Reproduce “Ghost Units Yield Biologically Plausible Backprop in Deep Neural Networks” 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
Thomas Mesnard, Gaetan Vignoud, Joao Sacramento, Walter Senn, Yoshua Bengio
Ghost Units uses inhibitory units in a recurrent network to predict and cancel routine feedback, allowing target-induced mismatches to carry local learning signals. Approximate gradient propagation is shown under converged cancellation, weak output nudging, and transpose-matched feedback; feedback alignment is tested separately. On MNIST, two-hidden-layer Model B reports 98.21% test accuracy versus a cited 98.04% for a prior network; this is not a controlled comparison.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.