Gaussian, multimodal latent distributions that yield diverse, data-consistent generations .

We develop a skew-adaptive extension of split conformal prediction for regression . The method starts from an asymmetric interval family centered at a point prediction and uses the gauge approach to deduce the conformity score induced by this family . We also develop a calibration-sample-based estimator for comparing the expected relative future width of skewness intervals .

Contemporary AI increasingly participates in attention allocation, reasoning, synthesis, and decision-making, shaping the very cognitive processes through which humans form beliefs, make decisions, and constitute their sense of self . We identify the risks of unstructured delegation: deskilling, automation bias, transfer of epistemic authority, and oracle-style centralization of knowledge .

AI emotional companions face a safety-rapport paradox: restrictive safeguards can damage supportive alliance, while permissive systems risk user harm . SLIP (Staged Layers of Intervention Protocol) is a four-stage graduated methodology deriving interventions from structured qualitative indicators .

Robots are increasingly present in human spaces, such as for conducting deliveries in hospitals, interacting with visitors at museums, and stocking items in warehouses . To ensure the seamless integration of robots into these spaces, a new role in human-robot interaction is emerging - the robot wrangler .

Demand for AI accelerators is rapidly increasing rack power density, with projections approaching 1MW per deployment by 2027 . Power utilization is particularly important as grid power capacity is a scarce resource in the AI era . Designing an efficient power delivery hierarchy for the long run is difficult .

Generative video models are increasingly used in design animation tasks . Unlike natural video generation, design animation imposes structured constraints . Specific components shall animate with prescribed motion types, directions, speed and timing . Non-animated regions must remain stable and layout structure must be preserved .

Training data attribution (TDA) for music generation must answer two questions that copyright analysis requires . Existing methods reduce influence to a single scalar, without revealing which musical aspects are dominant in that influence . We propose ARIA framework that decomposes attribution along musical aspects (five for symbolic music, three for audio)

Diffusion-based image synthesis has made AI-generated images (AIGI) increasingly photorealistic, raising concerns about authenticity in applications such as misinformation detection, digital forensics, and content moderation . We propose GenShield, a unified autoregressive framework that jointly performs explainable AIGI detection and controllable artifact correction .

Machine learning (ML) methods have been developed to aid with de novo molecule design . Data on the environmental impacts of chemical compounds are sparse, resulting in low-fidelity ML oracles and unreliable candidate proposals . We present GEMS-an interactive visual analytics tool that enables domain experts to directly collaborate with a genetic algorithm for molecule design. Users can integrate their expert knowledge to guide the evolutionary process by modifying the scoring function and molecule population without programming knowledge .

In virtual reality environments, the alignment of perceptual modalities is crucial for immersion and presence . In the AR domain, it is difficult to create such alignments because elements in the physical world are often beyond the user's control . Recent advances in generative AI enable on-demand content creation, enabling highly reactive AR experiences .

LLMs have shown impressive success in program synthesis, discovering programs that surpass prior solutions . Instead of scoring programs after evaluation, we check whether a candidate satisfies a formally defined property . When the property is violated, we stop evaluation early and provide the LLM with a concrete counterexample showing how the program failed . This feedback drastically reduces both the number of program generations and the evaluation cost .

Modeling long-term user interests with massive historical user behaviors enhances click-through rate (CTR) prediction performance in advertising and recommendation systems . GenLI consists of an interest generation module (IGM), a behavior retrieval module (BRM), and an interest fusion module (IFM)

Large Vision-Language Models (LVLMs) have shown significant progress in video understanding, but they face substantial challenges in tasks requiring precise spatiotemporal localization at the instance level . Existing methods primarily rely on text prompts for human-model interaction, but these prompts struggle to provide precise spatial and temporal references . VideoSeeker seamlessly integrates agentic reasoning with instance-level video understanding tasks .

Ada-Diffuser is a causal diffusion model that learns the temporal structure of observed interactions and the underlying latent dynamics simultaneously . The model leverages these dynamics for planning and control tasks . It has a modular design that supports both planning and policy learning tasks .

Recent advances in large language models (LLMs) have significantly enhanced automated legal reasoning . Yet, it remains unclear whether their performance reflects genuine legal reasoning ability or artifacts of data contamination . We show that performance can be inflated by contamination .

