3 Comments

    • Yes, it works also for deep networks (where it’s even more biologically plausible), however it can be less stable than standard back-propagation due to the the lack of information for each neuron. This condition can drive to larger oscillations that optimized global methods can mitigate.

Leave a Reply