Originally published in my blog.
Approaching a new semester, I eagerly took on the challenge of handling two courses once more, aiming to replicate my previous successes. This time, I enrolled in Reinforcement Learning (RL) and AI for Ethics (AIES). The RL course was intellectually stimulating yet demanded considerable effort. It comprised six homework assignments that involved answering questions using a notebook format, complete with coding tasks. Staying closely attuned to the lectures proved to be a valuable strategy for effectively tackling these assignments. AIES was easy.
The true essence of the RL course was encapsulated in its projects. The first project required delving into a research paper, implementing a random walk algorithm, and meticulously comparing the outcomes against the author’s findings. The second project delved into the realm of hands-on implementation, necessitating rigorous testing of Deep Q-Network (DQN) approaches. The pinnacle of challenge, however, was the third project — constructing an intricate multiagent RL system tailored to navigate the intricate Google Research Football environment. Achieving success here pivoted on crafting comprehensive reports that didn’t merely showcase code but demonstrated a profound comprehension of algorithms and strategies. This understanding, highlighted through these reports, became the linchpin for the assessment. Attending office hours and actively participating in Ed Discussions proved essential strategies. The office hours shed light on the rubrics for the reports and threw out some hints that helped in the implementation. In comparison to the Machine Learning (ML) course, this proved more engaging, partly due to my stronger background in ML. It demanded a substantial investment of about 20–25 hours weekly.
The lesson of not underestimating the initial workload was learned the hard way, especially evident in the formidable scope of the third project. The protracted training tasks, sometimes spanning days, underscored the importance of an early start. The exam format was straightforward — multiple-choice questions designed to be deceptively tricky, but the course material and lectures are enough to clear these. A welcome surprise was the infusion of Game Theory concepts, expanding the course’s scope beyond traditional RL topics.
Grading adhered to a curve, necessitating surpassing class averages for an A, but securing a B was a realistic goal.
On a contrasting note, AI for Ethics offered a serene counterpart to the semester’s demands. With its more theoretical nature, the course explored ethical complexities within the tech industry. Assignments encompassed a spectrum of activities including case studies, theoretical techniques, and lightweight coding tasks. The final exam encapsulated the theoretical essence of the course in a succinct manner. Offering a lighter workload of around five hours weekly, AIES harmoniously balanced out the rigors of the RL course. It presented a pathway to securing an A with relatively manageable effort.