arXiv Research Digest

AsymLoc is a novel distillation framework that aligns a Student to its Teacher through a combination of a geometry-driven matching objective and a joint detector-descriptor distillation objective, enabling fast, parameter-less nearest-neighbor matching . It achieves up to 95% of the teacher's localization accuracy using a order of magnitude smaller models .

UI-to-Code generation requires vision-language models to produce thousands of tokens of structured HTML/CSS from a single screenshot . Existing compression methods either select tokens at inference time using task-agnostic heuristics, or zero out low-attention features without actually shortening the sequence .

Token pruning has emerged as a mainstream approach for developing efficient Video Large Language Models . This work revisits and advances the two predominant token-pruning paradigms: attention-based selection and similarity-based clustering . We propose Tango, a novel framework designed to optimize the utilization of visual signals .

Chest X-ray report generation (CXR-RG) has potential to substantially alleviate radiologists' workload . But conventional autoregressive vision--language models suffer from high inference latency due to sequential token decoding . Diffusion-based models offer promising alternative through parallel generation, but they still require multiple denoising iterations .

Transformers have emerged as the dominant neural-network architecture, achieving state-of-the-art performance in language processing and computer vision . Softmax operations account for less than 1% of the total operation count, but they can disproportionately bottleneck overall inference latency . We use thin-film lithium niobate (TFLN) Mach-Zehnder modulators as analog nonlinear computational elements to drastically reduce the latency of nonlinear computations .

BERT-as-a-Judge is an encoder-driven approach for assessing answer correctness in reference-based generative settings . It is robust to variations in output phrasing and requires only lightweight training on synthetically annotated question-candidate-reference triplets . It consistently outperforms lexical baseline while matching performance of much larger LLM judges .

Online3R is a new sequential reconstruction framework that is capable of adapting to new scenes through online learning, effectively resolving inconsistency issues . We introduce a set of learnable lightweight visual prompts into a pretrained, frozen geometry foundation model to capture the knowledge of new environments .

Training large language models is constrained by memory requirements . OASIS is an online activation subspace learning algorithm for memory-efficient training . Intermediate activations are projected onto this evolving subspace, reducing memory without modifying forward pass computations .

Existing methods based on Transformer architecture or complex convolutional neural networks often suffer from the "memory wall" bottleneck . We propose novel real-time UHD low-light enhancement network based on geometric feature fusion using Clifford algebra in 2D Euclidean space .

Machine unlearning poses challenges in removing mislabeled, contaminated data from a pretrained model . New metric called relearning convergence delay captures changes in weight space and prediction space . This metric can be used to assess the risk of forgotten data being recovered from the unlearned model .

Large language models undergo alignment training to avoid harmful behaviors, yet the resulting safeguards remain brittle . Fine-tuning on narrow domains can induce emergent misalignment that generalizes broadly . Aligned models exhibit a greater compression of harm generation weights than unaligned counterparts .

Prompt learning is a parameter-efficient approach for vision-language models, yet its robustness under label noise is less investigated . Visual content contains richer and more reliable semantic information, but the prompt itself is highly susceptible to label noise . We propose VisPrompt, a lightweight and robust vision-guided prompt learning framework for noisy-label settings .

Autoregressive diffusion model advances trajectories through short, locally predictable transitions, explicitly modeling the growth of uncertainty over time . This dynamics-centric representation enables fast rollout of thousands of diverse futures from a single image, while maintaining physical plausibility and long-range coherence .

Tendon drives paired with soft muscle actuation enable faster and safer robots . Still, these systems are rarely used in practice due to inherent nonlinearities, friction and hysteresis . So far, these challenges have hindered policy transfer from simulation to real systems .

Immersive Volumetric Videos is a new volumetric media format designed to provide large 6-DoF interaction spaces, audiovisual feedback, and high-resolution, high-frame-rate dynamic content . ImViD is a multi-view, multi-modal dataset built upon a space-oriented capture philosophy .

Energy-Guided Latent Optimization for Concept Erasure (EGLOCE) removes unwanted concepts by re-directing noisy latent during inference . EGLOCE improves concept removal while maintaining image quality and prompt alignment .

