You can lead horses to water.............
Just watching again this presentation from Simon Thorpe in June 2019.
First noticed by me when put up by Uiux on the crapper 2 years ago.
Worth another look for anyone still grappling with the basics......
Thanks for the refresher Hoppy,
This slide is particularly important in the rate coding v rank coding debate and the backpropagation v feedforward debate.
D. Firing rate per se cannot be used during visual processing to code analog values with any real precision.
E. Coding of analog values could however be achieved by making use of the arrival times of spikes from different sources - the earliest arriving signals could be given priority (N-of-M coding)
Tragically, some of our competitors cling to rate coding.
Rate coding uses more computer operations and is thus slower and more power hungry. N-of-M coding provides a huge saving in computer operations, thus saving time and power.
G. Although feedback pathways between different layers are present, there may not be time to use them during normal visual processing.
This is because the minimum time to pass through the 10 layers of the visual recognition pathway in the brain is 1/10 seconds, which averages out at 1/100 seconds per layer (synapse/neuron), ie, backpropagation isn't used in object recognition in the brain.
Another tragedy for some of our competitors.
Backpropagation will slow the process and use more power because it requires more compute operations.