Explainability Tools and Techniques

#16 - Responsible AI and Building Trust in AI - Liu Feng-Yuan

Topic 8 of 16

Published Nov 23, 2020 · Clip 28:49–30:14 (1:25)

Feng-Yuan recommends TensorFlow Playground to see neuron-level thinking in a browser-trained neural network, showing early layers making simple linear classifications and deeper layers drawing nuanced boundaries. He points to research methods like SHAPLEY values and LIME, which can point to the pixels of an image that led to a prediction.