arXiv Research Digest

Large Vision Language Models (LVLMs) demonstrate impressive capabilities, but their substantial computational and memory requirements pose deployment challenges on resource-constrained edge devices . We study a complementary route: compressing existing LVLMs by applying structured pruning to the language model backbone, followed by lightweight recovery training . We assess the feasibility of conducting recovery training with only a small fraction of the available data .

Long-context reasoning is a critical capability of large language models, enabling applications such as long-document understanding, summarization, and code generation . However, efficient autoregressive inference relies on the key-value (KV) cache, whose memory footprint grows linearly with sequence length . To mitigate this overhead, KV cache pruning methods discard cached tokens with low attention scores during inference .

Flow-based vision-language-action (VLA) policies offer strong expressivity for action generation, but suffer from a fundamental inefficiency . Multi-step inference is required to recover action structure from uninformative Gaussian noise . We propose CF-VLA, a coarse-to-fine two-stage formulation that restructures action generation into a coarse initialization step that constructs an action-aware starting point .

Designing deep networks that meet latency and accuracy constraints on edge accelerators increasingly relies on hardware-aware optimization . We evaluate the proposed framework on vessel segmentation for spaceborne maritime monitoring, targeting the Intel Movidius Myriad X Visual Processing Unit .

On-device Small Language Models promise fully offline, private AI experiences for mobile users . But is this promise achievable in practice? This paper presents a case study documenting the engineering challenges of integrating SLMs into a production Android word-guessing game .

We consider online learning problems under a partial observability model capturing situations where the information conveyed to the learner is between full information and bandit feedback . For this setting, we propose the first algorithm that enjoys near-optimal regret guarantees without having to know the observation system before selecting its actions . We also define a new partial information setting that models online combinatorial optimization problems .

GradMAP trains independent neural-network policies for each agent without any parameter sharing . Each agent uses only its own local observation for online decision-making without communication . GradMAP embeds a differentiable three-phase AC power-flow model in a primal-dual learning loop .

Functional Task Networks (FTN) uses a high dimensional, self-organizing binary mask over a large population of small but deep networks . The mask is produced by a three-stage procedure: (1) gradient descent on a continuous mask identifies task-relevant neurons, (2) a smoothing kernel biases the result toward spatial contiguity, and k-winner-take-all binarizes the resulting group at a fixed capacity budget .

Single-image point cloud reconstruction must infer complete 3D geometry, including occluded parts, from a single RGB image . While diffusion-based reconstructors achieve high accuracy, they typically require many denoising iterations, resulting in slow and expensive inference . We propose Point-MF, a Mean-Flow-based framework for low-NFE single image reconstruction . It operates directly in point-cloud space to learn the mean velocity field and enables one-step reconstruction .

DenSNet is a density-first approach to machine-learned electronic structure . It learns the Hohenberg--Kohn map from nuclear configurations to the ground-state electron density . The approach uses an SE(3)-equivariant neural network to predict density coefficients of a flexible atom-centered Gaussian basis .

Chart-to-code generation converts a chart image into an executable plotting script . The same chart can be expressed by semantically equivalent scripts in different plotting languages . CharLuMA augments the multimodal projector with a language-conditioned mixture of low-rank subspaces .

Layerwise Convergence Fingerprinting (LCF) is a tuning-free runtime monitor that treats the inter-layer hidden-state trajectory as a health signal . LCF aggregates via Ledoit-Wolf shrinkage and thresholds via leave-one-out calibration on 200 clean examples, with no reference model, trigger knowledge, or retraining. LCF reduces mean backdoor attack success rate (ASR) below 1% on Qwen2.5-7B and Gemma

Despite advances in text-to-image generation, models still struggle to accurately render prompt-specified text with correct spatial layout . This challenge is driven not only by the lack of datasets that align prompts with the exact text and layout expected in the image . To address these issues, we introduce TextGround4M, a large-scale dataset of over 4 million prompt-image pairs .

SPLIT is a novel method for simulating image-based tactile sensors . Central to our approach is a latent space arithmetic strategy that explicitly disentangles contact geometry from sensor-specific optical properties . Unlike methods that require recalibration for every new unit, this disentanglement allows SPLIT to adapt to diverse DIGIT backgrounds .

Long-context ability has become one of the most important iteration direction of next-generation Large Language Models . The standard softmax attention exhibits quadratic time complexity with respect to sequence length . As the sequence length increases, this incurs substantial overhead in long-context settings, leading the training and inference costs of extremely long sequences deteriorate rapidly .

