4 ms·

apologies, reading further they refer to: > a randomly-initialized LSTM [12] with a learned linear output layer can predict time series where traditional RNNs

by s_Hogg 7y ago

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 7y ago
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 7y ago
  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 7y ago
    Added a few cites to these works, thanks.