An application of General Adversarial Imitation Learning using Sharpened Cosign Similarity instead of Convoluted Neural Networks.
To solve the technical test, I used the Generative Adversarial Imitation Learning model better known by the acronym, GAIL. The choice behind this was simple: AI experts in the video game space have used GAIL frequently to solve problems that require specialized knowledge to complete a task without nedding to possess that specialized knowledge themselves. GAIL is an algorithm that is placed squarely between the principles of Machine Learning and Control Theory. Proposed by Jonathan Ho and Stefano Erhman in 2016, this algorithm takes the approach of extracting an expert's cost function using Inverse Reinforcement Learning and extracting the policy from that cost function using reinforcement learning and improves upon it. The paper describing their work is here: https://arxiv.org/pdf/1606.03476.pdf.