DYMAPIA is a multi-domain Deepfake detection framework that fuses spatial, spectral, and temporal cues to capture subtle traces of manipulation in visual data . System builds dynamic anomaly masks by combining evidence from Fourier spectra, local texture descriptors, edge irregularities, and optical flow consistency .

Representation learning is fundamental to NLP, but building embeddings that work well at different computational budgets is challenging . We propose MIPIC (Matryoshka Representation Learning via Self-Distilled Intra-Relational Alignment and Progressive Information Chaining)

Scientific publication compresses a branching, iterative research process into a linear narrative, discarding the majority of what was discovered along the way . We introduce the Agent-Native Research Artifact (Ara) protocol that replaces the narrative paper with machine-executable research package . Ara raises question-answering accuracy from 72.4% to 93.7% on PaperBench and RE-bench .

Graph-based Retrieval-Augmented Generation (GraphRAG) extends traditional RAG by using knowledge graphs to give large language models a structured, semantically coherent context . GraphRAG reasoning process remains a black-box, limiting our ability to understand how specific pieces of structured knowledge influence the final output .

Large Language Models (LLMs) have recently been explored as fine-grained zero-shot re-rankers by leveraging attention signals to estimate document relevance . RouteHead is a query-dependent head selection method for attention-based re-ranking with LLMs .

Vision-language models transfer well in zero-shot settings but at deployment the visual and textual branches often shift asymmetrically . Under this condition, entropy-based test-time adaptation can sharpen the fused posterior while increasing error . We propose MG-MTTA, which keeps the backbone frozen and updates only a lightweight gate or adapter .

Recent advancements in reinforcement learning with verifiable rewards have significantly improved the complex reasoning ability of vision-language models . However, outcome-level supervision is too coarse to diagnose and correct errors within the reasoning chain . To this end, we propose Perceval, a process reward model (PRM) that enables token-level error grounding .

Multi-modal Evidence Grounding (MEG) is a metric that quantifies the contribution of retrieved evidence . Unlike standard confidence measures, MEG utilizes Semantic Certainty Anchoring, focusing on high-IDF information-bearing tokens that better capture the semantic core of the answer . MEG-RAG consistently outperforms strong baselines and demonstrates robust generalization across different teacher models .

Quantum error mitigation (QEM) is essential for extracting reliable results from near-term quantum devices . Practical deployments must balance mitigation strength against runtime overhead under time-varying noise . GSC-QEMit is a telemetry-driven framework for mitigation that switches between lightweight suppression and heavier intervention as drift evolves .

In many areas of medicine, security, and life sciences, we want to allocate limited resources to different sources in order to detect extreme values . In this paper, we study an efficient way to allocate resources sequentially under limited feedback . The most commonly optimized property is the regret with respect to the maximum mean reward .

The decentralization of modern energy systems is transforming consumers into prosumers who continuously exchange data with aggregators, peers, and market operators . While such data is essential for peer-to-peer trading, demand response, and distributed forecasting, it can reveal sensitive household patterns and introduce privacy risks .

The system is composed of three agents: a Verifier (who wants to check properties of a dataset), a Prover (who owns the dataset) and an Auditor . The Verifier is allowed to ask a limited number of simple binary true/false questions to the auditor . The Auditor uses the Model Context Protocol (MCP) to dynamically inspect the target dataset .

Fine-tuning unlocks large language models for specialized applications, but its high computational cost often puts it out of reach for resource-constrained organizations . Data privacy concerns make sharing sensitive information with third parties risky . A promising solution to this problem is split learning, which divides the model between clients and a server .

A t-private n-server Information-Theoretic Distributed Point Function allows one to convert any point function f_{alpha,beta}(x): [N] -> G into n shares (secret keys) This paper constructs a novel share conversion based on the private information retrieval (PIR) of Ghasemi, Kopparty and Sudan .

Light-activated drugs are a promising way to treat localized diseases for which existing treatments have severe side effects . Here we used computational techniques to find a set of promising candidates for the photoactive inhibition of the poly(ADP-ribose) polymerase 1 (PARP1) cancer target .

SAGE (Sparse Adaptive Guidance) is a novel LLM-based generation framework that enforces sparse and dynamic dependency guidance . SAGE discretizes features into value-aware pseudo-features and constructs a mutual information-based sparse dependency graph . This graph adaptively guides generation through explicit context selection or implicit logit correction .

