Browse Teach Curate

Representing a Computation as a Directed Computational Graph

Any numeric expression can be decomposed into a directed acyclic graph whose nodes are elementary operations and whose edges carry intermediate values from producer to consumer. Learners build the graph for an expression, execute it in forward topological order to obtain a value, identify each node's immediate inputs and outputs, and traverse the same graph in reverse order.

jeremy Video

Computation Graphs in Neural Networks Using Forward and Backward Propagation

In the domain of machine learning optimization theory, computation graphs formally represent a directed acyclic graph (DAG) that decomposes complex differentiable functions into elementary operations…