Basic codes for "Causal importance of low-level feature selectivity for generalization in image recognition"

Paper

Causal importance of low-level feature selectivity for generalization in image recognition. Jumpei Ukita, Neural Networks, 2020.

Paper Link
Short Version at NeurIPS2018 Workshop

Tested environment

• Keras 2.0.8
• TensorFlow 1.3.0

Usage

1. run model_training.py to train the CNN model on CIFAR10 dataset.
2. run grating_analysis.py to analyze the orientation selectivity of hidden layers.
3. run ablation.py to perform ablation analysis.