List your company
ResearchFormerly WizWand

Classification on CIFAR-100 (Accuracy)

92.5AccuracyCBM-Suite
21.280839.770458.2676.7496May 9, 2026May 17, 2026May 26, 2026Jun 4, 2026Jun 13, 2026Jun 22, 2026Jul 1, 2026
Updated 4mo ago

Evaluation results

MethodLinks
CBM-SuiteBottleneck type=CBMs with soft concepts, Backbone=ViT2026.0792.5
HybridCBMBottleneck type=CBMs with soft concepts, Backbone=ViT2026.0786.22
CaBMBottleneck type=Strictly text bottleneck, Protocol=text-only classifier2026.0781.14
Linear ProbingLoss=SCL, Passes=T × N, Backbone=ResNet182026.0573.9
Fixed PrototypesLoss=SCL, Passes=N, Backbone=ResNet182026.0573.9
Normalized Linear ProbingLoss=SCL, Passes=T × N, Backbone=ResNet182026.0573.6
PS-CBMBottleneck type=CBMs with soft concepts, Backbone=CNN2026.0772.1
LF-CBMBottleneck type=CBMs with soft concepts, Backbone=CNN2026.0768.8
VLG-CBMBottleneck type=CBMs with soft concepts, Backbone=CNN2026.0768.3
LaBoBottleneck type=CBMs with soft concepts, Backbone=CNN2026.0768.1
FLQMIAAlignment=frozen pretrained DINOv2 and GTE-1.5, Setting=Zero-shot2026.0553.31
PCBMBottleneck type=CBMs with soft concepts, Backbone=CNN2026.0752
FLVMIAAlignment=frozen pretrained DINOv2 and GTE-1.5, Setting=Zero-shot2026.0551.69
SAIL w/ SigLIPAlignment=frozen pretrained DINOv2 and GTE-1.5, Loss=SigLIP, Setting=Zero-shot2026.0550.5
SAIL w/ CLIPAlignment=frozen pretrained DINOv2 and GTE-1.5, Loss=InfoNCE, Setting=Zero-shot2026.0545.22
CSAAlignment=frozen pretrained DINOv2 and GTE-1.5, Setting=Zero-shot2026.0544.18
Multilinear-CovJac2026.0528.91
M-CovJac#p/n=4, L=6, k=256k, iterations=100k, τ=1.02026.0528.37
Soft-Mix2026.0528.21
Gumbel-ST2026.0528.08
Soft-Mix#p/n=16, L=6, k=256k, iterations=100k, τ=1.02026.0527.92
Gumbel-ST#p/n=16, L=6, k=256k, iterations=100k, τ=1.02026.0527.72
Multilinear-STE2026.0525.02
M-STE#p/n=4, L=6, k=256k, iterations=100k, τ=1.02026.0524.02