Fig. 7: The proposed brain-inspired modifications to the standard generative replay framework.

a Replay-through-feedback. The generator is merged into the main model by equipping it with generative feedback or backward connections. b Conditional replay. To enable the model to generate specific classes, the standard normal prior is replaced by a Gaussian mixture with a separate mode for each class. c Gating based on internal context. For every task or class to be learned, a different subset of neurons in each layer is inhibited during the generative backward pass. d Internal replay. Instead of representations at the input level (e.g., pixel level), hidden or internal representations are replayed.