Hybrid Quantum-Classical Physics-Informed Neural Network (HQC-PINN) integrates quantum quantum circuits into the PINN framework for hydrological PDE-constrained learning . The inherent stochasticity of quantum measurement provides a natural mechanism for uncertainty quantification without requiring Bayesian inference machinery .

FB-GNN-MBE can reproduce first-principles potential energy surfaces for hierarchically structured systems with manageable accuracy, complexity, and interpretability . It outperformed conventional non-FBN-based models and showed high practicality for large-scale molecular simulations .

The Half-Trek Criterion (HTC) is the primary graphical tool for determining generic identifiability of causal effect coefficients in linear structural equation models . However, HTC is inherently node-wise: it simultaneously resolves all incoming edges of a node, leaving a gap of "inconclusive" causal effects . We introduce a framework that decouples causal identification into two phases: (1) seed function S_0 that identifies an initial set of edges from any external source of

The persistent storage requirements for high-resolution, spatiotemporally evolving fields governed by large-scale and high-dimensional partial differential equations (PDEs) have reached the petabyte-to-exabyte scale . To address this bottleneck, we introduce ANTIC (Adaptive Neural Temporal in situ Compressor)

Model poisoning attacks pose a significant security threat to Federated Learning (FL) Most existing model poisoning attacks rely on collusion, requiring adversarial clients to coordinate by exchanging local benign models and synchronizing their poisoned updates . To address this challenge, we introduce and formalize the \textbf{non-collusive attack model .

In this study, we proposed a deep Swin-Vision Transformer-based transfer learning architecture for robust multi-cancer image classification . The proposed framework integrates a hierarchical Swin Transformer with ResNet50-based convolution features extraction . The model reached 100% test accuracy for lung-colon cancer, segmented leukemia datasets, and up to 99.23% accuracy for breast cancer classification .

Mode-Domain Architecture (MDA) is a bilateral predictive neural network architecture designed to restore missing signals on both the human and robot sides . MDA utilizes a novel Continuous-Orthogonal Mode Decomposition framework . The model achieves ultra-low inference latency of 0.065 ms, outperforming existing benchmarks .

AdaCubic is an auxiliary optimization problem with cubic constraints that dynamically adjusts the weight of the cubic term in Newton's cubic regularized method . We use Hutchinson's method to approximate the Hessian matrix, thereby reducing computational cost .

Just Recognizable Difference (JRD) boosts coding efficiency for machine vision through visibility threshold modeling, but is currently limited to a single-task scenario . To address this issue, we propose a Multi-Task JRD dataset and an Attribute-assisted MT-JRD model for Video Coding for Machines (VCM)

Trans-RAG implements a novel vector space language paradigm where each organization's knowledge exists in a mathematically isolated semantic space . At the core lies vector2Trans, a multi-stage transformation technique that enables queries to dynamically "speak" each organizations's vector space "language"

High-Level Synthesis compiles C/C++ into RTL but exploring pragma-driven optimization choices remains expensive because each design point requires time-consuming synthesis . We propose a differential learning framework for HLS Quality-of- Result (QoR) prediction that learns from kernel--design pairs .

Large Language Models for code generation can replicate insecure patterns from their training data . To mitigate this, a common strategy for security hardening is to fine-tune models using supervision derived from the final transformer layer . This design may suffer from a final-layer bottleneck: vulnerability-discriminative cues can be distributed across layers and become less detectable near the output representations optimized for next-token prediction .

The transition of Multi-Agent Reinforcement Learning (MARL) policies from simulated cyber wargames to operational Security Operations Centers (SOCs) is fundamentally bottlenecked by the Sim2Real gap . NetForge enforces Zero-Trust Network Access (ZTNA) constraints and requires defenders to process NLP-encoded SIEM telemetry .

In this paper, we propose a stochastic-dimension frozen sampled neural network for solving a class of high-dimensional Gross-Pitaevskii equations . SD-FSNN is unbiased across all dimensions and its computational cost is independent of the dimension . We randomly sample the hidden weights and biases of the neural network, outperforming gradient-based optimization methods in terms of training time and accuracy .