Graph neural ordinary differential equations extend graph learning from discrete message-passing layers to continuous-time representation flows . While it supports adaptive long-range propagation, we show that Graph ODEs with strictly positive irreducible mixing operators face an inherent monostability trap . We propose HGODE (HGODE) which couples feature evolution with a latent topological potential driven by a learned pairwise force .

Trusted Execution Environments (TEEs) on low-power microcontrollers (e.g., ARM TrustZone-M) enable isolation of Secure and Non-Secure software but still require both worlds to share resources, including interrupt controllers . Many RTOS-s rely on periodic interrupts (SysTicks) to advance their own notion of time (time-keeping), but the delivery of this interrupt is essential for preserving real-time behavior . On the other hand, the

The deployment of intelligent reinforcement learning (RL) agents on resource-constrained edge devices remains a challenge . We introduce BitRL, a framework for building RL agents using 1-bit quantized language models . BitRL achieves 10-16x memory reduction and 3-5x energy efficiency improvements over full-precision baselines .

As LLM-driven agents advance in cybersecurity, Jeopardy CTF benchmarks are approaching saturation . Cyber ranges, the natural next evaluation frontier, offer diminishing resistance under current design . To counteract this trend, we propose Dynamic Cyber Ranges .

CipherXRay is inspired by avalanche effect, but it only checks whether a "ripple effect" of avalanche effect exists, allowing a straightforward counterattack to succeed . In this work, we present a new approach that checks the avalanche effect itself .

Autonomous AI agents extend large language models into full runtime systems that load skills, ingest external content, maintain memory, plan multi-step actions, and invoke privileged tools . This paper presents AgentWard, a lifecycle-oriented, defense-in-depth architecture that organizes protection across these five stages .

Current cyber attribution approaches typically operate on a per-incident basis . We investigate whether cross-campaign attribution reduces ambiguity or whether structural limits persist under longitudinal data . We model adversary fingerprints as multi-dimensional feature vectors encoding behavioral, infrastructural, and temporal characteristics . We introduce ARCANE (Attacker Re-identification via Cross-campaign Attribution Network) framework .

Meta-CoT is a paradigm that performs a two-level decomposition of any single-image editing operation with two key properties: (1) Decomposability. (2) Generalizability. We observe that any editing intention can be represented as a triplet - (task, target, required understanding ability). We find that training on these five meta-tasks, together with the other two elements of the triplet, is sufficient to achieve strong generalization across unseen

The widespread adoption of social media has heightened interest in its psychological effects . This study seeks to address this by segmenting individuals according to their social media usage and psychological well-being . Data from 551 participants, collected via an online survey, were preprocessed using KNN imputation for missing values .

Backdoor attacks on detection transformers for OD tasks rely on patch-wise triggers optimized at fixed locations with minimal perturbations . We propose DETOUR, a practical backdoor attack by using semantic triggers that are effective in real-world object detection systems .

Existing fraud detection mechanisms typically rely on machine learning methods that treat each financial asset (i.e., token) and its related transactions independently . However, market manipulation strategies are rarely isolated events, but are characterized by coordination, repetition, and frequent transfers among related assets . This suggests that relational structure constitutes an integral component of the signal .

Recent work in hierarchical reinforcement learning has shown success in scaling to billions of timesteps when learning over a set of predefined reward functions . We show that, instead of using a single reward function per option, the reward functions can be effectively used to induce a space of behaviours . We call this method Hierarchical Behaviour Spaces (HBS)

Progression to dialysis or end-stage renal disease is a rare but clinically important outcome . Clinicians need evidence on how medication exposures influence downstream risk . We constructed a fixed-window EHR cohort (90-day observation, 730-day prediction; N=81401) and modeled sequences of diagnoses, procedures, and medications with kidney laboratory trends .

We study the problem of global maximization of a function f given a finite number of evaluations perturbed by noise . We consider a very weak assumption on the function, namely that it is locally smooth (in some precise sense) with respect to some semi-metric, around one of its global maxima . StoSOO follows an optimistic strategy to iteratively construct upper confidence bounds over the hierarchical partitions of the function domain to decide which point to sample next .

In multi-objective reinforcement learning (MORL) one widely studied approach addresses this by training a single policy network conditioned on preference-weighted rewards . We propose using the RFRL's training objective as an auxiliary task to enhance MORL .

