Deep Variational Bayes Filters: Unsupervised Learning of State Space Models from Raw Data

Citations Over Time

Abstract

We introduce Deep Variational Bayes Filters (DVBF), a new method for unsupervised learning and identification of latent Markovian state space models. Leveraging recent advances in Stochastic Gradient Variational Bayes, DVBF can overcome intractable inference distributions via variational inference. Thus, it can handle highly nonlinear input data with temporal and spatial dependencies such as image sequences without domain knowledge. Our experiments show that enabling backpropagation through transitions enforces state space assumptions and significantly improves information content of the latent embedding. This also enables realistic long-term prediction.

Related Papers

• Embed to control: a locally Linear Latent dynamics model for control from raw images(2015)
• → Deep Variational Bayes Filters: Unsupervised Learning of State Space\n Models from Raw Data(2016)182 cited
• → Deep Kalman Filters(2015)142 cited