search

coreylammie / MemTorch

A Simulation Framework for Memristive Deep Learning Systems
GNU General Public License v3.0
134 stars 45 forks source link

Implementation of RNNs #90

Open nikhil-garg opened 2 years ago

nikhil-garg commented 2 years ago

I think the current framework does not support vanilla RNNs, LSTMs, GRUs. These are really useful in time-series signal processing. Is it possible to add conversion modules for the above as is already present for linear, CNN?

coreylammie commented 2 years ago

Hi @nikhil-garg,

Definitely. I have created three separate issues for (torch.nn.RNN and torch.nn.RNNCell), (torch.nn.LSTM and torch.nn.LSTMCell), and (torch.nn.GRU and torch.nn.GRUCell) modules in #92, #93, and #94, respectively.

Currently, #85 has the highest priority, i.e., the modeling of source and line resistances in passive crossbars, but once this functionality has been implemented, I can start work on implementing the above modules.

Kind Regards,

Corey.