Random Neurons Across Layers 1 & 2 of ResNet-50 with ImageNet-Val + Broden Probing Dataset
Neuron examples:
We have color-coded the neuron descriptions by whether we believed they were accurate, somewhat correct, or vague/imprecise.
Describe-and-Dissect: Interpreting Neurons in Vision Networks with Language Models
Model: ResNet-50
Probing Data: ImageNet-Val + Broden
Random Neurons Across Layers 3 & 4 of ResNet-50 with ImageNet-Val + Broden Probing Dataset
Neuron examples:
We have color-coded the neuron descriptions by whether we believed they were accurate, somewhat correct, or vague/imprecise.
Model: ResNet-18
Probing Data: ImageNet-Val + Broden
Random Neurons Across Layers 1 & 2 of ResNet-18 with ImageNet-Val + Broden Probing Dataset
Neuron examples:
We have color-coded the neuron descriptions by whether we believed they were accurate, somewhat correct, or vague/imprecise.
Random Neurons Across Layers 3 & 4 of ResNet-18 with ImageNet-Val + Broden Probing Dataset
Neuron examples:
We have color-coded the neuron descriptions by whether we believed they were accurate, somewhat correct, or vague/imprecise.
Model: ResNet-50
Probing Data: CIFAR-100
Random Neurons Across Layers 1 & 2 of ResNet-50 with CIFAR-100 Probing Dataset
Neuron examples:
We have color-coded the neuron descriptions by whether we believed they were accurate, somewhat correct, or vague/imprecise.
Random Neurons Across Layers 3 & 4 of ResNet-50 with CIFAR-100 Probing Dataset
Neuron examples:
We have color-coded the neuron descriptions by whether we believed they were accurate, somewhat correct, or vague/imprecise.
Model: ViT-B/16
Probing Data: ImageNet-Val + Broden
#4 Random Neurons Across the Encoder of ViT-B/16 with ImageNet-Val + Broden Probing Dataset
Neuron examples:
We have color-coded the neuron descriptions by whether we believed they were accurate, somewhat correct, or vague/imprecise.