Previous work largely focuses on settings with a known binary state space, where the state is either 0 or 1 . We study prior-agnostic robust forecast aggregation in which the aggregator observes only experts' reports, yet is ignorant of the underlying joint information structure and the full prior, including the underlying state space .

SceneSelect uses unsupervised clustering on interpretable geometric and kinematic features to discover a latent scene taxonomy . A highly decoupled classification module is trained to assign real-time inputs to these taxonomy categories . A plug-and-play scheduling policy automatically dispatches the trajectory sequence to the optimal expert predictor .

Most foundation models in biology are trained within one modality or for a fixed forward task . We present MIMIC, a generative multimodal foundation model trained on our newly curated and aligned dataset, LORE . We link nucleic acid, protein, evolutionary, structural, regulatory, and semantic/contextual modalities within partially observed biomolecular states .

Graph-based anomaly detection methods show promise in protecting networks, but field lacks a standardized, reproducible environment to train these models and evaluate their efficacy . Gammaf is not a novel defense mechanism itself, but rather a comprehensive evaluation architecture designed to generate synthetic multi-agent interaction datasets and benchmark the performance of existing and future defense models .

The need to burn 100% H2 in high efficient gas turbines featuring low NOx combustion in premix mode require the complete redesign of the combustion system to ensure stable operation without any flashback . Generative design methods using latest AI technology will provide promising potential .

Citizen science is transforming how cognitive scientists study the human mind . By embedding experimental tasks into engaging, game-like experiences, researchers can reach large, diverse populations while collecting rich behavioral data outside the lab .

This review proposes an integrative framework grounded on interoception and embodied AI-termed the interoceptive machine framework- that translates biologically inspired principles of internal-state regulation into computational architectures for adaptive autonomy . Interoception is conceived as the monitoring, integration, and regulation of internal signals .

PersonaFlow generates editable user personas from OSS repository artifacts and integrates them alongside issue reports . In a user study with 13 OSS developers, most reported shifts in how they understood users . More than half modified their responses by adding empathetic language .

Large Audio-Language Models show consistent performance gains across speech and audio benchmarks . High scores may not reflect true auditory perception . If a model can answer questions without processing the acoustic signal, the benchmark fails as a measure of auditory understanding .

Automatic chord recognition (ACR) extracts time-aligned chord labels from music audio recordings . ACR still struggles with oversegmentation, data scarcity, and imbalance, especially in recognizing complex chords such as non-triads . We reformulate ACR as a segment-level sequence-to-sequence prediction task .

An attention-based deep learning model, inspired by transformer architecture, is used for short-term solar power forecasting . Unlike traditional models, SolarTformer leverages self-attention mechanisms to capture temporal and spatial variability in solar irradiance .

Frontier image generation has moved from artistic synthesis to synthetic visual evidence . This paper provides a source-grounded technical and policy analysis of synthetic visual risk . We argue for layered control: model-side restrictions, cryptographic provenance, visible labeling, platform friction, sector-grade verification, and incident response .

BITRec is a novel generative multi-behavior recommendation framework that introduces structured behavioral modeling through selective dependency activation . BITRec incorporates Hierarchical Behavior Aggregation (HBA), which explicitly models behavioral intensity differences through separated exploration and commitment pathways . Experiments on four large-scale datasets (RetailRocket, Taobao, Tmall, Insurance Dataset) achieve consistent improvements of 15-23% across multiple metrics .

Electroencephalography (EEG) is highly susceptible to artifact contamination, such as electrooculographic (EOG) interference, which severely degrades signal quality and hinders reliable interpretation in applications including neurological diagnosis, brain-computer interfaces (BCIs) and neurological diagnosis . We propose BandRouteNet, an adaptive frequency-aware neural network for EEG denoising .

RefEvo is a dynamic multi-agent framework designed for agile and reliable reference modeling . It has three key innovations: A Dynamic Design Planner that autonomously decomposes design specifications and constructs tailored execution workflows based on semantic complexity . A Co-Evolutionary Verification Mechanism employs a Dialectical Arbiter to simultaneously rectify the model and verification logic against the specification .

The Observational Medical Outcomes Partnership Common Data Model (OMOP CDM) enabled harmonisation of electronic health records data of nearly one billion patients in 83 countries . But generating real-world evidence (RWE) from these repositories remains a manual process requiring clinical, epidemiological and technical expertise .

