Hidden Trajectories Don't Lie: Correcting Corruption Drift for Test-Time Adaptation
We introduce Drift-TTA, which adds zero-initialized low-rank adapters to a frozen backbone and trains them to align corrupted hidden trajectories with those from cleaner inputs.
ProofPaper ↗
Key points
- Test-time adaptation (TTA) adapts pretrained models to distribution shifts encountered during deployment.
- We study hidden trajectories: the ordered sequence of intermediate representations produced across a deep neural network.
- We develop an online TTA approach that freezes the backbone and adapters while updating a scalar gate to control adaptation strength.
- We further cast trajectory alignment as MAP inference under a Gaussian observation model and derive a KL-based transfer bound characterizing test-time trajectory misalignment in terms of offline residual error, residual dimensionality, and distribution shift between offline and test corruptions.
Sources (1)
- [1]Hidden Trajectories Don't Lie: Correcting Corruption Drift for Test-Time AdaptationarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 16, 08:43 PM
We introduce Drift-TTA, which adds zero-initialized low-rank adapters to a frozen backbone and trains them to align corrupted hidden trajectories with those from cleaner inputs.
Test-time adaptation (TTA) adapts pretrained models to distribution shifts encountered during deployment.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 16, 2026FedRAPP: Federated Test-Time Prompt Tuning for Vision-Language Models with Reliability-Aware Prompt Personalization
- Oct 16, 2026PONDER: Learning to Generate in the Right Order
- Oct 16, 2026Data-Efficient Classifier Pretraining Through Diffusion Models
- Oct 15, 2026The Dormant Spiking Neuron: A State-Driven Mechanism for Efficient Spiking Neural Networks
- Oct 15, 2026Where, How, and When: Query-Guided Grounding for Video Understanding
- Oct 15, 2026Clean Data Can Still Carry Backdoors: Support-Persistent Backdoors for Model Reuse