We propose a hybrid physics-informed framework for solving families of parametric linear partial differential equations (PDEs) by combining a meta-learned predictor with a least-squares corrector . The predictor is a shallow task-conditioned model that maps query coordinates and PDE parameters to solution values while internally generating an interpretable, task-adaptive Gaussian basis geometry . This predictor-generated geometry is transferred to a second-stage corrector, which augments it with a

Trusted multi-view classification typically relies on view-wise evidential fusion process . Different views often differ in feature space, noise level, and semantic granularity . As a result, the uncertainty used for fusion can be dominated by branch-specific scale bias rather than true sample-level reliability .

The transition to electric mobility hinges on maximising aggregate adoption while facilitating equitable access . The study examines whether the 'charging divide' between households with and without off-street parking reflects a genuine infrastructure constraint or a by-product of socio-economic disparity .

Anomaly detection (AD) in chemical processes based on deep learning offers significant opportunities but requires large, diverse, and well-annotated training datasets . In a recent work, we introduced a large, fully annotated experimental dataset for batch distillation under normal and anomalous operating conditions . In the present study, we augment this dataset with a corresponding simulation dataset, creating a novel hybrid dataset .

CORA is a post-policy, pre-action safeguarding framework that provides statistical guarantees on harmful executed actions . CORA reformulates safety as selective action execution: we train a Guardian model to estimate action-conditional risk for each proposed step .

Synthetic data can be used in machine learning applications and research . The proposed solution surpasses the state-of-the-art generative methods and achieves reference MMD scores faster than modern deep learning solutions .

PDYffusion is a dynamics-informed diffusion framework that integrates PDE-based regularization and uncertainty-aware forecasting for stable long-term prediction . The proposed method consists of two key components: a PDE regularized interpolator and a UKF-based forecaster . The interpolator incorporates a differential operator to enforce physically consistent intermediate states, while the forecaster leverages the Unscented Kalman Filter to explicitly model uncertainty .

Evidence-grounded reasoning requires more than attaching text to a prediction . A model should make decisions that depend on whether the provided evidence supports the target claim . In practice, this often fails because supervision is weak, evidence is only loosely tied to the claim, and evaluation does not test evidence dependence directly . A key contribution is a supervision construction procedure that generates explicit support examples with semantically controlled non-support examples .

World models have emerged as a unifying paradigm for learning latent dynamics, simulating counterfactual futures, and supporting planning under uncertainty . This is because epidemic decision-making requires reasoning about latent disease burden, imperfect surveillance signals and policy-dependent surveillance signals .

We propose RecaLLM, a set of reasoning language models post-trained to make effective use of long-context information . We observe consistent gains at context windows of up to 128K tokens using training samples of at most 10K tokens .

Safety guarantees are a prerequisite to the deployment of reinforcement learning agents in safety-critical tasks . Often, deployment environments exhibit non-stationary dynamics or are subject to changing performance goals, requiring updates to the learned policy . We propose a novel a priori approach to safe policy updates in continual RL .

Data were collected from 23 healthy subjects (16 males and seven females) aged between 23 and 62 years using an ambulatory device . Participants followed a standardized activity routine involving five distinct activities: lying, sitting, standing, walking, and jogging .

Rays as Pixels is a Video Diffusion Model that learns a joint distribution over videos and camera trajectories . We represent each camera as dense ray pixels (raxels) and denoise them jointly with video frames through Decoupled Self-Cross Attention mechanism .

Combinatorial multi-armed bandits provide a fundamental online decision-making environment where a decision-maker interacts with an environment across $T$ time steps . The goal is to minimize regret, defined as the loss compared to the optimal fixed action .

Successful machine learning on graphs or networks requires embeddings that preserve the graph structure . NOMAD implements proximity-based models proposed in the widely popular LINE algorithm . We propose several practical trade-offs to improve the scalability and communication overheads of irregular and distributed graph embedding methods .

Automated Instruction Revision (AIR) is a rule-induction-based method for adapting large language models to downstream tasks . The paper argues that adaptation performance is strongly task-dependent: no single method dominates across all settings . AIR is most promising when task behavior can be captured by compact, interpretable instruction rules .

PhysInOne is a large-scale synthetic dataset addressing the critical scarcity of physically-grounded training data for AI systems . It provides 2 million videos across 153,810 dynamic 3D scenes, covering 71 basic physical phenomena .

