The Archive

Personalized federated learning for industrial failure prediction accommodates heterogeneous degradation across clients while preserving data privacy.

Dual-track LLM-assisted stylometry (embedding + register) detects biblical allusions in Cormac McCarthy via KJV vocabulary and semantic matching.

Unsupervised method calibrates confidence estimates in reasoning LLMs using self-consistency proxy targets and distillation, enabling reliable deployment with single generation.

Large-scale study across 15 LLMs and 8 tasks reveals zero-shot ability poorly predicts optimization performance in LLM-guided evolutionary search, identifying hidden trajectory factors.

CTLF branching-time logic framework formalizes bias detection in generative AI with counting worlds semantics and probability distribution verification.

AI music startup GRAI says fans want to remix tracks, not generate songs from scratch.

CAST model captures fine-grained item semantics for sequential recommendation by distinguishing true complementary relations from spurious popularity correlations.

Visual Contrastive Editing (VCE) mitigates object hallucination in large vision-language models by counteracting language priors without inference cost.

GOLD-BEV framework learns dense bird's-eye-view semantic maps for autonomous driving using time-synchronized aerial imagery as weak supervision.

HP-Edit applies RLHF to diffusion-based image editing via human-preference post-training framework, addressing scalability gaps in preference dataset collection.

MM-JudgeBench—first multilingual multimodal evaluation benchmark with 60K+ preference instances across 25 languages—reveals LVLM judge generalization failures.

M²GRPO integrates Mamba state-space policy with group-relative policy optimization for multi-agent cooperative pursuit in biomimetic underwater robots.

Vague complaint about Anthropic with no substantive details provided.

Study shows new entity embeddings cause interference with prior knowledge in continual KG embedding, extending understanding of catastrophic forgetting.

Analyzes routing optimization for federated learning over dynamic satellite networks, addressing in-orbit relay-based communication.

GRASPrune jointly prunes FFN channels and KV heads in LLMs under budget constraints via gated structured pruning.

User praises Claude Design's typography and font selection quality in graphic design task execution.

Evaluates self-consistency impact on encyclopedic knowledge recall vs. symbolic reasoning using MMLU knowledge-recall splits.

DeFineMed models bridge performance gap between 7B and 24B LLMs via continual pre-training on German medical corpus FineMed-de.

EXIT multimodal transformer predicts metal-organic framework properties accounting for sample-dependent structural variations.

OpenAI ships ChatGPT Images 2.0 with improved text rendering, multilingual support, and visual reasoning capabilities.

Compares centralized vs. decentralized ML architectures for energy efficiency in 6G IoT networks via railway testbed.

OpenAI teases GPT-Image-2 model with livestream announcement at 12pm PT.

TACENR provides task-agnostic contrastive explanations for node representations in graph neural networks.

EEG-based framework predicts driver intentions in real-world on-road driving using deep learning across 12 architectures.

FairTree audits ML model fairness across continuous and categorical subgroups via bias-variance decomposition.

LASER closed-loop active sensing framework reconstructs continuum physical fields using latent world models and POMDPs.

DeepRed benchmark evaluates LLM agents on realistic Capture The Flag challenges with partial-credit scoring beyond binary outcomes.

DASH-KV accelerates long-context LLM inference via asymmetric KV cache hashing to reduce attention complexity.

Memristive-friendly echo state networks for time series classification using reservoir computing with analog hardware.