Mono-Forward replaces Forward-Forward's contrastive goodness with local multi-class cross-entropy, outperforming vanilla FF and sometimes backpropagation while using 31% of its memory on MLP-Mixers for PathMNIST.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Mono-Forward: Revisiting Forward-Forward through Objective-Locality Decomposition
Mono-Forward replaces Forward-Forward's contrastive goodness with local multi-class cross-entropy, outperforming vanilla FF and sometimes backpropagation while using 31% of its memory on MLP-Mixers for PathMNIST.