Automated code review (ACR) bots are increasingly used in industrial software development . As adoption grows, a key challenge is how to evaluate the usefulness of bot-generated comments reliably and at scale . We examine the feasibility and limitations of automating the evaluation of LLM-powered ACR bots in an industrial setting .

A key limitation is suboptimal design of visualizations for small-screen devices . We argue that this gap is partly driven by a lack of specialized developer tools . We advocate for dedicated mobile visualization libraries tailored to personal health data and mobile contexts .

Diffusion language models generate without a fixed left-to-right order, making token ordering a central algorithmic choice . Existing systems mainly use random masking or confidence-driven ordering . We introduce DPRM (Doob h-transform Process Reward Model), a plug-in token-ordering module for diffusion language models .

Process Reward Models (PRMs) have achieved remarkable success in augmenting the reasoning capabilities of Large Language Models (LLMs) within static domains such as mathematics . However, their potential in dynamic data analysis tasks remains underexplored . We introduce DataPRM, a novel environment-aware generative process reward model that can serve as an active verifier .

The development of practical (multimodal) large language model assistants for Korean weather forecasters is hindered by the absence of a multidimensional, expert-level evaluation framework grounded in authoritative sources . To address this, we introduce K-MetBench, a diagnostic benchmark grounded in national qualification exams .

We evaluate the propensity of frontier models to sabotage or refuse to assist with safety research when deployed as AI research agents within a frontier AI company . We find no instances of unprompted sabotage across any model, with refusal rates close to zero for Mythos Preview and Opus 4.7 Preview, though all models sometimes only partially completed tasks .

ULLER (Unified Language for LEarning and Reasoning) offers a unified first-order logic (FOL) syntax . We show that these seemingly disparate semantics are all instances of one categorical framework based on monads . This enables the modular addition of new semantics and systematic translations .

Emerging extended reality technologies are reshaping how children play, learn, and socialize . Recent news has highlighted safety concerns such as car accidents, lower judgment for real-world situations, and exposure to disturbing content like virtual rape . This research examines how XR game design may lead to online safety risks for children .

Driving in compliance with traffic laws and regulations is a basic requirement for human drivers . But autonomous vehicles can violate these requirements in diverse real-world scenarios . Conventional approaches use formal logic languages to specify behavioral constraints . We propose a novel pipeline that grounds LLM reasoning in a traffic scenario taxonomy through node-wise anchors .

As LLM agents transition to autonomous digital coworkers, maintaining deterministic goal-directedness in non-linear multi-turn conversations emerged as an architectural bottleneck . We propose Self-Synthesizing Reasoning Protocols (SSRP), a metacognitive framework that implements a discrete separation between high-level architectural planning (Architect) and turn-by-turn procedural execution .

In projectors and head-mounted displays, an out-of-focus image appears blurred . Even when a display itself is in focus, computer operation may be hindered if the display is far from the user or if a user has poor visual acuity, because the user cannot see the screen clearly .

Multiple myeloma is managed through sequential lines of therapy over years to decades, with each decision depending on cumulative disease history distributed across dozens to hundreds of heterogeneous clinical documents . Whether LLM-based systems can synthesise this evidence at a level approaching expert agreement has not been established .

The Perceived Cooperativity Scale (PCS) and Teaming Perception Scale (TPS) are theoretically grounded scales . The PCS captures an agent's perceived cooperative capability and practice within a single interaction sequence . The TPS captures the emergent sense of teaming arising from mutual contribution and support . Both scales were adapted for human-human cooperation to enable cross-agent comparisons .

Emerging IoT-enabled cyber-physical applications demand low-latency, energy-efficient, and reliable execution across resource-constrained edge devices . Existing scheduling approaches fail to jointly address all these objectives and constraints . The proposed formulation jointly optimizes latency, energy, and reliability, while holistically addressing timing and resource constraints .

A system can satisfy accuracy-based validation, maintain output stability (Safety-Threshold Exceedance Rate, STER, equal to zero) and still violate timing constraints under deployment load . These are structurally independent properties that current pre-market validation protocols often do not operationalize at the inference layer . Joint STER and latency verification is proposed as a candidate method for operationalizing U.S. FDA Draft Guidance FDA-2024-D-4488 robustness requirements at

In Transformer models, non-GEMM operations often dominate hardware cost due to their nonlinear nature . We propose a hardware-efficient Softmax and LayerNorm with Guaranteed Normalization for Edge devices . Compared to the state of the art, we achieve 11x and 14x reduction in area .

