flâneur — a map of the web's best reading

Coroutines and backpropagation — Eugene Ha

eugeneha.ca · 2,519 words · saved by 1 readers

The backpropagation algorithm for feedforward neural networks is an immediate corollary of the backpropagation equations. We show that an implementation of the algorithm is itself an immediate corollary, provided that we interpret the backpropagation equations as a coroutine.

2024 March 29 Coroutines and backpropagation \( \def\∂#1/∂#2{\mathchoice {\frac{\partial #1}{\partial #2}} {{\partial #1}/{\partial #2}} {{\partial #1}/{\partial #2}} {{\partial #1}/{\partial #2}}} \newcommand\optensor\otimes \newcommand\tensor{{\mskip2mu}{\optensor}{\mskip2mu}} \newcommand\ophadamard\odot \newcommand\hadamard{{\mskip2mu}{\ophadamard}{\mskip2mu}} \newcommand\vec[1]{\mathbf{#1}} % vectorisation \newcommand\upsigma{\unicode{963}} % upright 'σ' \newcommand\act{\vec\upsigma} % activation function \newcommand\sact\sigma % scalar activation \newcommand\Dact{\vec{D\upsigma}} \newcomm

Explore this link on the map →

related reading