Learning-to-Defer routes each input to an expert that minimizes expected cost, but it assumes that the information available to every expert is fixed at decision time . We show that a broad family of natural separated surrogates, which learn routing and advice with distinct heads, is inconsistent even in the smallest non-trivial setting . We then introduce an augmented surrogate that operates on the composite expert--advice action space .

We study the population loss landscape of two-layer ReLU networks of the form $k=1/k$ in a realisable teacher-student setting with Gaussian covariates . We show that local minima admit an exact low-dimensional representation in terms of summary statistics . We further establish a direct link with one-pass SGD . This perspective reveals a hierarchical structure of minima: they are typically isolated in the well-specified regime, but become connected by flat directions as

Training Transformer language models is expensive, as performance typically improves with increasing dataset size and computational budget . We isolate dataset-size effects using a strongly reduced attention-only decoder architecture . We observe smooth performance improvements accompanied by clear diminishing returns .

This paper argues that a one-size-fits-all approach to specifying consent for the use of creative works in generative AI is insufficient . Real-world ownership and rights holder structures make the status quo of binary consent with opt-in by default untenable .

Persona-E$^2$ (Persona-Event2Emotion) is a large-scale dataset grounded in annotated MBTI and Big Five traits to capture reader-based emotional variations across news, social media, and life narratives .

Three-dimensional (3D) data visualizations, such as surface plots, are vital in STEM fields from biomedical imaging to spectroscopy, yet remain largely inaccessible to blind and low-vision (BLV) people . We conducted an Experience-Based Co-Design with BLV co-designers with expertise in non-visual data representations .

Audio large language models (ALLMs) enable rich speech-text interaction, but they also introduce jailbreak vulnerabilities in the audio modality . Authors propose GRM, a utility-aware frequency-selective jailbreak framework . GRM achieves an average Jailbreak Success Rate of 88.46% while providing a better attack-utility trade-off than representative baselines .

Camera Artist is a multi-agent framework that models a real-world filmmaking workflow to generate narrative videos with explicit cinematic language . Camera Artist builds upon established agentic pipelines and introduces a dedicated Cinematography Shot Agent .

Audio super-resolution aims to recover missing high-frequency details from bandwidth-limited low-resolution audio, thereby improving the naturalness and perceptual quality of the reconstructed signal . LatentFlowSR leverages conditional flow matching (CFM) within a latent representation space .

Contrastive Language-Audio Pre-training (CLAP) can support abusive speech detection directly from audio . Abusive speech detection is becoming increasingly important as social media shifts towards voice-based interaction .

Neurocognitive disorders (NCDs) are globally prevalent and require scalable screening methods for proactive management . Prior research has explored the potential of technologies like conversational AI (CAI) to administer NCD screening tests . But challenges remain in designing CAI-based solutions that make routine screening socially acceptable, engaging, and encouraging early medical consultation .

Most in-the-wild two-speaker dialogue is available only as degraded monaural mixtures . We propose DialogueSidon, a model for joint restoration and separation of degraded mixtures of dialogue audio .

A growing body of research has found that AI can produce large persuasive effects on people's attitudes, but whether AI can persuade people to take consequential real-world actions has remained unclear . In two large preregistered experiments, we used conversational AI models to persuade participants on a range of attitudinal and behavioural outcomes, including signing real petitions and donating money to charity . We found sizable AI persuasion effects on these behavioural outcomes .

Drug instructions are crucial for guiding the rational use of medication . We conduct a visualization design study to enhance the comprehension of OTC drug instructions . We devise two tailored drug instruction designs for different audience groups through an iterative design process .

In most privacy-policy consent flows, agreement is operationalized as a single click at the end of a long, opaque policy document . Recent privacy-law scholarship has argued for a standard of demonstrably informed consent . Authors say pedagogical friction can strengthen evidentiary basis of consent and clarify what it costs in time and burden .

Large Language Models are widely used by students, yet their tendency to provide fast and complete answers may discourage reflection and foster overconfidence . We examined how alternative LLM interaction designs support deeper thinking without excessively increasing cognitive burden . Future-self explanations imposed higher cognitive workload yet yielded the closest alignment between perceived and actual understanding .

Many-objective optimisation involves optimising problems with more than three objectives . As the number of objectives increases, number of solutions needed to adequately represent the entire Pareto front typically grows substantially . This makes it challenging, if not infeasible, to design a search algorithm capable of effectively exploring the entire . Bayesian optimisation, where sample efficiency is critical, may be more useful to focus on finding a single solution .

