Get Started
Home
Topics
Search
Library
Research questionHow does structured model compression change a model’s ability to adapt at test time under distribution shift?A compressed model may retain high accuracy under supervised adaptation yet adapt poorly at test time. Reduced representational diversity and structural constraints can limit recovery from shifted inputs, with effects that vary across compression strategies.
AI
Computer Vision
Inference Optimization
Machine Learning
Small / On-device Models
Latest papersRecent research connected to this question, newest first.On the Interaction Between Model Compression and Test-Time AdaptationThe evidence concerns ResNet-18 and ViT-Base models evaluated on CIFAR-10-C and ImageNet-C with multiple structured compression methods and standard test-time adaptation techniques. The analysis examines representational expressivity and adaptation-subspace compatibility, so its conclusions are grounded in these image-model and corruption-benchmark settings.research paper · Sep 3, 2026
Related questions
How can spectral compressive imaging models adapt to unseen optical configurations without heavy computation?How can dominant-channel activations enable lightweight adaptation of frozen DiT-based super-resolution models?How can image-to-shape diffusion transformers be compressed for resource-constrained 3D generation without losing geometric fidelity?Can text-to-image models match web-scale performance using smaller, reproducible datasets and models?