Probability Functional Descent: A Unifying Perspective on GANs, Variational Inference, and Reinforcement Learning
It provides a theoretical unification for diverse ML problems, which is incremental in nature.
The paper introduces a unifying framework that frames generative adversarial networks, variational inference, and reinforcement learning as minimization problems on probability measures, and presents a generic optimization algorithm called probability functional descent that recovers existing methods in these areas.
This paper provides a unifying view of a wide range of problems of interest in machine learning by framing them as the minimization of functionals defined on the space of probability measures. In particular, we show that generative adversarial networks, variational inference, and actor-critic methods in reinforcement learning can all be seen through the lens of our framework. We then discuss a generic optimization algorithm for our formulation, called probability functional descent (PFD), and show how this algorithm recovers existing methods developed independently in the settings mentioned earlier.