Research map / Label-efficient and robust learning / Test-time adaptation
Synchronizing Task Behavior: Aligning Multiple Tasks during Test-Time Training
Wooseong Jeong, Jegyeong Cho, Youngho Yoon, Kuk-Jin Yoon
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add Task Behavior Synchronizer (TBS) (ablation) | S4T · Taskonomy to NYUD-v2 | Delta_TTT: +0.00 -> +26.16 | Improved |
| add Task Behavior Synchronizer (TBS) (ablation) | S4T · Taskonomy to PASCAL-Context | Delta_TTT: +0.00 -> +28.32 | Improved |
| add task-specific projection to TBS (ablation) | S4T · Taskonomy to NYUD-v2 | Delta_TTT: +26.16 -> +33.79 | Improved |
| add task-specific projection to TBS (ablation) | S4T · Taskonomy to PASCAL-Context | Delta_TTT: +28.32 -> +32.27 | Improved |
| add feature masking (ablation) | S4T · Taskonomy to NYUD-v2 | Delta_TTT: +33.79 -> +34.94 | Improved |
| add feature masking (ablation) | S4T · Taskonomy to PASCAL-Context | Delta_TTT: +32.27 -> +34.20 | Improved |
| use image reconstruction as auxiliary task instead of task labels (ablation) | S4T · Taskonomy to NYUD-v2 | Delta_TTT: +34.94 -> +26.76 | Hurt |
| use same masking across all tasks instead of random masking (ablation) | S4T · Taskonomy to NYUD-v2 | Delta_TTT: +32.76 -> +34.94 | Improved |
| use same masking across all tasks instead of non-overlapping masking (ablation) | S4T · Taskonomy to NYUD-v2 | Delta_TTT: +32.56 -> +34.94 | Improved |
| use same masking across all tasks instead of hiding specific tasks (ablation) | S4T · Taskonomy to NYUD-v2 | Delta_TTT: +33.06 -> +34.94 | Improved |
| raise masking ratio to about 0.7-0.8 (ablation) | S4T · Taskonomy to NYUD-v2 | Delta_TTT | Improved |
| apply S4T when adapting from NYUD-v2 to Taskonomy | S4T · NYUD-v2 to Taskonomy | Delta_TTT: +2.13 | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 5 of 5 on most shared results.
- ActMAD — ActMAD: Activation Matching To Align Distributions for Test-Time-Training (CVPR 2023) · better on 3 of 4 shared results
- NC-TTT — NC-TTT: A Noise Constrastive Approach for Test-Time Training (CVPR 2024) · better on 3 of 3 shared results
- TIPI — TIPI: Test Time Adaptation With Transformation Invariance (CVPR 2023) · better on 3 of 3 shared results
- ClusT3 — ClusT3: Information Invariant Test-Time Training (ICCV 2023) · better on 3 of 3 shared results
- TENT — Tent: Fully Test-Time Adaptation by Entropy Minimization (ICLR 2021) · better on 3 of 3 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
- In-Place Test-Time Training (ICLR 2026 · similarity 0.85)
- Test-Time Training Provably Improves Transformers as In-context Learners (ICML 2025 · similarity 0.85)
- Compute Aligned Training: Optimizing for Test Time Inference (NeurIPS 2026 · similarity 0.85)
- NTKMTL: Mitigating Task Imbalance in Multi-Task Learning from Neural Tangent Kernel Perspective (NeurIPS 2025 · similarity 0.84)
- Learning What to Remember: Test-Time Training via Context Distillation (NeurIPS 2026 · similarity 0.84)
- DPCore: Dynamic Prompt Coreset for Continual Test-Time Adaptation (ICML 2025 · similarity 0.84)
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.
