In my Security-Gym post I described the environment I built to generate realistic cybersecurity log streams for RL experiments. The next conditions I wanted to test were to see if a continual learning agent can actually detect attacks in this data running as a …
Over the past several months I've been running streaming RL experiments on real attack data extracted from a Cowrie SSH honeypot. The RL agent performance on the Cowrie data is promising and I wanted to continue testing different algorithms. However I …
I'm continuing to focus on RL prediction steps 1/2 of the Alberta Plan. In my first experiment, I showed that IDBD diverges almost immediately on the real honeypot data I'm collecting and testing with. Given enough time Autostep fared much better but …
My strategy for learning the foundations and contributing to the Alberta Plan for AI Research is to begin at Step 1 and work my way through the plan, learning and reading the associated literature as I go. This should give me a solid …
I continue to be impressed with the JAX library as I learn how to use it. In my last post I mentioned switching from PyTorch to JAX for the alberta-framework. I had seen about a 2.78x speedup just …
As I officially begin my D.Eng, I've been eager to start producing something. My method of learning has always been to understand things "under the hood". I don't feel I fully understand concepts until I can learn them from the …