Dr. Chatzis published a new article in the prestigious IEEE Transactions on Neural Networks and Learning Systems. In his new article entitled “A Latent Manifold Markovian Dynamics Gaussian Process,” Dr. Chatzis introduces a new Deep Learning architecture that, instead of relying on conventional stochastic neural networks principles, uses methods from Bayesian non-parametrics which allow for much better modeling and recognition abilities when training data is limited, or the modeled distributions change in a dynamic fashion. His experimental evaluations have shown that his method completely outperforms state-of-the-art approaches in several benchmark tasks dealing with modeling data with temporal dynamics.