Semantic code search has been widely adopted in both academia and industry . These approaches embeds natural-language queries and code snippets into a shared embedding space and retrieve results based on vector similarity . But these approaches often suffer from poor explainability and generalization . We propose XSearch, an intrinsically explainable code search framework .

Learning latent representations from complex data is central to modern machine learning . In such settings, representations are better understood as latent states capturing underlying system dynamics . Yet current approaches remain fragmented, relying on distinct assumptions about what these states should represent . We propose constrained latent state modeling (CLSM) as a unifying perspective .

Current toxic speech datasets are predominantly text-based, limiting the development of models that can capture paralinguistic cues . We present ToxiAlert-Bench, a large-scale audio dataset with over 30,000 audio clips annotated with seven major toxic categories and twenty fine-grained toxic labels .

We report a fixed-base driving-simulator study with a 2x2 manipulation of added latency (100/300 ms) and bitrate (500/2000 kbit/s) We measured effective glass-to-glass (G2G) latency per condition . Physiological measures (heart rate, RR interval, heart rate, skin conductance) exhibited sub-additive interactions, whereas performance and oculomotor interactions were small or non-significant .

Probabilistic forecasting of infectious diseases is crucial for public health but relies on labor-intensive manual model curation by expert modeling teams . Here, we present an autonomous system using Large Language Model (LLM)-guided tree search to generate, evaluate, and optimize executable forecasting software .

Major League Baseball's seven-year experimentation with the Automated Ball-Strike System (ABS) shows how even seemingly straightforward rules require a complex translation process to operationalize via technological systems . ABS is envisioned to call balls and strikes accurately: a seemingly straightforward use of technology to objectively determine the distance between a pitch and the strike zone .

This paper provides an algebraic exposition of the theory of dyadic morality (TDM) We formalize TDM using structural causal modeling (SCM) notation . This algebraic formalization enables neurosymbolic AI systems to compute morality in a way that is both mathematically rigorous and faithful to human moral cognition .

Flow matching and Schrödinger bridges define probability paths, yet their inference grids are usually heuristic or inherited from one-endpoint diffusion . We derive a conditional-marginal entropy-rate objective for bridge-aware discretization . We use it to build a training-free entropic inference-time scheduler from first principles .

The field lacks a scalable way to construct evaluation settings that are realistic, diverse, controllable, inspectable, and reproducible . We introduce ShopGym, an integrated framework for realistic simulation and scalable benchmarking of e-commerce web agents . We validate the framework through graph-based structural analysis and agent-based behavioral evaluation .

This paper develops a sign-separated finite-time error analysis for constant step-size Q-learning . The analysis identifies a max-induced asymmetry in error dynamics . Negative errors admit an optimal-policy lower comparison, the authors say .

Self-supervised pretraining on molecular graphs has emerged as a promising approach for molecular property prediction . Most existing methods operate at a single structural granularity and treat bond information as auxiliary edge attributes rather than as an independent semantic layer . We propose MolCHG, a multi-level self-supervisory pretraining framework built upon a novel Compositional Hierarchical Graph .

Spiking Neural Networks (SNNs) are capable of mimicking the spike-based data processing typical of biological neurons . The spread of digital neuromorphic hardware is slowed down by the prohibitive costs that the silicon tape out of circuits brings . FPGAs could represent a viable alternative, offering a flexible and cost-effective platform .

Neuromorphic computing offers an energy-efficient alternative to conventional machine learning through event-driven computation . The growing adoption of edge computing has created an increasing need for workloads capable of operating under strict resource and energy constraints .

The process of engineering and deploying applications in the edge/embedded space is massively complicated by the non-homogeneous nature of the software stack and the complexity of diagnostics & debugging . This paper presents a work-in-progress vision for a unified language and runtime system that allows applications to scale seamlessly across the edge and cloud . Using a single language and runtimes, applications can be developed and tested in a single environment, and then deployed to any component of the system .

Edge-AI deployment is bottlenecked by data-movement energy . Pair event-driven vision sensors with in-memory analog compute could lift that ceiling by orders of magnitude .

