[R] DeepMind: Neural networks suffer from catastrophic forgetting when tasks are encountered sequentially. We overcome this by Bayesian inference in function space, using inducing point sparse GP methods and by optimising over rehearsal data points

Written by torontoai on June 17, 2019. Posted in Reddit MachineLearning.

• Previous
• Next