Search the full wire by company, model, lab, or keyword. Every story we have ever aggregated.
PRISM multimodal LM for clinical prognosis integrates structured measurements, ECG waveforms, and population-retrieved embeddings via dual-modality memory.
FedPCD proposes priority-constrained federated optimization to improve fairness across heterogeneous clients without symmetric weighting.
Explanatory World Models use LLM-based attribution to improve credit assignment for distant actions in long-horizon recommendation RL.
Multi-agent session alternation framework mitigates long-horizon deceleration and harness staleness in LLM-based ML engineering agents.
POrchestra enables dynamic, proactive agentic orchestration with adaptive sub-agent goals and reinforced policy learning from sparse feedback.
BranchOPD extends on-policy distillation by resampling student trajectories from decision branches to increase supervision coverage for math/coding.
LISCO learns primal-dual solver steps via neural networks to reduce real-time cost of nonlinear MPC for constrained optimization.
Local agentic framework generates edge case scenarios autonomously for autonomous system safety assurance without heavy human-in-the-loop.
Harmonic analysis of Sobolev functions on quantum spaces via Peter-Weyl decompositions and quantum circuit approximation bounds.
Representational similarity metrics (RSA, CKA, Procrustes) are limited by input selection and grouping choices made before measurement; study across vowel acoustics in 27 representations.
MLLMs fail to recognize coarse text in images (SquintBench: 1076 moderation images); frozen-weight analysis shows glyph identity weakly represented in vision encoders.
SqueezeBench prevents saturation in math benchmarks via open-ended integer functions with machine-checked Lean proofs; raising the bar as systems improve.
Evidence-driven workflow for transferring knowledge from simulation and operational data back into cyber-physical system design-time models.
Book/survey on agentic LLMs: how language models act in real world, interact with each other, and influence decisions through direct intervention.
Oído: 13M-parameter Conformer-CTC speech recognition on ESP32-S3 microcontroller ($5, 512KB SRAM) via int8 quantization and SIMD optimization.
Compression-based risk bounds for learning: prove logarithmic factor is removable, achieving law-of-large-numbers rate O(k/N) without log penalty.
DistilHuBERT accent classification via speaker-disjoint splits and cross-corpus generalization reveals large gap between in-corpus accuracy and real-world robustness.
Multi-agent RL for autonomous vehicle routing: marginal cost reward internalizes externalities on human drivers, enabling socially-aware coexistence in mixed traffic.
Rollout-Time Intervention (RTI): RL technique to elicit LLM agent trajectories beyond policy capability by overcoming on-policy exploration bottleneck.
Physics-domain anomaly detection using learned representations; interpretability technique for LHC collision events.
PPG-Bench unifies evaluation of five photoplethysmography foundation models across nine datasets and 26 cardiovascular tasks.
CNN-Transformer with sigmoid label function for direction-of-arrival estimation; signal processing application.
Study of continual learning regularization (EWC, mini-KFAC) on Whisper-small for financial speech ASR; finds adaptation improves rather than degrades.
Tutorial survey unifying time series causal discovery methods across signal processing, statistics, dynamical systems, and ML.
Web$hell unified interface for LLM agents to access open web via search, browser, and filesystem for long-horizon task completion.
In-context learning for system identification; meta-model maps training data and queries to outputs zero-shot, bypassing explicit model estimation.
Persistent Candidate LinUCB reduces computational cost in linear bandits with large action sets via candidate set maintenance.
ConvLSTM forecasts land degradation in Ceará, Brazil using satellite Earth Observation data (2000–2023); measures predictability limits.
Analysis: linear models generalize better than flexible nonlinear models in complex domains due to environment shift in nonlinear components.
RegulaTutor applies self-regulated learning theory to personalize long-term LLM-based tutoring by linking learner history to explicit pedagogical strategies.