Can Neural Networks Dream? Applying Crick and Mitchison's Theory of Reorganization to Multi-Layer Perceptrons for Reduction of Catastrophic Interference

Date: February 28, 2026
Abstract: According to Crick and Mitchison’s theory of reorganisation, the reverse learning mechanism is responsible for preventing the brain from getting overloaded. This mechanism occurs in REM sleep and is closely associated with dreams. In this study, the reverse learning mechanism is applied to artificial neural networks and its effects on catastrophic forgetting are analysed. The performance of the network with respect to catastrophic forgetting before and after the application of reverse learning is compared.

[URL] | [PDF] | [Google Scholar]

---