The Archive

Systematic bibliometric survey of Tibetan NLP literature (2004–2026) reveals rapid growth since 2018 across corpora, segmentation, and speech tasks.

Model-free Bayesian clustering framework using martingale posteriors and Wasserstein summarization to avoid MCMC and likelihood specification.

ProgramGym introduces scalable training environments for coding agents to reproduce long-horizon programs, addressing agent design discovery and verification.

PlacePulse creates transferable geospatial embeddings from structural demographic and business data at fine spatial resolution for foundation models.

MoPrO-SD optimizes LLM prompts for stance detection by decomposing into semantic modules and refining via class-specific feedback.

BLOKS accelerates kernel ridge regression via sketched block preconditioning to enable scalability without excessive memory overhead.

Mechanistic analysis of LLM rubric design shows structural organization, semantic specificity, and score aggregation each impact evaluation effectiveness.

Category-conditioned expert-selection biases in Mixture-of-Experts models mitigate data imbalance by preferentially routing minority categories to capacity.

Information bottleneck analysis explains classifier-free guidance activation timing in diffusion models, replacing heuristic schedules with principled theory.

AgentSWE benchmark evaluates whether coding agents faithfully build agents matching user requirements across creation, editing, and deployment stages.

SWE-Critix introduces post-training verifiers for judging long-horizon coding-agent trajectories without oracles in software engineering tasks.

AlphaNIST applies RL from binary task rewards to learn complex robotic assembly without dense guidance, advancing NIST benchmark performance.

EmoGavel framework uses grounded RL with verifiable rewards for multimodal emotion preference learning with explicit localization supervision.

Study systematically quantifies temporal, semantic, and population alignment violations in MIMIC-IV preprocessing pipelines, exposing clinical ML risks.

T²Mem defers memory construction to inference time, enabling LLMs to progressively build query-specific context over long interaction histories.

Interprets Mamba-3's input-dependent rotations as underdamped oscillatory dynamics, showing benefits for state-tracking and length generalization.

Explores attention sink phenomena in text-to-image Transformers across diffusion and autoregressive models, extending LLM findings to vision tasks.

AgentTune source-available library unifies agent lifecycle: building, training, evaluation, distillation with EventLog-based trajectory management.

PDE-PFN applies prior-data fitted networks to PDE solving via in-context learning, using self/cross-attention for posterior predictive inference.

ChemDiffAgent applies masked diffusion to chemical reasoning agents, enabling bidirectional molecular understanding and multi-source tool integration beyond autoregressive constraints.

DuoRate optimizes MoE serving by allocating per-expert precision separately for storage and decode traffic, addressing the store-read gap in quantized models.

GraphTok offloads tokenization to GPU as a replayable CUDA graph, eliminating CPU bottleneck in agentic LLM serving and improving time-to-first-token.

Analysis of web-crawler budget allocation across hosts for LLM pretraining, showing quality variance primarily occurs between hosts rather than within.

SBNO uses diffusion Schrödinger bridge matching for PDE surrogate modeling with sparse, partial observations and unknown physics constraints.

JEPA-kMMVAE combines joint-embedding prediction with Markov VAE for wireless channel state information prediction from noisy temporal observations.

The escalating environmental footprint of the Information and Communication Technol- ogy (ICT) sectorwith global Data Center electric- ity usage potentially reaching 3% by 2030 and the imperative for real-time responsiveness in smart buildings demand a foundational shift towards Edge Computing. This article proposes a holis- tic architectural and algorithmic approach for cre- ating autonomous, resilient, and privacy-preserving learning systems within smart buildings, advancing the core concept of the building itself as the data center. Our contributions, demonstrating originality compared to ...

Local online training on resource-constrained platforms requires low computational and training-state overhead. We study the numerical stability and computational efficiency of half-precision proximal policy optimization (PPO) on commodity GPUs. Directly converting PPO training to FP16 is unreliable because numerical errors feed back through policy updates and subsequent data collection, while PPO repeatedly compares a current policy with a fixed behavior policy. We propose PPO-FP16 with three structure-aware modules. Global scalar protection stabilizes statistics shared across a minibatch or...

We introduce MORAD, an online reinforcement learning framework for backbone-conditioned RNA diffusion models that trains a single shared policy across diverse target backbones. Structure-conditioned RNA inverse design seeks nucleotide sequences compatible with a target three-dimensional backbone, yet existing reinforcement learning approaches either optimize a separate policy for each target backbone or rely on offline preference optimization, leaving online post-training of a reusable backbone-conditioned model largely unexplored. MORAD combines backbone-conditioned group sampling, group-rel...

DAG structures can be learned from observations by incorporating acyclicity via smooth constraints, such as DAGMA's log-determinant function. Typically, observations and the underlying DAG are related through linear structural equation models, with Gaussianity assumption on the disturbances at each node. Extending DAG learning methods that account for scale and tail behavior of disturbances that vary across nodes remains a challenge. We propose a node-varying Student-$t$ DAGMA that jointly estimates the graph structure, node-specific disturbance scales, and the degrees-of-freedom parameter. A...

Ever since people first seriously contemplated giving machines an intelligence modeled on our own, there has never been any question that they would, like us, be able to say no. The sci-fi canon is full of stories of robotic disobedience. Most of these capers are, of course, cautionary. But recently, the idea that AI shouldn’t…