
Deep learning methods based on gradient descent gradually lose plasticity and the ability to continually learn in dynamic environments, requiring additional techniques to maintain variability and adaptability.


coremsg

The pervasive problem of artificial neural networks losing plasticity in continual-learning settings is demonstrated and a simple solution called the continual&nbsp;backpropagation algorithm is described to prevent this issue.

Loss of plasticity in deep continual learning - Nature

continual-learning-challenges-for-deep-neural-networks-loss-of-plasticity-and-inability-to-adapt-over-time


Continual Learning Challenges for Deep Neural Networks: Loss of Plasticity and Inability to Adapt Over Time
