Meta-learnt priors slow down catastrophic forgetting in neural networks

Giacomo Spigler

Current training regimes for deep learning usually involve exposure to a single task / dataset at a time. Here we start from the observation that in this context the trained model…