Feb 28, 2020 · Our results suggest two routes towards the discovery of neural implementations for credit assignment without weight symmetry.

This space encompasses existing approaches (including feedback alignment and weight mirror), but also elucidates novel learning rules. We show that several of ...

Feb 28, 2020 · Taken together, our results suggest two routes to- wards the discovery of neural implementations for credit assignment without weight symmetry:.

The essential idea of our code base is that by implement- ing custom layers that match the TensorFlow API, but use custom operations for matrix multiplication ( ...

▸ Requires the derivatives of the activation functions. ▸ Requires separate forward and backward passes. ▻The weight updates require access to transposes of the.

Taken together, our results suggest two routes towards the discovery of neural implementations for credit assignment without weight symmetry: further ...

This work investigates a recently proposed local learning rule that yields competitive performance with backpropagation and finds that it is highly ...

Apr 1, 2020 · Our results suggest two routes towards the discovery of neural implementations for credit assignment without weight symmetry.

Jul 12, 2020 · Our results suggest two routes towards the discovery of neural implementations for credit assignment: further improvement of local rules so that ...

Mar 4, 2020 · Excited to share our work "Two Routes to Scalable Credit Assignment without Weight Symmetry". Joint work with amazing collaborators Daniel ...