Controller guides when to update AI models at test time
When Should We Intervene at Test Time?
Machine Learning
Summary
AI models sometimes need help adjusting when the environment changes, but updating them too often or at the wrong time can hurt performance. The paper presents PORTAL, a system that decides when to adjust the model using limited feedback, like labels, to avoid harmful interventions. PORTAL learns from the feedback to choose the best action and requests new information only when it can improve accuracy. This approach keeps the model reliable over time and helps it adapt efficiently without constant human input.
What this means in practice
- •For machine learning engineers: Ensure AI models adapt safely in production by selectively updating only when beneficial, reducing errors caused by harmful test-time changes.
- •For autonomous vehicle developers: Manage real-time AI adjustments in self-driving cars to improve perception accuracy during changing environmental conditions without constant human labeling.
Authors
Jiahao Zhang, Xinling Wen, Liangliang Li, Yu Chen
Abstract
Test-time adaptation helps a deployed model respond to changing inputs, but the same intervention can rescue one batch and damage the next. Updating normalization statistics, for example, can improve corrupted images yet fail when a batch contains mostly one class. Average accuracy hides these failures and cannot tell a model when to use an intervention. Target labels can guide this decision, but they are scarce and some choices are already clear. We introduce PORTAL, a controller that chooses when to deploy an intervention, which inputs to label, and when to stop querying. Each available prediction rule is a candidate, and its predictions are fixed before current feedback arrives. Every acquired label therefore evaluates all candidates on the same input. PORTAL combines these shared comparisons and directs queries toward candidates that can still overtake the current leader. It stops when the remaining labels cannot change the choice or when the possible accuracy gain no longer justifies another query. Our analysis links the required feedback to accuracy gaps and shared mistakes, explaining why some choices need much less evidence. Experiments show that this control turns an intervention that loses accuracy overall into a useful choice on selected batches. It prevents severe failures caused by changes in arrival order and preserves gains across complete corruption streams. When a fixed rule already works well, PORTAL uses little feedback to retain its performance. Periodic label checks reveal when past evidence needs refreshing and help the controller recover as conditions change. The acquired labels also train stronger candidates for future batches, and selection brings nearly all of their accuracy gains into deployment. The same feedback therefore improves both the choice made now and the predictions available to later batches.