Test Time Adaptation
Test-time adaptation (TTA) focuses on improving the performance of pre-trained machine learning models on unseen data during inference, without requiring additional labeled training data. Current research emphasizes developing robust TTA methods across diverse tasks, including image classification, segmentation, object detection, and speech recognition, often employing techniques like batch normalization updates, pseudo-labeling, and adversarial training within various model architectures (e.g., transformers, neural implicit representations). The ability to adapt models efficiently at test time is crucial for deploying machine learning systems in real-world scenarios characterized by domain shifts and data variability, impacting fields ranging from medical imaging to robotics.
Papers - Page 5
ESCAPE: Energy-based Selective Adaptive Correction for Out-of-distribution 3D Human Pose Estimation
Luke Bidulka, Mohsen Gholami, Jiannan Zheng, Martin J. McKeown, Z. Jane WangRealistic Evaluation of Test-Time Adaptation Algorithms: Unsupervised Hyperparameter Selection
Sebastian Cygert, Damian Sójka, Tomasz Trzciński, Bartłomiej Twardowski