On: Weight-Agnostic Neural Networks

[s_Hogg]

Is there meant to be a direct link between this sort of thing and the randomly-initialised approach to Echo State Networks?

[s_Hogg → s_Hogg]

apologies, reading further they refer to:

> a randomly-initialized LSTM [12] with a learned linear output layer can predict time series where traditional RNNs fail

[hardmaru → s_Hogg]

Yeah, though I think we could have referenced a few more works in the reservoir computing area. Will keep that in mind for the next draft.

[theSage → hardmaru]

just starting to read this. It reminds me of meiosis networks + stochastic delta rule where the weights were drawn from a random distribution (normal I think).

[hardmaru → theSage]

Added a few cites to these works, thanks.