Biologically-Inspired Continual Learning of Human Motion Sequences
It addresses continual learning for human motion generation, which is incremental as it builds on an existing brain-inspired replay model.
This work tackled the problem of continual learning for generating human motion sequences by proposing a biologically-inspired conditional temporal variational autoencoder (BI-CTVAE), which achieved a final classification accuracy of 78% after sequential learning, a 63% improvement over no-replay and only 5.4% lower than a state-of-the-art offline model.
This work proposes a model for continual learning on tasks involving temporal sequences, specifically, human motions. It improves on a recently proposed brain-inspired replay model (BI-R) by building a biologically-inspired conditional temporal variational autoencoder (BI-CTVAE), which instantiates a latent mixture-of-Gaussians for class representation. We investigate a novel continual-learning-to-generate (CL2Gen) scenario where the model generates motion sequences of different classes. The generative accuracy of the model is tested over a set of tasks. The final classification accuracy of BI-CTVAE on a human motion dataset after sequentially learning all action classes is 78%, which is 63% higher than using no-replay, and only 5.4% lower than a state-of-the-art offline trained GRU model.