2 reports · 2013–2016

Interpretability

What is actually happening inside a trained network, and how we can tell.

Fig. 02 / Field notesA local view of the image
A convolution scans a local three-by-three receptive field to produce an output feature map.
A small receptive field produces one part of a feature map.

Visualizing Convolutional Networks

Interpretability · 5 min read · Beginner

Zeiler and Fergus visualized CNN layers with deconvolution, revealing edges, textures and objects emerging layer by layer inside the network.

interpretabilitycnnvisualization Read

Grad-CAM: Visual Explanations from Deep Networks

Interpretability · 5 min read · Beginner

Grad-CAM explains CNN decisions with gradient-weighted heatmaps, revealing which image regions drove each prediction without retraining.

grad-caminterpretabilityexplainability Read