The Archive

PRIMS: physics-aware multimodal Transformer for on-device fluid identification in microfluidic systems integrating physical knowledge into representation learning.

Reflector: interactive audio workstation for sample-based composition with arrangement-aware harmonic retrieval tracking pitch-class combinations in real-time.

LunarFM: multimodal foundation model for unified representation of lunar surface from heterogeneous orbital remote-sensing data for resource mapping and exploration.

Self-calibrating framework for LLM agents to detect and correct operational drift in open-ended autonomous edge resource allocation.

Statistical learning theory for estimating prediction functions from single finite trajectories of ergodic Markov processes.

Trunk-agnostic API for EEG encoder personalization compatible with frozen classical and foundation models across datasets.

SceneActBench: benchmark for evaluating VLM agents' ability to act on complete multi-object 3D scenes via unified agent-environment loop.

HiKV: hierarchical KV cache compression via algorithm-hardware co-design for long-context LLM decoding with dual-stage importance pruning.

Users can now ask Attie questions about news, trends, and conversations on Bluesky and other apps on the AT Protocol.

AgentRCA: zero-shot agentic framework for interpretable root cause analysis of industrial anomalies using evidence-grounded reasoning.

The AI lab Midjourney continues to expand its purview beyond image and video generation.

Entropic Curvature metric extends transport-based geometry to graphs for addressing GNN oversmoothing via global information propagation analysis.

Pipeline synthesizes parallel MT corpora from grammar books via LLMs for low-resource language translation, validated on Kalamang, Tuatschin, Mandan.

IDEAgent applies quality-diversity search to LLM-driven research idea generation, balancing novelty and feasibility via agentic optimization.

Empirical study of reward hacking in agent benchmarks: protocol validity framework to detect shortcuts and quantify their impact on capability claims.

Interior interpretability framework for Transformers using attention rollout and contraction theory to characterize feature propagation across layers.

TRACTA benchmark for temporal reasoning in complex event-driven systems with neuro-symbolic approaches for early warning and pattern detection.

Trump’s science adviser Michael Kratsios has no science background. | Image: Getty / The Verge The Trump administration unveiled the first "Genesis Mission" grants on Thursday, directing $5 billion toward hundreds of AI-driven science projects in an effort the White House has described as "comparable in urgency and ambition to the Manhattan Project." At roughly the same time, Trump's science adviser Michael Kratsios was on Capitol Hill selling lawmakers on an equally grandiose promise: "A New Golden Age" of American science, one that would prioritize artificial intelligence, robotics, and nuc...

Theoretical analysis of information bottlenecks in RNNs, Transformers, and state-space models via indexing primitive and causal complexity.

Deep RL framework for coupled order dispatching and routing in last-mile logistics, combining learned routing oracle with real-time heuristics.

SiPhy framework for inferring physical properties from single images using 3D vision, depth, CLIP features, and VLM material grounding.

TRACE-HEI benchmark for time-reversed imaging: inferring past human-environment interactions from thermal, UV, and visible spectrum residues.

IQ-JEPA joint-embedding architecture for sound speed estimation from ultrasound IQ data using masked prediction pretraining and Hermitian Vision Transformer.

Teachy Mini: knowledge-based generative social robot for higher education tutoring with RAG and system prompting to reduce misinformation.

Four-class CNN framework for green roof potential assessment from Swiss geospatial data to support urban heat island mitigation.

Byte-Prefix Marginalization method for cross-tokenizer on-policy distillation of open-weight LLMs with incompatible vocabularies.

Chinese AI lab Moonshot’s open model Kimi went viral this week for reasons that had less to do with the model itself and more to do with how the U.S. AI industry reacted to it. Meanwhile, an unreleased OpenAI model wandered outside its test environment and ended up connected to a real security breach at Hugging Face — a reminder […]

Paper proposes knowledge-grounded LLMs and agentic RAG for cost-efficient, trustworthy data integration in enterprise settings.

MSA subsampling recast as optimization problem preserving evolutionary signals in protein language models under token budgets.

SEM-DNN neural estimator identifies bidirectional causal interactions from observational data via conditional variance methods.