Surya Dantuluri's Blog

Hyperparameter Behavior in Reinforcement Learning

How do hyperparameters affect traditional RL algorithms?

At the moment this project is on halt until funding/another way of running these models are possible.

Essentially we look at the benefits of using different optimizers in RL.


Started the project as a spinoff of my previous post. I decided that since I'm writing so much Tex in Mathjax, why not just write this paper (that I was eventually going to do) in LaTex then easily use the VPG explanation into my previous blog post (which is why it isn't finished yet).

I worked on the project heavily in February and have finished all necessary testing for the k3s Kubernetes cluster program to begin and start on Google Cloud Compute.

Why I can't finish the paper


From the initial testing the project will take 5-8 weeks to complete with 6 asynchronous simulation and rollout servers. The estimated cost comes to around $8,000, which for a HS student, I don't have to any capacity. I've applied to Google's Research Credits program and hoping to gain this funding to start the cluster.


writes articles on Machine Learning, Full Stack Development, and Insightful Topics