Efficient hardware implementation of nonlinear activation functions is a crucial task in deploying artificial neural networks on resource-constrained and edge devices such as Field-Programmable Gate Arrays . The proposed approach leverages the mathematical relationship between the sigmoid and hyperbolic tangent functions .

Tessera is a reference architecture for inline, cache-line granularity weight decryption on UMA edge accelerators . The design intercepts 64-byte AXI bursts, computing AES-256-CTR keystreams in parallel with DRAM fetches . This streams plaintext directly into isolated NPU SRAM, creating a transient memory footprint confined to the active tile .

Large language models (LLMs) have advanced rapidly, emerging as versatile tools across fields . However, performing LLM inference at the network edge remains challenging due to their large memory and compute demands . This survey outlines the challenges specific to LLM edge inference and provides a comprehensive overview of recent progress .

AI WiFi offload is emerging as a promising approach for providing large language model services to resource-constrained wireless devices . Unlike conventional edge computing, LLM inference over WiFi must jointly address heterogeneous model capabilities, wireless contention, uncertain task complexity .

Public clouds increasingly expose heterogeneous hardware, but allocation interface remains built around rigid on-demand and spot service classes . LaissezCloud keeps allocations contestable during execution: tenants and operators update bids online .

Each candidate server is characterised by predictive mean and uncertainty summaries of network latency . Risk is evaluated using a tight Normal approximation complemented by a conservative Cantelli bound . percentile-based scoring coupled with hysteresis stabilizes decisions and suppresses oscillatory switching under short-lived network fluctuations .

In-memory architectures have emerged as a complementary solution to conventional von Neumann systems . This study proposes a parallel in-memory stochastic computing (SC) architecture that implements an end-to-end computation pipeline .

SPAC (Switch and Protocol Adaptive Customization) automates generation of FPGA-based network switches co-optimized for custom protocols and application-specific traffic patterns . SPAC delivers latency reductions ranging from 7.8% to 38.4% across various tasks .

AGNT2 is a three-tier stack purpose-built for agent and microservice coordination on-chain . Autonomous AI agents generate high-frequency, semantically rich service invocations among mutually untrusting principals . Existing chains treat those interactions as generic calldata forcing identity, escrow, dependency ordering, and session state to be encoded above execution layer at the wrong cost point .

The packing stage constitutes 58% and 94% of the entire Versatile Place and Route (VPR) flow runtime on average when mapping a wide variety of benchmarks to the AMD 7-series-like and Altera Stratix-10-like VTR architecture captures .

Cloud vendors offer discounted spot instances to maximize surplus resource utilization, but these instances are subject to the risk of sudden interruption . Traditional pricing datasets have been employed to predict this risk, yet recent policy changes by cloud vendors have diminished their effectiveness . Paper proposes SpotVista, a system that recommends a resource pool of reliable and cost-efficient multi-node spot instances .

Incisor is a cloud HPC job submission system for the ex ante instance selection problem: choosing suitable hardware in the challenging but common setting where only the executable, inputs, and invocation commands are available at submission time . In practice, this task is manual and expertise-intensive, requiring users to combine incomplete knowledge of rapidly evolving cloud offerings with workload-specific intuition, static analysis, and systems reasoning to infer hardware constraints .

FreeScale reduces computational bubbles caused by severe stragglers and slow blocking communications . FreeScale achieves up to 90.3% reduction in computational bubbles when applied to real-world workloads running on 256 H100 GPUs .

KubePACS is a Kubernetes-native spot instance provisioning system that constructs node pools optimized for both cost and performance while guaranteeing high availability . Spot instances are widely adopted due to steep discounts compared to on-demand pricing .

Large Language Models (LLMs) have achieved strong performance across natural language and multimodal tasks, yet their practical deployment remains constrained by inference latency and kernel launch overhead . This paper presents a hybrid runtime framework that combines Just-In-Time compilation with CUDA Graph execution to reduce launch overhead while preserving runtime flexibility during decoding .

Multi-Process Service (MPS) improves performance by up to 30% and reduces energy by about 20% . However, under memory contention, MPS suffers severe degradation, worsening performance by around 30% . MIG's full hardware isolation resolves memory contention leading to more consistent improvements .

