arXiv preprints from January 1, 2026 through September 10, 2026 — 11:19:30 EST
Robust generalization under distribution shift remains a key challenge for conditional generative modeling: conditional flow-based methods often fit the training conditions well but fail to extrapolate to unseen ones. We introduce SP-FM, a shortest-path flow-matching framework that improves out-of-distribution (OOD) generalization by...
How can populations of learners develop coordinated, diverse behaviors without explicit communication or diversity incentives? We demonstrate that competition alone is sufficient to induce emergent specialization -- learners spontaneously partition into specialists for different environmental regimes through competitive dynamics,...
We study the emergence of multi-step reasoning in deep Transformer language models through a geometric and statistical-physics lens. Treating the hidden-state trajectory as a flow on an implicit Riemannian manifold, we analyze the layerwise covariance spectrum of activations, where $C^{(\ell)}=\mathbb{E}[h^{(\ell)}h^{(\ell)\top}]$,...
Research waste in biomedical science is driven by redundant studies, incomplete reporting, and the limited scalability of traditional evidence synthesis workflows. We present an AI co-scientist for scalable and transparent knowledge synthesis based on explicit formalization of Population, Intervention, Comparator, Outcome, and Study...
Two widely adopted techniques for LLM inference serving systems today are hybrid batching and disaggregated serving. A hybrid batch combines prefill and decode tokens of different requests in the same batch to improve resource utilization and throughput at the cost of increased latency per token. In contrast, disaggregated serving...
Time series foundation models have recently gained a lot of attention due to their ability to model complex time series data encompassing different domains including traffic, energy, and weather. Although they exhibit strong average zero-shot performance on forecasting tasks, their predictions on certain critical regions of the data...
Warning: This paper may contain examples and topics that may be disturbing to some readers, especially survivors of miscarriage and sexual violence. People affected by abortion, miscarriage, or sexual violence often share their experiences on social media to express emotions and seek support. On public platforms like Reddit, where...
Most privacy regulations function as a passive defensive shield that users must wield themselves. Users are incessantly asked to "opt-in" or "opt-out" of data collection, forced to make defensive decisions whose consequences are increasingly difficult to predict. Viewed through the Johari Window, a psychological framework of...
Enterprise back office workflows require agentic systems that are auditable, policy-aligned, and operationally predictable, capabilities that generic multi-agent setups often fail to deliver. We present POLARIS (Policy-Aware LLM Agentic Reasoning for Integrated Systems), a governed orchestration framework that treats automation as...
Causal discovery based on Independent Component Analysis (ICA) has achieved remarkable success through the LiNGAM framework, which exploits non-Gaussianity and independence of noise variables to identify causal order. However, classical LiNGAM methods rely on the strong assumption that there exists an ordering under which the noise...
With the rapid development of artificial intelligence (AI), machines are increasingly evolving into intelligent agents, and the human-machine relationship is shifting from traditional "human-computer interaction" toward a new paradigm of "human-AI collaboration." However, technology-centered approaches to AI development have gradually...
The group-in-a-box (GIB) layout is an efficient graph drawing method designed to visualize the group structure of graphs. The layout communicates group sizes and both within-group and between-group network structures simultaneously. The layout is characterized by its composition of multiple elements, including nodes, edges, and boxes....
Connected automated vehicles (CAVs) possess the ability to communicate and coordinate with one another, enabling cooperative platooning that enhances both energy efficiency and traffic flow. However, during the initial stage of CAV deployment, the sparse distribution of CAVs among human-driven vehicles reduces the likelihood of...
GPU-accelerated Inverted File (IVF) index is one of the industry standards for large-scale vector search but relies on static VRAM layouts that hinder real-time mutability. Our benchmark and analysis reveal that existing designs of GPU IVF necessitate expensive CPU-GPU data transfers for index updates, causing system latency to spike...
Remote palpation enables noninvasive tissue examination in telemedicine, yet current tactile displays often lack the fidelity to convey both large-scale forces and fine spatial details. This study introduces a hybrid fingertip display comprising a rigid platform and a $4\times4$ soft pneumatic tactile display (4.93 mm displacement and...
With the growing deployment of small satellites (such as CubeSats, Nanosats, Picosats, and Femtosats) in Low Earth Orbit (LEO) for targeted applications like imaging, communication, data storage, and rendezvous-docking mission, there is increasing attention on orbit maintenance and attitude control. A common approach for active orbit...
Robot design is a nontrivial process that involves careful consideration of multiple criteria, including user specifications, kinematic structures, and visual appearance. Therefore, the design process often relies heavily on domain expertise and significant human effort. The majority of current methods are rule-based, requiring the...
In this paper, we study the problem of quantitative group testing (QGT) and analyze the performance of three models: the noiseless model, the additive Gaussian noise model, and the noisy Z-channel model. For each model, we analyze two algorithmic approaches: a linear estimator based on correlation scores, and a least squares estimator...
Wildfire monitoring requires high-resolution atmospheric measurements, yet low-cost sensors on Unmanned Aerial Vehicles (UAVs) exhibit baseline drift, cross-sensitivity, and response lag that corrupt concentration estimates. Traditional deep learning denoising approaches demand large datasets impractical to obtain from limited UAV...
Mathematical problem generation (MPG) is a significant research direction in the field of intelligent education. In recent years, the rapid development of large language models (LLMs) has enabled new technological approaches to problem-generation tasks. Although existing LLMs can achieve high correctness rates, they generally lack...
The next-token prediction (NTP) objective has been foundational in the development of modern large language models (LLMs), driving advances in fluency and generalization. However, NTP operates at the \textit{token} level, treating deviations from a single reference continuation as errors even when alternative continuations are equally...
This paper explores the suspicious alignment phenomenon in stochastic gradient descent (SGD) under ill-conditioned optimization, where the Hessian spectrum splits into dominant and bulk subspaces. This phenomenon describes the behavior of gradient alignment in SGD updates. Specifically, during the initial phase of SGD updates, the...
Groundwater vulnerability is a major concern in arid regions worldwide, where population growth and intensive agriculture increase the risks of depletion and contamination. This study proposes a hybrid groundwater vulnerability assessment framework that improves the conventional DRASTIC model by integrating land-use data and applying...
We present the first comprehensive analysis of ARM MTE hardware performance on four different microarchitectures: ARM Big (A7x), Little (A5x), and Performance (Cortex-X) cores on the Google Pixel 8 and Pixel 9, and on Ampere Computing's AmpereOne CPU core. We also include preliminary analysis of MTE on Apple's M5 chip. We investigate...
The enterprise governance of Generative AI (GenAI) in regulated sectors, such as Human Resources (HR), demands scalable yet reproducible auditing mechanisms. While Large Language Model (LLM)-as-a-Judge approaches offer scalability, their reliability in evaluating adherence of different types of system instructions remains unverified....