ActMAD: Activation Matching to Align Distributions for Test-Time-Training
This addresses distribution shifts in test-time training for machine learning models, offering a domain-specific but incremental improvement over existing methods.
The paper tackled the problem of adapting models to out-of-distribution data at test-time by aligning activation statistics across multiple network layers, achieving state-of-the-art performance on benchmarks like CIFAR-100C and Imagenet-C and a 15.4% improvement in object detection on KITTI-Fog.
Test-Time-Training (TTT) is an approach to cope with out-of-distribution (OOD) data by adapting a trained model to distribution shifts occurring at test-time. We propose to perform this adaptation via Activation Matching (ActMAD): We analyze activations of the model and align activation statistics of the OOD test data to those of the training data. In contrast to existing methods, which model the distribution of entire channels in the ultimate layer of the feature extractor, we model the distribution of each feature in multiple layers across the network. This results in a more fine-grained supervision and makes ActMAD attain state of the art performance on CIFAR-100C and Imagenet-C. ActMAD is also architecture- and task-agnostic, which lets us go beyond image classification, and score 15.4% improvement over previous approaches when evaluating a KITTI-trained object detector on KITTI-Fog. Our experiments highlight that ActMAD can be applied to online adaptation in realistic scenarios, requiring little data to attain its full performance.