notesum.ai

Published at November 6

Beyond Model Adaptation at Test Time: A Survey

cs.LG

cs.AI

Released Date: November 6, 2024

Authors: Zehao Xiao¹, Cees G. M. Snoek¹

Aff.: ¹Unknown

Arxiv: http://arxiv.org/abs/2411.03687v1

Refer to caption

	Preparation-agnostic	Training preparation	Training & Data preparation
Model adaptation
Auxiliary self-supervision	-	[202, 81, 180, 241, 45, 293]	[4, 19, 40, 43, 103, 388]
Entropy minimization	[319, 389, 296, 375, 237, 173]	-	-
Pseudo-labeling	[264, 324, 34, 134, 95]	-	[7]
Feature alignment	[325, 285, 141, 76, 222]	-	-
Inference adaptation
Batch-wise inference	[131, 124, 379]	[63]	-
Sample-wise inference	[377, 147, 369]	[355, 151, 326]	[341, 294]
Normalization adaptation
Target statistics	[90, 142, 179, 228]	-	[376]
Statistics combination	[271, 148, 359, 123, 349]	[186, 401]	-
Statistics inference	-	-	[61, 343]
Sample adaptation
Feature adjustment	-	[243, 233, 127]	[340, 248]
Input adjustment	-	[83, 239, 301]	-
Prompt adaptation
Text-prompting	[262, 261, 66, 378]	-	-
Embedding-prompting	[279, 358, 86, 211, 385]	-	[338]