a576eafbce762079f7d1f77fca1c5cc2-AuthorFeedback.pdf

The novelty of our contribution is2 two-fold: First, our proposed learning rule with modifications3 to both sides of the gradient update is novel. Tasks are best retained using the double-sided approach.R4raised that9 theydidnotunderstand themotivation forthedouble-sided learning rule. For example, when aMemory15 task is learned first, followed by aDelay task with the same input/output structure (e.g. Our work ismotivated by the question ofhowthesameneural population may be36 involved in computations relating to multiple tasks (lines 23-24). Fixed point structures were highly overlapping upon visual inspection in TDR subspaces.49