search

fabrizioaymone / suitability-of-Forward-Forward-and-PEPITA-learning

This repository contains the spreadsheet of the quantitative analysis performed for the paper "Suitability of Forward-Forward and PEPITA Learning to MLCommons-Tiny benchmarks".
Other
4 stars 1 forks source link

Accuracy evaluation of the same approaches #1

Open yttuncel opened 1 year ago

yttuncel commented 1 year ago

Hi,

Thanks much for this work, I've been thinking of doing a similar study myself. However, I couldn't make FFA work with CNNs, which was also highlighted by Hinton. For example, did you actually make FFA work with RESNET for CIFAR10 classification? or did you only study the memory/time comparisons and ignored the accuracy?

Looking forward to your answer! Once again, thank you for making this public!

whubaichuan commented 11 months ago

@yttuncel Hi, I think in this repository, they did not actually run the classification with the acc/error performance. They just analyze the computation and memory usage.

There is another repository want to implement the FF with CNN but is still under work and can not work.

By the way, you mentioned that ''which was also highlighted by Hinton'', how do you explain it? I think for CIFAR10, hinton uses a kind of CNN. Is that right?