Writing

Explore reinforcement learning through tic-tac-toe self-play, comparing reward designs and testing how training against a minimax opponent affects later self-play. [Read More]

Train an actor-critic agent to swing up Acrobot, then examine how training duration, neural-network size, and reward discounting affect its performance. [Read More]

Train an actor-critic agent to balance a pendulum, using a Gaussian policy and radial basis features to explore continuous control and training stability. [Read More]

Explore policy-gradient learning on Continuous MountainCar, including state discretization, exploration, variance reduction, and discrete versus continuous actions. [Read More]

Apply reinforcement learning to Brawl Stars using only what the agent can see on screen, with a closer look at perception, rewards, planning, and failure cases. [Read More]