Tag: distribution shift

A physics-guided digital-twin framework estimates electric vehicle battery energy minute by minute and is stress-tested across route, operating-regime, vehicle, and ...

Researchers have developed AdaTent++, a test-time adaptation framework that routes each unlabeled frame through reliability-gated paths to improve traffic sign ...

Researchers at Tsinghua University and collaborators developed Occupancy-Matching Policy Optimization, a reinforcement learning method that aligns historical robot experience with ...

Researchers show that conformal prediction set sizes and miscoverage changes can distinguish label noise from covariate shift in deployed classifiers ...

A Spanish team ran the same quantum kernel benchmark on Qiskit, Cirq and PennyLane under simulated drought, wet and extreme-heat ...

Researchers at Hanyang University have developed TACCO, a contrastive learning method that explicitly identifies and suppresses out-of-distribution actions to stabilize ...

Researchers have developed a feedback-guided framework that adapts decision thresholds instead of retraining classifiers, achieving near-equal accuracy with lower false ...

A new survey of 161 studies maps how out-of-distribution data undermines AI-based mechanical fault diagnosis and argues that generalization, calibrated ...

A new theoretical review argues that reward hacking, benchmark failure, and distribution shift share a single structural root: observation regimes ...

A new AI retrosynthesis system called RetroChimera combines models with complementary inductive biases to outperform leading baselines and earn the ...