Modulation of early visual processing alleviates capacity limits in solving multiple tasks

Code accompanying a CCN 2019 submission.

train_NNs.py lets you train MLPs with 1 hidden layer while varying the number of tasks the network has to perform and the number of units in the hidden layer.

early-late-task-mod-perm-mnist_analysis.ipynb lets you analyse the trained networks to assess the nature of the task-based modulations and perform ablation analyses, among other things.

*** Coded with Python 2.7 and Tensorflow 1.3.0 (CPU version) ***