Graph analytics powers smart cities, cyber-physical infrastructure, IoT security, and large-scale social networks . Execution in heterogeneous edge-cloud environments results in higher energy use and carbon emission footprint . MERSEM integrates evolutionary search with reinforcement learning to solve the problem of graph workload allocation and scheduling .

Speculative decoding can significantly accelerate LLM inference, especially given that its cloud-edge collaborative deployment offers cloud workload offloading, offline robustness, and privacy enhancement . PipeSD outperforms state-of-the-art baselines, achieving 1.16x speedup and reducing energy consumption .

Serverless computing has emerged as a promising computing paradigm for edge computing . Efficiently allocating requests to containers is therefore critical to reduce resource over provisioning and unnecessary data movement . This paper proposes Scale, a Service Level Objective aware container scheduling and resource allocation framework .

We study distributed zero-knowledge proofs, introduced by Bick, Kol, and Oshman (SODA 2022) While distributed interactive proofs have advanced rapidly, distributed interactive proof techniques remain limited and mostly problem-specific . Our main contribution is a distributed zero knowledge implementation of Sumcheck . For non-k-colorability, we obtain an $O(n)$-round distributed statistical zero knowledge proof deciding whether a graph is not k-colorable, for any constant k,

One of the central models in distributed computing is Linial's LOCAL model [SIAM J. Comp. 1992]. LCLs are graph problems whose valid solutions can be characterized by a finite set of allowed constant-radius neighborhoods . The last decade has seen major progress in understanding their complexity .

Vision Transformers (ViTs) achieve state-of-the-art performance on challenging vision tasks, but deployment on edge devices is severely hindered by the computational complexity and global reduction bottleneck imposed by layer normalization . Recent methods attempt to bypass this by replacing normalization layers with hardware-friendly scalar approximations .

The constitution assigns, per amendable component including itself, a metric space, aggregation rule, and supermajority threshold . Amendments proceed by members voting with their ideal elements, followed by members submitting public proposals carrying supermajority public support under the revealed votes . Public proposals can be sourced from deliberation among members, vote aggregation, or AI mediation .

Sorting is a fundamental operation across numerous computational domains . The proposed architecture achieves a significant 3.4x reduction in latency compared to memristor-based IMC sorting . To our knowledge, this work represents the first exploration of in-memory sorting using 6T SRAM .

Temporal random walks are widely used to analyze time-stamped interactions in domains such as microservices, finance, and online platforms . Tempest combines a GPU-native dual-index organization over a shared edge store with a hierarchical cooperative scheduler that dispatches walks at thread, warp, or block granularity based on per-step node convergence .

Autonomous multi-agent systems hit critical reasoning, coordination and scaling bottlenecks as the size and complexity of their tasks grow . APWA facilitates parallel execution by decomposing workflows into non-interfering subproblems that can be processed using independent resources without cross-communication .

Mat2Boundary is a DSL and compiler for boundary computations that models a broad class of boundary-conditions as affine sparse linear operators . This abstraction unifies halo copying, circular and symmetric mappings, zero padding, block-edge synchronization, and user-defined interpolation .

Dynamic Management of Resources middleware enables MPI process malleability in Slurm via a simple API decoupled from scheduler internals . We evaluate this design on the MareNostrum~5 supercomputer .

We study the fundamental problem of graph exploration in dynamic graphs using mobile agents . For a graph with $n$ nodes and $k$ agents, we show that exploration is impossible if the adversary can deactivate at least $ \left\lceil \frac{k}{n-2} \right\rceil - 1$ agents per round . We further establish that achieving exploration at this threshold requires agents to have both $1$-hop visibility and global communication .

Machine learning (ML) is increasingly applied to optimize system performance in tasks such as resource management and network simulation . Unlike traditional ML tasks, networked systems often operate in heterogeneous, long-running, and dynamic environment states . EMA takes a system-driven, data-centric approach that accommodates diverse system and model designs .

Much of the world's most valuable information is private, especially in highly regulated sectors such as healthcare and finance, where data include patient histories or customer communications . Unlocking this data could represent a major leap forward, enabling LLMs with deeper domain expertise and stronger real-world utility .