Digital Signal Processing (DDSP) pipelines for voice conversion rely on subtractive synthesis . In DDSP-QbE, the excitation is generated via phase accumulation, producing a sawtooth-like waveform whose abrupt discontinuities introduce aliasing artefacts that manifest as buzziness and spectral distortion .

Novices often face cognitive biases such as premature closure and over-reliance on heuristics . Novices struggle to transfer diagnostic strategies to new cases . Scenario-based learning (SBL) enhanced by Learning Analytics (LA) and large language models offers a promising approach .

EquiformerV3, the third generation of the $SE(3)$-equivariant graph attention Transformer, is designed to advance all three dimensions: efficiency, expressivity, and generality . SwiGLU-$S^2$ activations and smooth-cutoff attention enable accurate modeling of smoothly varying potential energy surfaces .

Large Vision Language Models (LVLMs) achieve strong multimodal reasoning but frequently exhibit hallucinations and incorrect responses with high certainty . VL-Calibration decouples confidence into visual and reasoning confidence . We propose token-level advantage reweighting to focus optimization on tokens based on visual certainty .

Modern fake news arises through human-AI collaboration, where strategic inaccuracies are embedded within otherwise accurate and credible narratives . Mixed-truth cases represent a realistic and consequential threat, yet they remain underrepresented in existing benchmarks . We introduce a synthetic benchmark containing 6,798 fake news articles .

Visual Retrieval-Augmented Generation (VRAG) empowers Vision-Language Models to retrieve and reason over visually rich documents . The accumulation of visual tokens across retrieved pages dilutes context and causes cognitive overload, leading agents to deviate from their search objective .

AI agents increasingly operate in multi-agent environments where outcomes depend on coordination . While LLMs coordinate extremely well on similar actions, they lag behind humans in sustaining heterogeneity when divergence is rewarded .

Process Reward Agents (PRA) is a test-time method for providing domain-grounded, online, step-wise rewards to a frozen policy . In contrast to prior retrieval-augmented PRMs, PRA enables search-based decoding to rank and prune candidate trajectories at every generation step . Experiments on multiple medical reasoning benchmarks demonstrate that PRA consistently outperforms strong baselines .

SafeMind unifies probabilistic Control Barrier Functions with semantic context understanding and meta-adaptive risk calibration . A semantics-to-constraint encoder modulates safety margins using perceptual or language cues . SafeMind reduces safety violations by 3--10x and energy consumption by 10--15% relative to state-of-the-art CBF, MPC and hybrid RL baselines .

Social media users embedded within larger audiences, with lower posting frequency and perceived support, are less likely to express their opinions . Those who do speak often adjust their expressed views to align with perceived group norms .

We introduce Intent Lenses, a conceptual primitive for intent-mediated note generation and sensemaking . We present an interactive system that infers lenses from presentation captures to generate structured visual notes on a spatial canvas . Users can further add, link, and arrange lenses across captures to support exploration .

EdgeFlow is a mobile LLM inference framework that mitigates the cold start issue by adaptively adjusting the precisions of LLM parameters . EdgeFlow reduces cold-start latency by up to 4.07x compared with three state-of-the-art mobile LLMs .

Large Language Model (LLM) inference on edge Neural Processing Units (NPUs) is fundamentally constrained by limited on-chip memory capacity . We propose SHIELD, a lifecycle-aware segmented eDRAM architecture that jointly exploits temporal residency and bit-level sensitivity in bfloat16 (BF16) activations . SHIELD isolates the sign and exponent fields from the mantissa, disables refresh for transient QO mantissas, and applies relaxed refresh to persistent

We study a lightweight ledger protocol for intermittent and noisy networks . We evaluate the protocol with a discrete-event simulator under controlled partitions and two network regimes . The main result is that quarantine alone does not materially improve agreement or recovery under noisy conditions .

Memory Dependence Prediction (MDP) is a speculative technique to determine which stores, if any, a given load will depend on . Area-constrained cores are increasingly relevant in various applications such as energy-efficient or edge systems . This paper proposes that targeting the predictor working set is as effective as growing the predictor .

