Current browse context:
stat.ML
Change to browse by:
References & Citations
Statistics > Machine Learning
Title: Flow Contrastive Estimation of Energy-Based Models
(Submitted on 2 Dec 2019 (this version), latest version 1 Apr 2020 (v2))
Abstract: This paper studies a training method to jointly estimate an energy-based model and a flow-based model, in which the two models are iteratively updated based on a shared adversarial value function. This joint training method has the following traits. (1) The update of the energy-based model is based on noise contrastive estimation, with the flow model serving as a strong noise distribution. (2) The update of the flow model approximately minimizes the Jensen-Shannon divergence between the flow model and the data distribution. (3) Unlike generative adversarial networks (GAN) which estimates an implicit probability distribution defined by a generator model, our method estimates two explicit probabilistic distributions on the data. Using the proposed method we demonstrate a significant improvement on the synthesis quality of the flow model, and show the effectiveness of unsupervised feature learning by the learned energy-based model. Furthermore, the proposed training method can be easily adapted to semi-supervised learning. We achieve competitive results to the state-of-the-art semi-supervised learning methods.
Submission history
From: Ruiqi Gao [view email][v1] Mon, 2 Dec 2019 06:29:36 GMT (2153kb,D)
[v2] Wed, 1 Apr 2020 17:53:41 GMT (2167kb,D)
Link back to: arXiv, form interface, contact.