The Archive

ACCT enables KV cache reuse across different-scale LLMs in same family, reducing context processing overhead in heterogeneous deployments.

Hyperbolic IRL incorporates geometric structure of feature spaces into inverse reinforcement learning for improved interpretability.

TradeWar benchmark evaluates LLM agent negotiation via tariff policy execution with economic outcome metrics.

Geo-FSRL: optimization method for stochastic policy geometry in few-step flow reinforcement learning models.

Analysis of discourse-level side effects from LLM activation steering, showing unintended changes beyond perplexity metrics.

MetaRefine improves image quality assessment by preserving geometry and aspect ratio while refining scores with multi-level visual features.

New trajectory-level metric (PEC, FED, TER, Recovery Rate) evaluates tool-using agents on intermediate execution validity, not just final success.

CBE ensemble method for imbalanced classification using boundary-aware sampling and complexity metrics.

Metric learning for synthetic image pre-training using fractal datasets and parameter-aware triplet loss.

SymKL unifies cross-modality signal degradation measurement using implicit neural representations and Fisher information geometry.

Structure-driven adversarial attack framework for GNNs using spectral graph properties; separates geometry from learned weights.

ProACT adaptive VR system adjusts firefighter training difficulty in real-time based on performance metrics.

You can now query AI traffic in to understand how AI bots are accessing your site.Vercel Observability AI traffic insights show: To view AI traffic insights, open in the Vercel dashboard.Observability > AI Traffic You and your agents can also query these metrics with in the . CLI queries require Observability Plus.vercel metricsVercel CLI type: embedded-entry-inline id: 6E78NBZeL598kMHZ7c4zyk Learn more in the .AI traffic documentation Read more how many requests each bot makesRequests by verified AI bot: pages most frequently requested by AI crawlers for indexing or model trainingMost-crawle...

Proposes latent rollout consistency metric to evaluate whether learned world models capture faithful environmental dynamics for agent reasoning.

Constraint-Referenced Strategy metric evaluates robot manipulation by separating progress along object joints from wasted motion.

Parametrized activation steering rescales intervention strength for consistent, interpretable control of LLM behavior at inference.

R package for semantic factor analysis using LLM embeddings to analyze psychometric scales; niche tooling for psychology research.

Reinterprets evolution strategies as center-of-mass dynamics on reward-weighted particles, offering geometric insight into ES optimization.

Analyzes failure modes in linear diffusion models learning data symmetries, quantifying initialization and finite-sample effects.

Autoencoder latent representations fail to improve bus delay detection over gradient boosting on transit GPS telemetry.

AI Factories are some of the most technologically complex operational environments ever created. Hundreds of thousands of GPUs, NVIDIA Spectrum-X Ethernet... AI Factories are some of the most technologically complex operational environments ever created. Hundreds of thousands of GPUs, NVIDIA Spectrum-X Ethernet scale-out, scale-across and scale-in fabrics, NVIDIA NVLink scale-up fabrics, distributed schedulers, and multi-week training runs operate in continuous lockstep.These layers function as one production system even when different teams own their… Source

Benchmarks generative AI for turbulence modeling, proposing validation metrics beyond visual realism for multiscale stochastic dynamics.

Proposes Spherical Brownian Diffusion for continuous language modeling on unit sphere geometry with principled forward process.

Counter-Strike 2 telemetry reveals rapid reaction-time signatures via millisecond-precision visibility reconstruction and fight-or-flight event labeling.

Differential neural tangent kernel symmetries under orthogonal groups characterized for overparameterized physics-informed networks.

Information geometry analysis of neural representation formation trajectories using Bures geometry and quantum Fisher information.

Fed-Universe proposes semantic-geometric-topological-human stack for negotiated alignment in federated learning under heterogeneity.

Tensor regression method for asymmetric/heavy-tailed distributions using modal regression; domain-specific statistics work.

Empirical study quantifying example-level data quality metrics (diversity, length, instruction structure) predictive of instruction tuning utility.

Isometric Centroid-Encoder (ICE) improves reproducibility of supervised dimensionality reduction by preserving prescribed class-centroid geometry in latent space.