AdecPilot employs a Hierarchical Implicit Implicit Termi- nation protocol to enforce deterministic stops and prevent post- completion hallucinations . The source code is available at https://anonymous.4open.science/r/Anonymous_code- B8AB .

RL-driven compiler jointly optimizes ASIC architecture, memory hierarchy, and workload partitioning for AI inference across 3nm to 28nm . The design space is formulated as a single Markov Decision Process with mixed discrete-continuous actions and a unified Power-Performance-Area objective .

Chain Federated Fine-Tuning (ChainFed) forgoes end-to-end updates in favor of a sequential, layer-by-layer manner . ChainFed trains the initial adapter to convergence, freezes its weights, and then proceeds to the next . This iterative train-and-freeze process forms an optimization chain, gradually enhancing the model's task-specific proficiency . Extensive experiments demonstrate the superiority of ChainFed over existing methods .

MATCHA is a unified DNN deployment framework that generates highly concurrent schedules for parallel, heterogeneous accelerators . It uses constraint programming to optimize L3/L2 memory allocation and scheduling . Pattern matching, tiling, and mapping across individual HW units enables parallel execution .

Real-time tsunami early warning relies on distributed sensor networks to infer seismic sources and seafloor motion . Optimizing these networks via Bayesian optimal experimental design (OED) is exceptionally challenging for systems governed by hyperbolic partial differential equations . We present a scalable Bayesian OED framework for linear time-invariant systems .

The Hardware Lottery posits that research directions are dictated by available silicon compute platforms . We identify a derivative phenomenon where model architectures are optimized for cloud throughput at the expense of algorithmic efficiency . We argue for decoupling cloud-scale saturation strategies from core architectural design to preserve the viability of single-user, real-time edge intelligence .

Transformer neural networks achieve state-of-the-art accuracy across language and vision tasks, but deployment on embedded hardware is hindered by stringent area, latency, and energy constraints . During inference, performance and efficiency are increasingly dominated by the Key--Value (KV) cache .

Reference-Oriented Storage (ROS) is a new storage abstraction for RL weight transfer . ROS presents the illusion that certain versions of the model weights are stored and can be fetched on demand . ROS does not physically store any copies of the weights; instead it tracks the workers that hold these weights on GPUs for inference .

The rapid advancement of neuromorphic technology aims to address the memory wall challenge inherent in conventional von Neumann architecture . While designed to bring computation closer to memory through distributed architectures, our findings indicate that on-chip memory systems have become significant consumers of area and energy .

Placement of Kubernetes control-plane nodes is critical to ensuring cluster reliability, scalability, and performance . Existing initialisation procedures typically select control-planes arbitrarily, without considering node resource capacity or network topology .

Self-attention in Transformers generates dynamic operands that force conventional Compute-in-Memory accelerators into costly non-volatile memory (NVM) reprogramming cycles, degrading throughput and stressing device endurance . We present TrilinearCIM, a Double-Gate FeFET (DG-FeFET)-based architecture that uses back-gate modulation to realize a three-operand multiply-accumulate primitive .

Foundry is a template-based CUDA graph context materialization system . It persists both graph topology and execution context during offline processing . Foundry reduces cold-start latency by up to 99%, cutting the initialization time of Qwen3-235B-A22B from 10 minutes to 3.9 seconds .

Sustaining exascale performance in production requires engineering choices and operational practices that emerge only under real deployment constraints and demand coordination across system layers . HPL-MxP reached 11.64EF/s, an 11.5x speedup over FP64 enabled by mixed-precision arithmetic .

This work introduces an ultralow-power voltage-to-spike encoder . The encoder achieves a deviation of less than 5.6 percent from linearity over 0.1-0.4 V input . Fabricated in TSMC 0.18-um CMOS and operating at VDD = 0.5 V with 2-27 nA reference current .

Assertion-Based Verification (ABV) is key to reducing debugging time . But existing methods pursue indiscriminate verification, aiming for maximal coverage without considering signal criticality . We propose AgileAssert, a key signal-driven assertion generation framework .

Advances in GPU compute throughput and memory capacity brings significant opportunities to a wide range of workloads . Multi-Instance GPU (MIG) is a promising approach to improve utilization by partitioning GPU compute and memory resources into fixed-size slices with isolation .

