A stochastic extension of the Recurrent Neural Network with Parametric Biases (RNNPB) that injects Gaussian noise into the parametric-bias latent space using the variational-autoencoder reparameterization trick, so the model learns a continuous probabilistic latent representation of multidimensional sequences rather than point estimates. Grounded in predictive coding and the Bayesian-brain view, it quantifies and adjusts uncertainty during both learning and inference, resisting overfitting and improving generation and recognition of robotic motion sequences and generalization to novel patterns compared with the deterministic RNNPB.
The ability to generate and recognize sequential data is fundamental for autonomous systems operating in dynamic environments. Inspired by the key principles of the brain-predictive coding and the Ba…