Closed gngdb closed 9 years ago
This is done, initial results were not promising. Although, we should probably return to this after properly dealing with learning rate, momentum etc. Could still be important.
Done, suspect other factors need to be dealt with before this becomes limiting factor. Unclear if it will make results better or worse. Illustrated in notebook Adding MLP Layers
.
Building on what AlexNet has, which is one more than we have at the moment; try some nets with the current best that have more MLP layers. Running with dropout and the same size of the layers we had before.