COM SCI 260R
Reinforcement Learning
Description: Lecture, four hours; discussion, two hours; outside study, six hours. Fundamentals and advanced topics of reinforcement learning (RL), computational learning approach where agent tries to maximize total amount of reward it receives while interacting with complex and uncertain environments. Includes introduction of Markov decision processes, model-free RL and model-based RL methods, policy optimization, RL distributed system design, as well as case studies of RL in game playing such as AlphaGo, traffic simulation, autonomous driving, and other machine autonomy applications. Advanced topics of RL such as multi-agent RL, human-in-loop method, and imitation learning. Letter grading.
Units: 4.0
Units: 4.0
1 of 1
Most Helpful Review
Winter 2025 - The assignments are interesting, sometimes time consuming, in particular the last assignment and course project. The material is also very interesting, and quite a bit of it is SOTA toward the end. The lectures are OK, the professor mainly reads from slides, but he is clearly passionate about the topic. That said, the exams are either intentionally designed for you to fail and generate a curve, or designed and evaluated in the laziest manner possible. If you do not EXACTLY match what their predefined rubric states, you will receive a zero. This includes rederiving formulas which were derived in class (on an open-note exam!) and matching exact keywords in short answer problems. On top of this, the course staff chose deliberately to hide the correct solutions and rubrics on Gradescope to discourage requesting regrades on solutions which appeared to be fully correct. The final grade is also not curved whatsoever (had 89% raw score, received a B+, scored over 100% on all assignments/project). So your entire grade comes down to this sketchy exam where despite knowing 80-90% of the material, you can easily receive a 60% or lower. If you value your GPA, audit this class. Material is interesting, projects are great, but dealing with this nonsense final is not worth the risk.
Winter 2025 - The assignments are interesting, sometimes time consuming, in particular the last assignment and course project. The material is also very interesting, and quite a bit of it is SOTA toward the end. The lectures are OK, the professor mainly reads from slides, but he is clearly passionate about the topic. That said, the exams are either intentionally designed for you to fail and generate a curve, or designed and evaluated in the laziest manner possible. If you do not EXACTLY match what their predefined rubric states, you will receive a zero. This includes rederiving formulas which were derived in class (on an open-note exam!) and matching exact keywords in short answer problems. On top of this, the course staff chose deliberately to hide the correct solutions and rubrics on Gradescope to discourage requesting regrades on solutions which appeared to be fully correct. The final grade is also not curved whatsoever (had 89% raw score, received a B+, scored over 100% on all assignments/project). So your entire grade comes down to this sketchy exam where despite knowing 80-90% of the material, you can easily receive a 60% or lower. If you value your GPA, audit this class. Material is interesting, projects are great, but dealing with this nonsense final is not worth the risk.
1 of 1