Modern RISC vector processors rely on the synergy of multi-lane parallelism and chaining to achieve high sustained throughput . But achieved performance often falls substantially short of the theoretical performance bound due to microarchitectural inefficiencies . In this work, we take the open-source RVV processor Ara as the target platform and analyze the sources of its sustained-throughput loss and optimize the design accordingly .

Modern computing workloads commonly involve matrix-matrix multiplication (mmul) as a core computing pattern . Coarse-Grained Reconfigurable Arrays (CGRAs) can flexibly and efficiently support it . Here, we introduce a specialized mmul CGRA kernel schedule, parametrizable across different CGRA sizes .

Lookup-table (LUT) based neural networks can deliver ultra-low latency and excellent hardware efficiency on FPGAs by mapping arithmetic operations directly onto the logic primitives . However, state-of-the-art LUT-aware training (LAT) approaches remain difficult to use in practice . They are often orders of magnitude slower to train than conventional networks and require non-trivial manual tuning .

High-level synthesis (HLS) transforms an algorithmic description of hardware from a higher abstraction (e.g., C/C) into a register-transfer level (RTL) design . Large size of the generated RTL often causes model checking to struggle to conclude within reasonable time or resource limits . We propose utilizing the high-level design features from the HLS flow to construct a set of helper assertions aimed at guiding the model checker and accelerating the verification process .

GR-Evolve is a code evolution framework that leverages an agentic large language model to modify global routing source code using QoR-driven feedback . The framework equips the LLM with persistent contextual knowledge of open-source global routers along with an integrated toolchain for evaluation within the OpenROAD infrastructure . We demonstrate up to 8.72% reduction in post-detailed-routing wirelength over existing baseline routers .

CUDA Graph-based multi-path communication approach achieves up to a 2.95x bandwidth improvement, compared to the single-path UCX (UCT::CUDA-IPC) approach . To the best of our knowledge, our proposed approach is the first to seamlessly integrate CUDA graphs into UCX .

In the noisy intermediate-scale quantum (NISQ) regime, quantum devices contain hardware-specific noise sources which restrict device-invariant error mitigation strategies . We explore transfer learning approaches to apply noise models learned on one quantum device to a different device with the help of a small amount of data .

Diffusion models are a leading paradigm for data generation, but training-free editing typically re-runs the full denoising trajectory for every edit strength . To address this issue, we instead edit near the data manifold, where small local updates can replace repeated re-synthesis .

We study offline-to-online learning in linear contextual bandits with biased offline regression data . We propose Ellipsoidal-MINUCB, which combines a standard online branch with an offline-informed pooled branch and uses offline information only when it tightens uncertainty .

Temporal Graph Neural Networks (TGNNs) have become increasingly popular in recent years due to their superior predictive performance . However, how these models utilize the information to make predictions is rather unexplored, leading to potentially faulty or biased models . This work introduces two novel model-agnostic explainers for local explanations of TGNNs based on Shapley and Owen values .

Self-Supervised Learning (SSL) differs in that it directly optimises the representation space using a loss function to enhance the model's performance across multiple downstream tasks . We demonstrate that the evolution of linear regions correlates with the representation quality by using SplineCam to extract two-dimensional polytopes near the data distribution . Contrastive methods rapidly expand regions over time, whereas self-distillation methods tend to consolidate by merging neighbouring regions .

The PARS model employs an advanced reinforcement learning (RL) architecture, incorporating a cutting-edge soft-actor critic (SAC) model and hyper-parameter optimization methods . Negative-g punishments and other handcrafted features are also taken into account by the system .

The continuous advancement of autonomous driving introduces challenges across multiple disciplines to ensure safe and efficient driving . One promising approach is the use of crowdsourced data from a vehicle fleet, representing road topology and lane-level features . We adopt a Detection Transformer (DETR)-based approach .

Diffusion Templates is a unified and open plugin framework that decouples base-model inference from controllable capability injection . The framework is organized around three components: Template models that map arbitrary task-specific inputs to an intermediate capability representation . All resources will be open sourced, including code, models and datasets .

Embodied AI agents increasingly rely on large language models (LLMs) for planning, yet per-step LLM calls impose severe latency and cost . In AgenticCache, each agent queries a runtime cache of frequent plan transitions . Cache-based plan reuse offers a practical path to low-latency, low-cost embodied agents .

Study attempts to formalize emergent intelligence from the perspective of limit theory . Intelligence emerges as transition from finite to effectively infinite knowledge, authors say . They say intelligence originates from the existence of a parameter-limit architecture .

