notesum.ai

Published at October 30

Adaptive Paradigm Synergy: Can a Cross-Paradigm Objective Enhance Long-Tailed Learning?

cs.CV

cs.AI

Released Date: October 30, 2024

Authors: Haowen Xiao¹, Guanghui Liu², Xinyi Gao¹, Yang Li¹, Fengmao Lv³, Jielei Chu³

Aff.: ¹School of Information Science and Technology, Southwest Jiaotong University; ²College Of Computer Science And Technology, Harbin Engineering University; ³School of Computing and Artificial Intelligence, Southwest Jiaotong University

Arxiv: http://arxiv.org/abs/2410.22883v1

Method	KNN@1	KNN@10	MS LP	1%S LP	LT LP	Full LP
ResNet18
SimCLR	59.84	60.19	68.29	62.98	61.86	75.37
SDCLR	61.92	60.63	71.74	71.74	65.38	80.49
TS	63.09	62.91	71.86	71.29	65.03	79.08
FASSL	—	—	—	68.54	—	80.69
APS	64.06	63.70	72.78	72.60	65.46	80.35
ResNet50
SimCLR	62.51	61.49	72.83	72.17	65.78	83.75
SDCLR	62.59	61.27	73.30	72.73	67.07	85.59
TS	63.11	61.94	73.54	73.28	66.80	84.76
APS	64.17	62.83	74.16	74.27	68.15	85.46