Formalizing the Generalization-Forgetting Trade-off in Continual Learning

We formulate the continual learning (CL) problem via dynamic programming and\nmodel the trade-off between catastrophic forgetting and generalization as a\ntwo-player sequential game. In this approach, player 1 maximizes the cost due\nto lack of generalization whereas player 2 minimizes the cost due to\ncatastrophic forgetting. We show theoretically that a balance point between the\ntwo players exists for each task and that this point is stable (once the\nbalance is achieved, the two players stay at the balance point). Next, we\nintroduce balanced continual learning (BCL), which is designed to attain\nbalance between generalization and forgetting and empirically demonstrate that\nBCL is comparable to or better than the state of the art.\n