Kang Suhyun ; Park Jungwon ; Lee Wonseok ; Rhee Wonjong

Abstract

Cross-Domain Few-Shot Learning~(CDFSL) methods typically parameterize modelswith task-agnostic and task-specific parameters. To adapt task-specificparameters, recent approaches have utilized fixed optimization strategies,despite their potential sub-optimality across varying domains or target tasks.To address this issue, we propose a novel adaptation mechanism calledTask-Specific Preconditioned gradient descent~(TSP). Our method firstmeta-learns Domain-Specific Preconditioners~(DSPs) that capture thecharacteristics of each meta-training domain, which are then linearly combinedusing task-coefficients to form the Task-Specific Preconditioner. Thepreconditioner is applied to gradient descent, making the optimization adaptiveto the target task. We constrain our preconditioners to be positive definite,guiding the preconditioned gradient toward the direction of steepest descent.Empirical evaluations on the Meta-Dataset show that TSP achievesstate-of-the-art performance across diverse experimental scenarios.