ai w/o rl
Continual learning
also called "Incremental Learning", "machine Learning 2.0", or "Lifelong Machine Learning"
Goal:
1. Retain learned knowledge from previous tasks & use it to help furure learning
2. Minimize **catastrophic forgetting**
Three solution concept
1. Replay methods (Rehearsal)
2. Regularization-based methods (Memory)
3. Parameter isolation methods (Architectural)
Popular Solution
1. GDumb (Random Sampling concept)