This week our team gave our first Qualification Review Board (QRB) presentation! The QRB presentation is meant to provide an overview of our plans and our current progress, so we can receive feedback for improving our project.

Among the things we explained was our plan for designing our reinforcement learning agents. In order to develop effective and specialized sub-strategies, we set up and ran an experiment on value function hyperparameters. Hyperparameters are parameters for the model that affect the learning process. The goal of this experiment was to identify correlations between the input hyperparameters and different winning strategies. We found that certain hyperparameters had a higher correlation with different win strategies than others. For example, a model that was incentivized to decrease other players’ resource production tended to use a more aggressive strategy than one purely rewarded on its own production.

We received a lot of great feedback! The day after the presentation, we got together to reflect on what went well, what didn’t, and how we can improve ourselves in the future. There will be another QRB presentation later in February, and we plan to do even better then!

That’s all for now! See you next week!