Visibility Graph Analysis (VGA) is a key space syntax method for understanding how spatial configuration shapes human movement . Its reliance on all-pairs BFS computation limits practical application to small study areas . We present a system combining three techniques to scale VGA to city-scale problems .

LegoDiffusion outperforms existing diffusion workflow serving systems, sustaining up to 3x higher request rates and tolerating up to 8x higher burst traffic . The system decomposes a workflow into loosely coupled model-execution nodes that can be independently managed .

With the development of deep neural network (DNN) enabled applications, achieving high hardware resource efficiency on diverse workloads is non-trivial in heterogeneous computing platforms . Compared with prior works, our design can achieve 1.3x - 5x throughput and hardware efficiency . We also evaluate the FILCO framework on the 7nm AMD Versal VCK190 board .

We develop the first theoretically-efficient algorithm for maintaining the maximal independent set (MIS) of a graph in the parallel batch-dynamic setting . In this setting, a graph is updated with batches of edge insertions/deletions . For a batch of $b$ updates, our algorithm has $O(b \log^3 n)$ expected work and polylogarithmic depth with high probability .

GROMACS is a de-facto standard for classical Molecular Dynamics (MD) The rise of AI-driven interatomic potentials that pursue near-quantum accuracy at MD throughput now poses a significant challenge: embedding neural-network inference into multi-GPU simulations retaining high-performance .

Drift-Aware Multi-Scale Dynamic Learning (DA-MSDL) framework is proposed to maintain robust multi-output predictive performance via online adaptive mechanisms on nonstationary data streams . The framework employs a multi-scale bi-branch convolutional network as its backbone to disentangle local fluctuations from long-term trends .

Distributed online convex optimization (D-OCO) is a powerful paradigm for modeling distributed scenarios with streaming data . However, the communication cost between local learners and the central server is substantial in large-scale applications .

Robust geo-localization in changing environmental conditions is critical for long-term aerial autonomy . Existing continual learning methods often fail here because geographic features exhibit severe intra-class variations . To respect strict onboard storage constraints, our pipeline decouples geographic knowledge into static satellite anchors and a dynamic experience replay buffer .

Pretraining is the cornerstone of Large Language Models, dominating the vast majority of computational budget and data to serve as the primary engine for their capabilities . We hypothesize that the geometric "closeness" of task-specific minima is intrinsically linked to downstream generalization . We propose the Nexus optimizer, which encourages the closeness of these minima by maximizing gradient similarity during optimization . Nexus reduces the out-of-distribution loss by 0.012 and yields up to a 15

Temporal Patch Shuffle (TPS) is a simple and model-agnostic data augmentation method for forecasting . It extracts overlapping temporal patches, selectively shuffles a subset of patches using variance-based ordering as a conservative heuristic . This design increases sample diversity while preserving forecast-consistent local temporal structure .

In partial multi-label learning, each instance is associated with a set of candidate labels containing both ground-truth and noisy labels . The presence of noisy labels disrupts the correspondence between features and labels, degrading classification performance . We propose a novel PML method based on feature-label modal alignment (PML-MA)

PE-MAMoE equips each UAV with a sparsely gated mixture of experts actor whose router selects a single specialist per step . Phase Controller injects brief, expert-only stochastic perturbations after phase switches, resets the action log-standard-deviation, entropy and learning rate, and schedules the router temperature, all to re-plasticize the policy without destabilizing safe behaviors .

CrossAbSense is a framework of property-specific neural oracles that combine frozen protein language model encoders with configurable attention decoders . On the GDPa1 benchmark of 242 therapeutic IgGs, our oracles achieve notable improvements of 12--20\% over established baselines on three of five developability assays .

Label noise in multi-label learning (MLL) poses significant challenges for model training . We propose a novel weakly-supervised clustering approach for PML . WSC-PML employs a three-stage process: initial prototype learning from noisy labels, adaptive confidence-based weak supervision construction and joint optimization via iterative clustering refinement .

HFD-TM (Hierarchical Flow-Decomposition for Turning Movement Prediction) is a deep learning framework that predicts turning movements by first forecasting corridor through-movements and then expanding these predictions to individual turning streams . The design is motivated by empirical traffic structure, where corridor flows account for 65.1% of total volume, exhibit lower volatility than turning movements .

