The Archive

ER-JEPA applies hierarchical self-supervised learning to ECG time-series analysis with limited labeled data.

IMPFM framework for sample-efficient online feedback-driven search via multi-particle flow-maps in high-dimensional spaces.

Agent Skill Supply Chains framework applies software dependency management to LLM agent skill composition and provenance tracking.

DiscoPER framework enables open-ended autonomous scientific discovery via iterative meta-reflection and cross-finding synthesis in LLM agents.

GAIA operator learning model extends geometry-adaptive neural operators to inverse problems and boundary value problems with domain mismatches.

Log_bQuant logarithmic quantization method improves 4-bit LLM compression by adapting to non-uniform weight distributions.

ZO-Act enables zeroth-order fine-tuning of LLMs via activation-informed low-rank subspace optimization without backpropagation.

Paper proposes Muon optimizer acts as implicit residual connection during training, improving representation preservation via orthogonalization trade-offs.

University RAG-based multimodal chatbot prototype for institutional Q&A, combining LLMs with semantic retrieval for domain-specific policy queries.

FAR framework enables robots to learn from test-time failures via preference learning, improving autonomous task completion without human intervention.

Google hosts NYC education summit with industry leaders to discuss AI in classrooms; no new models, tools, or technical announcements reported.

SynLaD latent diffusion framework generates synthesizable drug molecules by unifying 3D pharmacophore design with synthetic accessibility constraints.

CausalMix formulates LLM data mixture optimization as causal inference to handle non-static distributions, enabling scaling without costly retraining.

MedQADE benchmark reveals LLM evaluators (Gemini 3 Flash) match clinician agreement on German medical QA but lack clinical caution in safety assessment.

GRINCO framework reduces active learning labeling costs by selecting samples in quotient space, exploiting data symmetries to avoid redundant queries.

Case study on software engineering with frontier AI coding agents reveals shift from implementation scarcity to governance, inspection, and maintainability challenges.

LongVQUBench introduces 1200+ long-duration videos to benchmark vision-language models on temporal video quality understanding beyond short-clip tasks.

OpenAgent formalizes generalization gaps in LLM tool-use agents across query, action, observation, and domain shifts via controlled sandbox evaluation.

Theoretical analysis of staleness effects in asynchronous GRPO-based RLHF reveals per-step surrogacy gap and learning-rate scaling laws.

AlphaEarth embeddings as spatial context improve spatio-temporal point-process forecasting for EMS across sparse regions.

Gaussian process bandit optimization with quantum kernels balances expressivity and learnability for NISQ tasks via kernel projection.

Message Passing Language Models enable efficient multi-threaded LLM reasoning by enabling inter-thread communication, reducing chain-of-thought bottlenecks.

MemSyco-Bench benchmarks agent sycophancy in memory retrieval, measuring how retrieved memories bias reasoning and decision-making.

GSRQ applies gain-shape residual quantization to push KV cache to sub-1-bit regime, addressing centroid shrinkage in high-dimensional codebook learning.

Multi-agent LLM framework automatically generates and verifies reaction rules across 665k patent reactions for chemistry synthesis planning.

Theoretical characterization of separable graphical models with mixed edge types for encoding independence structures.

LLM agent collectives with persistent memory and tools exhibit emergent complexity while remaining interpretable substrates.

Test-time control framework DART-VLN mitigates memory decay and loop inefficiencies in vision-language navigation agents.

EchoRisk: multicentre longitudinal echocardiography dataset with cardiotoxicity labels for automated breast cancer treatment risk stratification.

Study shows moderate LLM agent personality expression outperforms extremes on trust and goal-adoption in conversational behavior-change tasks.