Reproduce “Local learning with imperfect analog synapses” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Théo Dupuis, Adrien Renaudineau, Guillaume Bourgeois, Joel Minguet Lopez, Gabriele Navarro, Jean-Michel Portal, Elisa Vianello, Jacques-Olivier Klein and 1 more
CwSG trains convolutional layers locally, rewarding spatial differences in feature maps rather than high activation alone. Each layer uses its own output and label, avoiding backward array reads. In device-calibrated simulations replaying measured trajectories, CwSG reached 82.5% SVHN test accuracy with 6.4 pulses per device on PCM, versus BP’s 82.2% with 378 pulses; this was not closed-loop hardware training.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.