Describe-by-Recipe vs Interpret-the-Weights

Definition

A framing (from Kording & Lillicrap) that we may never fully interpret the arbitrarily complex learned computations inside a network, but we can understand it by describing the recipe that produced it: architecture, learning rule, initialization, hyperparameters, and training data.

Cast within

No topic-region cast yet — this appears once Describe-by-Recipe vs Interpret-the-Weights's compiled claims are aligned into a topic region's argument tree.

Claims mentioning Describe-by-Recipe vs Interpret-the-Weights (2)

Your notes

Loading notes...