List your company
ResearchFormerly WizWand

Active Learning Image Classification on F-MNIST (test)

99.56AccuracyTD-FT (Ours)
84.147288.148692.1596.1514Oct 29, 2025
Updated 8mo ago

Evaluation results

MethodLinks
TD-FT (Ours)Acquisition Strategy=Power batch acquisition, Representation Learning Method=Task-Driven Fine-Tuning, Encoder Backbone=ResNet182025.1099.56
US+EPIG (SimCLRv2)Acquisition Strategy=EPIG, Representation Learning Method=SimCLRv2, Encoder Backbone=ResNet182025.1098.53
TD-SPLIT (Ours)Acquisition Strategy=Power batch acquisition, Representation Learning Method=Task-Driven SPLIT, Encoder Backbone=ResNet182025.1098.46
TD-FT RandomAcquisition Strategy=Random, Representation Learning Method=Task-Driven Fine-Tuning, Encoder Backbone=ResNet182025.1096.23
US+EPIG (VAE)Acquisition Strategy=EPIG, Representation Learning Method=VAE, Encoder Backbone=Burgess encoder2025.1094.5
Cluster MarginAcquisition Strategy=Cluster Margin, Representation Learning Method=Original Baseline, Encoder Backbone=ResNet182025.1094.24
SIMILARAcquisition Strategy=SIMILAR, Representation Learning Method=Original Baseline, Encoder Backbone=ResNet182025.1093.82
US Random (SimCLRv2)Acquisition Strategy=Random, Representation Learning Method=SimCLRv2, Encoder Backbone=ResNet182025.1092.76
TD-SPLIT RandomAcquisition Strategy=Random, Representation Learning Method=Task-Driven SPLIT, Encoder Backbone=ResNet182025.1088.19
US Random (VAE)Acquisition Strategy=Random, Representation Learning Method=VAE, Encoder Backbone=Burgess encoder2025.1086.5
GALAXYAcquisition Strategy=GALAXY, Representation Learning Method=Original Baseline, Encoder Backbone=ResNet182025.1084.74