Continual Learning of Predictive Models in Video Sequences via Variational Autoencoders

This paper proposes a method for performing continual learning of predictive\nmodels that facilitate the inference of future frames in video sequences. For a\nfirst given experience, an initial Variational Autoencoder, together with a set\nof fully connected neural networks are utilized to respectively learn the\nappearance of video frames and their dynamics at the latent space level. By\nemploying an adapted Markov Jump Particle Filter, the proposed method\nrecognizes new situations and integrates them as predictive models avoiding\ncatastrophic forgetting of previously learned tasks. For evaluating the\nproposed method, this article uses video sequences from a vehicle that performs\ndifferent tasks in a controlled environment.\n