AI Paper Atlas

Research map / Label-efficient and robust learning / Test-time adaptation

Revisiting Realistic Test-Time Training: Sequential Inference and Adaptation by Anchored Clustering Regularized Self-Training

Yongyi Su, Xun Xu, Tianrui Li, Kui Jia

TPAMI 2024 · Label-efficient and robust learning › Test-time adaptation · approach: Realistic and stable test-time adaptation · 21 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add self-training (ablation)TTAC++ · CIFAR10-C (N-O-SL)error rate (%): 10.94 -> 9.78Improved
add anchored clustering (ablation)TTAC++ · CIFAR10-C (N-O-SL)error rate (%): 29.15 -> 14.32Improved
add pseudo label filtering (ablation)TTAC++ · CIFAR10-C (N-O-SL)error rate (%): 29.15 -> 15.00Improved
combine anchored clustering and pseudo label filtering (ablation)TTAC++ · CIFAR10-C (N-O-SL)error rate (%): 14.32 -> 11.33Improved
add global feature alignment with KLD (ablation)TTAC++ · CIFAR10-C (N-O-SL)error rate (%): 11.33 -> 10.94Improved
use KLD instead of L2 distance for global alignment (ablation)TTAC++ · CIFAR10-C (N-M-SL)error rate (%): 11.87 -> 10.8Improved
add self-training (ablation)TTAC++ · CIFAR10-C (N-M-SL)error rate (%): 9.42 -> 7.23Improved
add contrastive branch to source training (ablation)TTAC++ · CIFAR10-C (Y-O-SL)error rate (%): 9.78 -> 10.05Improved
use filtering instead of no filtering for cluster updates (ablation)TTAC++ · CIFAR10-Cerror rate (%): 10.58 -> 9.78Improved
use filtering instead of soft assignment (ablation)TTAC++ · CIFAR10-Cerror rate (%): 10.35 -> 9.78Improved
use sample queue of 4096 instead of single mini-batch (ablation)TTAC++ · CIFAR10-Cerror rate (%): 11.18 -> 10.31Improved
increase model update epochs from 1 to 4 with queue (ablation)TTAC++ · CIFAR10-Cerror rate (%): 10.31 -> 8.82Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy