KRNet: Towards Efficient Knowledge Replay
This work addresses storage efficiency for researchers and practitioners in continual learning and domain adaptation, though it is incremental as it builds on existing knowledge replay techniques.
The paper tackles the problem of inefficient storage and redundancy in knowledge replay methods by proposing KRNet, which maps sample identities directly to data, reducing storage costs by 400x compared to autoencoders.
The knowledge replay technique has been widely used in many tasks such as continual learning and continuous domain adaptation. The key lies in how to effectively encode the knowledge extracted from previous data and replay them during current training procedure. A simple yet effective model to achieve knowledge replay is autoencoder. However, the number of stored latent codes in autoencoder increases linearly with the scale of data and the trained encoder is redundant for the replaying stage. In this paper, we propose a novel and efficient knowledge recording network (KRNet) which directly maps an arbitrary sample identity number to the corresponding datum. Compared with autoencoder, our KRNet requires significantly ($400\times$) less storage cost for the latent codes and can be trained without the encoder sub-network. Extensive experiments validate the efficiency of KRNet, and as a showcase, it is successfully applied in the task of continual learning.