A novel stability-enhanced Gaussian process variational autoencoder (SEGP-VAE) is proposed for indirectly training a low-dimensional linear time invariant (LTI) system, using high-dimensional video data . The mean and covariance function of the novel SEGP prior are derived from the definition of an LTI system .

This paper presents an online intention prediction framework for estimating the goal state of autonomous systems in real time . The problem is formulated as an inverse optimal control / inverse reinforcement learning task, with intention treated as a parameter in the objective .

We consider machine learning tasks with low-rank functional tree tensor networks (TTN) as the learning model . We propose a natural Riemannian gradient descent type approach applicable to arbitrary losses which is based on the natural gradient by Amari .

Standard object detectors typically treat architectural elements independently . We address this limitation by augmenting the YOLOv8 training objective with a custom lightweight alignment loss . This regularization encourages grid-consistent arrangements of bounding boxes during training .

The King Wen sequence of the I-Ching (c. 1000 BC) orders 64 hexagrams in a pattern that has puzzled scholars for three millennia . We present a rigorous statistical characterization of this ordering using Monte Carlo permutation analysis against 100,000 random baselines . We find that the sequence has four statistically significant properties: higher-than-random transition distance, negative lag-1 autocorrelation, yang-balanced groups of four, and asymmetric within-pair

We develop a predictive-first optimisation framework for streaming hidden Markov models . We assume access to regime-specific predictive models whose parameters are learned online while maintaining a fixed transition prior over regimes . Our objective is to sequentially identify latent regimes while maintaining accurate step-ahead predictive distributions .

Cloud providers must assign heterogeneous compute resources to workflow DAGs while balancing competing objectives such as completion time, cost, and energy consumption . We identify specific out-of-distribution conditions under which GNN-based deep reinforcement learning schedulers fail . We demonstrate that performance degradation stems from structural mismatches between training and deployment environments .

Existing benchmarks evaluate models against external standards but do not measure whether models understand and enforce their own stated boundaries . We introduce the Symbolic-Neural Consistency Audit (SNCA), a framework that extracts a model's self-stated safety rules via structured prompts, formalizes them as typed predicates .

MixFlow trains a flow model on linear mixtures of a fixed unconditional distribution and a $κ\texttt{-FC$-based distribution . This simple mixture improves the alignment between the source and data, provides better generation quality with less required sampling steps, and accelerates training convergence considerably .

We develop a theory of generalization and scaling for Mixture-of-Experts (MoE) Transformers . By conditioning on fixed routing patterns and union-bounding across them, we derive a sup-norm covering-number bound whose metric entropy scales with the active parameter budget and incurs a MoE-specific routing overhead .

Maximum entropy reinforcement learning (MaxEnt RL) has become a standard framework for sequential decision making . MaxEnt RL's standard Gaussian policy parameterization is inherently unimodal, limiting its ability to model complex multimodal action distributions . We propose Truncated Rectified Rectified Flow Policy (TRFP), a framework built on a hybrid deterministic-stochastic architecture .

We introduce a novel learning framework for accelerated Monte Carlo (MC) dose calculation termed Energy-Shifting . This approach leverages deep learning to synthesize 6 MV TrueBeam Linear Accelerator (LINAC) dose distributions directly from monoenergetic inputs under identical beam configurations . We propose a novel 3D architecture termed TransUNetSE3D, featuring Transformer blocks for global context and Residual Squeeze-and-Excitation (SE) modules for adaptive channel-

This paper introduces a score-driven rating system that employs the score, i.e. the gradient of the log-likelihood, as the updating mechanism for player and team ratings . The proposed framework extends beyond simple win/loss game outcomes and accommodates a wide range of game results, such as point differences, win/draw/loss outcomes .

Unobserved confounding is a key challenge when estimating causal effects from a treatment on an outcome in scientific applications . We assume that we observe a single, potentially multi-dimensional proxy variable of the unobserved confounder . We develop a neural network based estimation framework, SPICE-Net, to estimate causal effects .

Composed image retrieval (CIR) aims to retrieve a target image that depicts a reference image modified by a textual description . Recent vision-language models (VLMs) achieve promising CIR performance by embedding images and text into a shared space . Instead of relying solely on embedding similarity, FIRE-CIR performs question-driven visual reasoning .