Asynchronous stochastic gradient descent (ASGD) is a standard way to exploit heterogeneous compute resources in distributed learning . We show theoretically that the method is biased toward a frequency-weighted average of the local objectives rather than the desired global objective . We recover the correct objective by rescaling worker-specific stepsizes in proportion to their computation times, so that each worker contributes the same aggregate learning rate over a cycle .

Generative recommendation (GR) has emerged as a promising paradigm that replaces fragmented, scenario-specific architectures with unified Transformer-based models . But deploying GR at scale on Ascend NPUs faces fundamental system-level challenges .

Swarm Network-as-a-Service (SNaaS) leverages fleets of drones to provide on-demand connectivity at scale . SNaaS explicitly models drone-to-device and drone interactions as composable services . A dedicated enforcement module monitors queue stability and SLA latency, adaptively reconfiguring the swarm when violations occur .

This paper presents an LLM-empowered workflow for RISC-V supply chain analysis . The proposed approach addresses the challenges of heterogeneous and unstructured supply chain data by leveraging LLMs for textual understanding and VLMs for extracting information from visual artifacts such as diagrams and tables .

Nature-inspired computing approaches have gained significant attention for solving difficult optimization problems . SACHI is an all-digital Ising architecture implemented by repurposing the L1 cache of a CPU using SRAM-based processing-in-memory techniques .

Charm++ is a parallel programming system which has demonstrated the utility of overdecomposition for many applications and in multiple contexts . The emergence of GPGPUs as a dominant compute component has created some real and perceived challenges for this paradigm .

FocalPolicy is a foresight-aware visuomotor policy that combines Frequency-Optimized Chunking with Locally Anchored flow matching . It combines time-domain alignment within the proximal actions while regularizing frequency-domain structure over multiple future action chunks to improve cross-chunk coherence .

We developed a tool for detecting subtle differences in the probability distributions of datasets . We identify these shifts using an algorithm designed to detect localised density anomalies in high-dimensional feature spaces . If an anomaly is present, we then identify the feature subspace in which the anomaly is most pronounced . This allows us to trace the domain shift to a small set of features, making the shift interpretable .

Industrial Water Treatment Systems exposed to cyber threats that can manipulate process behaviour without creating obvious devices outliers . Logic-layer deception anomalies can preserve numerically plausible measurements while breaking expected cause-and-effect relationships .

Sharpness-aware minimization (SAM) encourages flat minima by perturbing parameters along directions of high loss curvature . But SAM treats all directions uniformly, ignoring the underlying loss geometry . LLQR+SAM combines SAM with a learned preconditioner obtained from the recently proposed LLQr framework .

Multi-Fidelity Flow Matching (MFFM) is a cascade refinement framework for parametric PDE solutions . Source distribution in conditional flow matching is a design parameter that can be calibrated to data, not a default isotropic prior . Conditioning makes residual refinement problem substantially easier than unconditional field generation .

Model-Induced Noise Decoupling (MIND) is a theoretically grounded framework addressing model-induced label noise . We demonstrate that the high-dimensional noise manifold can be decoupled into tractable, subspace-dependent components via Latent Manifold Disentanglement . MIND significantly outperforms state-of-the-art methods on complex benchmarks .

State Space Models (SSMs) are inherently recurrent along the sequence dimension . Yet depth-recurrence - reusing the same block repeatedly across layers - has not been explored in this model family . We show that a looped SSM with $k$ parameters iterated $L$ times consistently closely matches or outperforms a standard SSM . We further show that input reshaping is an equally neglected design axis .

LLM-as-a-judge has become the dominant paradigm for grading model outputs at scale . Yet the same model assigns systematically different scores when its output format changes . Using Position-aware Edge attribution Patching, we causally investigate the internal mechanism in Gemma-3, Qwen2.5, and Llama-3.5 .

Using tabular Q-learning to train an olfactory search agent with a minimal memory of past observations . This agent learns an interpretable strategy to recover the plume which combines well-known behaviors observed in insects: surging, casting, and a return downwind .

Shapley Neuron Valuation (SNV) quantifies Neuron importance in continual learning . SNV selectively freezes important Neurons while keeping others plastic, enabling buffer-free continual learning without expanding architecture. SNV improves accuracy by +2.88% in the class incremental learning and +6.46% in task incremental learning scenarios .

