A javascript emulator inspired by Data Aging use case. We proved - though simplistically - that we could potentially let an RL agent learn a repeated workload to place data in hot and cold partition just by designing a rewarding process.
To test technical feasibility, we deploy an RL agent tune a single In-Mem DB parameter (cpu concurrency) to optimize the performance of a particular workload. Our RL Agent can successfully learn that it does not need all CPU in the system in order to serve the query
A game of Pong based on JavaScript that pits RL player vs original computer player. The RL agent rewarded only when it successfully score a goal at his opponent, and be punished when it lose the ball.
This our first game and because our rewarding design (just hovering the ball) RL Agent learns a creative way to earn endless reward. In the world of RL, be careful of your rewarding design
Deep Reinforcement Learning Agent learnt from first experience including getting penalized for place X on top of O. Eventually, it learns the rules of the game and start winning or draw with the random Agent
Check out our updated interactive slide