Skip to Content

Reinforcement Learning with Randomized Rewards

Current reinforcement learning models are brittle and unable to accept change in reward structure. In order to obtain a true general AI our machine learning algorithms must be able to expect changes in a reward structure. That is why the research is taking a look at randomizing the reward that is obtained in simple machine learning experiments. One of the areas that have been developed by CUbic reasarches is the implementation of infrastructure and monitoring software. To solve this problem, CUbIC has developed custom software using python to show training time and hardware statistics.

Symposium Participant

Kyle Shumway

Project Details

Symposium Date: Spring 2018

Research Theme: Education

Presentation Type: FURI

Faculty Mentors

  • Troy McDaniel