Federated learning enables a population of clients to collaboratively train machine learning models without exchanging raw data . Standard algorithms such as FedAvg suffer from slow convergence and high communication and memory costs . FedSLoP is guaranteed to converge to a first-order stationary point at a rate of $O(1/\sqrt{NT)$.

Randomized neural networks (RaNNs) are attractive for partial differential equations (PDEs) because they replace expensive end-to-end training with a linear least-squares solve over randomized hidden features . AD-RaNN parameterizes the hidden-feature sampling distribution by a low-dimensional vector p and optimizes only p . The method uses a two-stage strategy: ridge-regularized reduced training for stable distribution-parameter optimization .

Molecular similarity plays a central role in ligand-based drug discovery, such as virtual screening, analog searching and goal-directed molecular generation . Traditional similarity measures, ranging from fingerprint-based Tanimoto coefficients to 3D shape overlays, are often computationally expensive at scale or rely on hand-crafted molecular descriptors .

Speech-only spoken language models (SLMs) lag behind text and text-speech models in performance . Recent discrete autoregressive (AR) SLMs indicate significant computational and data demands to match text models . Since discretizing continuous speech for AR creates bottlenecks, we explore whether continuous diffusion (CD) SLM is more viable .

This article evaluates an artificial intelligence (AI)-based Automatic Ground Collision Avoidance System (AGCAS) designed for advanced jet trainers to enhance operational effectiveness . The system utilizes line-of-sight queries on a terrain server to ensure precise and efficient collision avoidance .

Image processing functions are often required to be insensitive to small geometric perturbations such as rotation, scaling, shearing or translation . Safety-critical applications are required to perform as expected in normal operations . By doing so, Super-DeepG achieves both precision and computational efficiency .

Cancer survival prediction from multi-omics data remains challenging because prognostic signals are high-dimensional, heterogeneous, and distributed across interacting genes and pathways . PathMoG reorganizes genome-scale inputs into 354 KEGG-informed pathway modules, introduces a Hierarchical Omics Modulation module to condition gene-expression representations on mutation, copy number variation, pathway, and clinical context .

All three operate over the continuum--real-valued states evolved by real-valued dynamics--even when the target functions are discrete . We prove primitive recursion admits equivalent characterizations in all three frameworks . No fixed polynomial map can round uniformly to the nearest integer or realize exact phase selection .

Fast Adversarial Training (FAT) has attracted significant attention due to its efficiency in enhancing neural network robustness against adversarial attacks . However, FAT is prone to catastrophic overfitting (CO) wherein models overfit to the specific attack used during training and fail to generalize to others . In this work, we innovatively interpret CO through the lens of backdoor . We conceptualize CO as a weak trigger variant of unlearnable tasks .

Variational Segmentation from Label Proportions (VSLP) is a two-stage framework that infers dense segmentations from global label proportions without any pixel-level annotations . We validate our approach on two public datasets, achieving superior performance over existing weakly supervised and unsupervised methods .

This paper evaluates an advanced jet trainer's utilization of artificial intelligence (AI)-based aircraft aerobatic maneuvers . A multitude of aircraft maneuvers have been simulated using reinforcement learning (RL) agents, which will serve as a training tool for future pilots .

In this work, we extend the class of non-Euclidean neural quantum states (NQS) to new variants including Poincaré RNN and Lorentz GRU . In particular, using larger systems consisting of 100 spins, we found that all four hyperbolic RNN/GRU NQS variants always outperformed their respective Euclidean counterparts .

Fast Adversarial Training (FAT) has proven effective in enhancing model robustness by encouraging networks to learn perturbation-invariant representations . But FAT often suffers from catastrophic overfitting (CO) where the model overfits to the training attack and fails to generalize to unseen ones .

Self-Abstraction Learning (SAL) is a hierarchical framework for deep learning . In SAL, networks are arranged by structural complexity, where the simplest topmost network is trained first and its hidden and output layers serve as guidance for the successively more complex networks below . This top-down sequential guidance effectively mitigates optimization issues, enabling stable training of deep architectures .

Relative Entropy Inverse Reinforcement Learning (RE-IRL) is employed to account for environments where transition probabilities are unknown or inaccessible . To address the challenge of data sparsity, we utilize a $K$-nearest neighbor approach to estimate the observed behavior policy . We propose a statistical testing framework to evaluate the validity of the estimated results .