RoboWM-Bench: Benchmark for Robotic Manipulation Models
RoboWM-Bench offers a unified benchmark to evaluate video world models in robotic manipulation, focusing on physical plausibility and executable actions.
ProjLens: Enhancing Safety in Multimodal Models
Discover how ProjLens improves multimodal model safety by exposing backdoor vulnerabilities and boosting robustness in large language models.
Unified ASR Transducer: Closing Offline-Streaming Gap
Discover how consistency regularization unifies ASR transducers, improving accuracy and reducing latency in both offline and streaming modes.
S2MAM: Robust Semi-Supervised Model for Variable Selection
Discover S2MAM, a semi-supervised meta additive model enhancing robust estimation and variable selection in noisy data environments.
Product-of-Experts Training Cuts Dataset Bias in NLI
Discover how Product-of-Experts training reduces dataset artifacts and improves accuracy in Natural Language Inference models for better reasoning.