What is in a Neural Network?
Dissecting DNNs with CLIP-Dissect
Home
How To Use
Project Paper
Model:
ResNet 50 trained on ImageNet;
Probing Data:
Broden;
Concept Set:
top-20k English Words;
Similarity Function:
Soft-WPMI
We are dissecting the layers with the star symbol ✦
Search by Neuron ID:
Layer 1
Layer 2
Layer 3
Layer 4
Top-K Concepts (up to 10):
Search by Concept:
Layer 1
Layer 2
Layer 3
Layer 4
Search
Previous Neuron
Next Neuron
Searched Neuron IDs History
Neuron ID
Layer
Top Dissected Concept
Top Concept Similarity
Searched Concepts History
Concept
Neurons Found
Layers
Top Matching Neuron ID
Top Matching Neuron Similarity
