The Archive

Mixing-time analysis of weighted Dikin walks for sampling from polytopes and PSD cones under self-concordance.

Sharp asymptotic characterization of kernel ridge regression under power-law anisotropic Gaussian data in high dimensions.

Information-theoretic bounds on LLM meaning recovery from text alone; formal proof that textual form cannot encode full semantic content.

Comprehensive survey of neural network optimizers in 2025–2026 covering temporal estimation, geometry, scheduling, and distributed systems.

Logos: cross-process agent bus architecture separating component failure domains from process lifecycle via spatiotemporal composability.

Expanded Relief-based feature selection algorithms optimized for interaction detection in high-dimensional biomedical data.

Video diffusion models repurposed for joint geometry learning (depth, surface normals) without task-specific fine-tuning overhead.

DARTS addresses representation bias in decoder-based LLM merging via lightweight correction modules, extending prior vision-model work to autoregressive architectures.

Theoretical analysis of certified code world models and their knowledge bounds under sampling-gate constraints using topological gauge-choice formalism.

InstructMesh enables interactive refinement of generative 3D models for fabrication via natural language and slider-based selective repair operations.

Open AI models are evolving faster than ever, but bringing them into native applications can still require model-specific conversion, preprocessing,... Open AI models are evolving faster than ever, but bringing them into native applications can still require model-specific conversion, preprocessing, post-processing, and runtime code. NVIDIA TensorRT Model Connect open collection of reference implementations helps to address this challenge. TensorRT Model Connect shows you how to run supported models with NVIDIA TensorRT in native C++… Source

DWT_AlexNet_DNN hybrid framework combines discrete wavelet transform and deep learning for improved texture image classification across industrial and medical domains.

Stratechery weekly digest covering breaker advantage, HDMI standards battles, and data center discourse—no specific AI announcements.

Study maps ASR error propagation in embodied AI safety, showing how speech recognition failures reduce EAI model refusal behavior via SafeAgentBench/POEX benchmarks.

Split conformal framework provides pointwise uncertainty quantification guarantees for neural operator surrogates via calibrated residual bands and spatial quantiles.

CE-MoE reduces training communication overhead in mixture-of-experts LLMs via heterogeneous layer patterns that concentrate expert capacity while maintaining depth.

PCMI and WACS metrics enable reference-free forced alignment evaluation for speech using self-supervised representations, supporting multilingual analysis.

Replicant is a deep RL framework learning realistic black-box evasion policies against ML-based malware detectors under strict label-only threat model.

Just as new data centers face growing backlash from neighboring communities, the US Environmental Protection Agency (EPA) is about to make it harder for people to weigh in on any pollution those centers create. The EPA plans to toss out a federal rule requiring public notice and an opportunity to comment when certain industrial sites apply for an air permit. Advocates warn that the move would allow data center developers and other industries to break ground without giving nearby residents a say, or even a warning that construction is planned. "You can see why that would be so interesting to a...

Empirical study of Claude Code plugin marketplace structure and maintenance dynamics, examining co-evolution of NL instructions, scripts, and configs vs. traditional packages.

Sequential test-time scaling outperforms parallel sampling for LLM machine translation, offering higher ceilings and sample efficiency on smaller budgets.

AcrossVAM1.0 factorizes robot video prediction into object-centric motion and appearance using text-assisted 0.28M-parameter Transformer on frozen SAM3-DLP codec.

Systematic survey of LLM-based agents for software and systems security, covering design patterns, applications, and evaluation methods for autonomous security workflows.

Proper scoring rule choice significantly affects LLM forecaster calibration and bias profiles; Brier score training yields best calibration despite theoretical equivalence.

NL2AGBench evaluates LLM auto-formalization of natural-language geometry problems into AlphaGeometry's domain-specific language for theorem proving.

ElephantBench probes whether LLMs retain divergent long-tail facts via 1,094 multi-account QA questions from low-exposure web corpus.

ContextPilot uses fine-grained RL to teach agents proactive context management with global planning and adaptive compression for long-horizon tasks.

COVER proposes evaluation contract framework for multi-agent coalition routing to isolate team-composition effects from accuracy gaps.

Real-time virtual circuits for tokamak plasma control on MAST Upgrade using neural network surrogates instead of offline lookup tables.

Study factorizes persona-based dialogue generation across training/inference/evaluation visibility of speaker biographies to isolate interlocutor effects.