Balancing stability and plasticity in continual learning: the readout-decomposition of activation change (RDAC) framework.
Daniel AnthesSushrut ThoratPeter KönigTim C. KietzmannPublished in: CoRR (2023)
Keyphrases
- learning mechanism
- main contribution
- learning scheme
- learning process
- learning algorithm
- distributed learning
- reinforcement learning
- biologically plausible
- dynamic bayesian networks
- design principles
- knowledge acquisition
- online learning
- supervised learning
- multi task
- training set
- learning environment
- training data
- genetic algorithm