Stipple patterns, point sets whose local density tracks a target image, are traditionally produced by per-density iterative optimizers, which are slow, non-differentiable . We present the first diffusion-based sampler that simultaneously satisfies a learned local point-distribution prior and a continuous, image-defined capacity constraint at inference .

Continual learning (CL) aims to train models sequentially over multiple domains without forgetting previously learned knowledge . Existing CL methods optimize for in-domain performance and are prone to learning spurious, domain-specific cues (``shortcut learning'')

Graph-Optimized Multimodal Alignment (GOMA) is a structure-driven post-alignment framework . GOMA decouples three key design choices: where messages should flow, how multimodal evidence should propagate, and which smoothing depth should be retained . It learns modality-aware propagation operators, performs finite-step coupled smoothing without diagonal cross-modal shortcuts .

We propose a scalable neuromorphic architecture based on spiking dynamics emerging from the autonomous time-continuous evolution of clockless (asynchronous) digital circuits . Implemented on commercially available field-programmable gate arrays, our system implements networks of interacting Boolean spiking neurons with configurable excitatory and inhibitory synaptic weights .

Neural network surrogate models have emerged as a promising approach to model solution fields for boundary value problems encountered in physical modeling . Stochastic problems represent an area of interest because of the potential to significantly reduce the repeated evaluation of expensive forward models via traditional numerical solvers when conducting parametric analysis .

We propose a variational quantum classifier operating on high dimensional deep representations via amplitude encoding, stabilized by a learnable classical pre encoding layer . Model reliability is assessed using SAFE-AI metrics derived from the Cramer von Mises divergence .

Machine learning systems increasingly make life-changing decisions about individuals . How can individuals respond to negative decisions made by these opaque systems? We propose an operational definition of contestability as a complement to recourse .

Model-Based Reinforcement Learning (MBRL) leverages latent imagination for sample efficiency, yet remains constrained by Historical Tethering . We propose Mind Dreamer (MD), a framework that operationalizes Active Latent Intervention (ALI) to transcend Markovian continuity . MD reformulates discovery as the minimization of a global Relay Manifold Expected Free Energy (R-EFE)

Monte Carlo methods are essential across diverse scientific fields, yet their efficiency is frequently hampered by slowing down . In this paper, we demonstrate that incorporating physical priors into the model significantly enhances performance . Building upon existing strategies that integrate spin-spin interactions, we propose a framework that utilizes a prior probability distribution as a starting point for training .

Deep Reinforcement Learning (DRL) for quadrotor flight control typically relies on Domain Randomization for sim-to-real transfer . We propose a novel adaptive control architecture that actively perceives and reacts to instantaneous perturbations . Real-world validations on a Crazyflie micro-quadrotor demonstrate that our adaptive controller significantly outperforms baselines .

We consider the problem of testing properties of graphs underlying high-dimensional graphical models . We adopt the model of covariance queries introduced by Lugosi, Truszkowski, Velona, and Zwiernik (2021) We study the case when the underlying graph is a tree .

Variational Autoencoders (VAEs) often exhibit a polarised regime in which latent variables separate into active, passive, and mixed subsets . Existing criteria for identifying active dimensions depend on a Gaussian prior, limiting their applicability to variational models and specific priors . We propose a simple information-theoretic classification based on the entropy of the mean representation . We show theoretically how this entropy couples to KL minimisation through entropy--variance bounds .

We propose a novel definition of model exploitation in reinforcement learning . Informally, a world model is exploitable if it implies that one policy should be strictly preferred over another while the environment's true transition model implies the reverse . We analogize our definition with a prior characterization of reward hacking .

Connectivity-Constrained and Retrieval-Enhanced (CCRE) deep learning framework proposes a deep-learning framework . CCRE achieves a 72.3% first-destination accuracy and a 61.4% average three-step accuracy, outperforming baselines like CatBoost and LSTM .

Parameter-efficient fine-tuning (PEFT) has emerged as a critical technique for adapting large-scale foundation models across natural language processing and computer vision . We introduce a novel PEFT method that constructs orthogonal transformations through low-rank skew-symmetric matrices and compositional rotation chains . We propose an approximation scheme that enables fully parallel computation of compositional rotations .