Direct Feedback Alignment with Sparse Connections for Local Learning — mentioned by raindeer3

Sun, 11 Aug 2019 07:47:39 GMT

Very interesting also is that you can directly send the error to intermediate layers through sparse random projections without the need for any layerwise backpropagation. This relaxation of the structure of the backward pass makes bp even more plausible from a biological perspective. https://arxiv.org/abs/1609.01596 https://arxiv.org/abs/1903.02083

Paper and discussion history. A mention is not an endorsement.

Direct Feedback Alignment with Sparse Connections for Local Learning · HackerPapers discussions

Direct Feedback Alignment with Sparse Connections for Local Learning — mentioned by raindeer3