Timezone: »

NOTE: Robust Continual Test-time Adaptation Against Temporal Correlation
Taesik Gong · Jongheon Jeong · Taewon Kim · Yewon Kim · Jinwoo Shin · Sung-Ju Lee

Thu Dec 01 09:00 AM -- 11:00 AM (PST) @ Hall J #919

Test-time adaptation (TTA) is an emerging paradigm that addresses distributional shifts between training and testing phases without additional data acquisition or labeling cost; only unlabeled test data streams are used for continual model adaptation. Previous TTA schemes assume that the test samples are independent and identically distributed (i.i.d.), even though they are often temporally correlated (non-i.i.d.) in application scenarios, e.g., autonomous driving. We discover that most existing TTA methods fail dramatically under such scenarios. Motivated by this, we present a new test-time adaptation scheme that is robust against non-i.i.d. test data streams. Our novelty is mainly two-fold: (a) Instance-Aware Batch Normalization (IABN) that corrects normalization for out-of-distribution samples, and (b) Prediction-balanced Reservoir Sampling (PBRS) that simulates i.i.d. data stream from non-i.i.d. stream in a class-balanced manner. Our evaluation with various datasets, including real-world non-i.i.d. streams, demonstrates that the proposed robust TTA not only outperforms state-of-the-art TTA algorithms in the non-i.i.d. setting, but also achieves comparable performance to those algorithms under the i.i.d. assumption. Code is available at https://github.com/TaesikGong/NOTE.

Author Information

Taesik Gong (KAIST)
Jongheon Jeong (KAIST)
Taewon Kim (Korea Advanced Institute of Science & Technology)
Yewon Kim (Korea Advanced Institute of Science and Technology)
Yewon Kim

I am a second-year Master’s student in Networking & Mobile Systems Lab at Korea Advanced Institute of Science and Technology (KAIST) advised by Professor Sung-Ju Lee. By developing interactive systems and AI technologies, I explore how the state-of-the-art advances in AI can bring positive and beneficial outcomes to users and society.

Jinwoo Shin (KAIST)
Sung-Ju Lee (KAIST)

More from the Same Authors