Posts
-
On Auxiliary Tasks in Deep Reinforcement Learning
In the world of reinforcement learning (RL), a newborn agent sometimes struggles at reaching goals that we wish it could achieve. RL practicers always augment it with auxiliary tasks, which help the agent to learn much faster, more robustly, and ideally perform better. In this post, I will summarize those auxiliary tasks used in deep RL, share ideas behind intuitions, and hopefully inspire you in related work.
-
ELBO — What & Why
ELBO (evidence lower bound) is a key concept in Variational Bayesian Methods. It transforms inference problems, which are always intractable, into optimization problems that can be solved with, for example, gradient-based methods.
subscribe via RSS