Test-Time Classifier Adjustment Module for Model-Agnostic Domain Generalization

By Yusuke Iwasawa, Yutaka Matsuo. Neural Information Processing Systems, 2021

424 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Continual Test-Time Domain Adaptation (2022, methodology): “[22] only updates the final classification layer during inference time using pseudo-prototypes.”
Efficient Test-Time Model Adaptation without Forgetting (2022, background): “…adaptation (Nado et al., 2020; Schneider et al., 2020; Khurana et al., 2021), test-time entropy minimization (Wang et al., 2021; Fleuret et al., 2021), prediction consistency maximization over different augmentations (Zhang et al., 2021b), and classiﬁer adjustment (Iwasawa & Matsuo, 2021).”
Generalizing to Unseen Domains: A Survey on Domain Generalization (2021, background): “Very few recent works [266, 267] paid attention to this setting.”
A Comprehensive Survey on Test-Time Adaptation Under Distribution Shifts (2023, unknown): “Besides, T3A [15] proposes merely adjusting the classifier layer by computing class prototypes using online unlabeled data and classifying each unlabeled sample based on its distance to these prototypes.”
MEMO: Test Time Robustness via Adaptation and Augmentation (2021, methodology): “Recently, several works have proposed methods for improving accuracy via adaptation after seeing the test data, typically by updating a subset of the model’s weights [44, 45, 18], normalization statistics [40], or both [46, 52].”