If you could fill your ideas here -
- In training the network, what does this line help to do? I had noticed omitting them had no noticeable effect in evaluation.
synapse_0_direction_count = np.zeros_like(synapse_0)
synapse_1_direction_count = np.zeros_like(synapse_1)
if(j > 0):
synapse_0_direction_count += np.abs(((synapse_0_weight_update > 0)+0) — ((prev_synapse_0_weight_update > 0) + 0))
synapse_1_direction_count += np.abs(((synapse_1_weight_update > 0)+0) — ((prev_synapse_1_weight_update > 0) + 0))
2. If the model had to be updated with new classes or add new data to existing classes, what could be an efficient method to “patch” them? Re training them entirely from start would take time. A lot of time.