Convolutional Neural Network

Multiple Convolution layers combined to form a network. They can learn abstract features and concepts from raw image pixels.

Perception (Feature Detection) on high dimensional data with a neighborhood condition

Explanation Methods

We can generate Explanations for CNNs by doing Data Visualization of the learned features with Activation Maximization. How can we be sure that the model actually understood the visualized concepts or whether it just remembered parts of the dataset? → Network Dissection