一、課程說明(Course Desc<x>ription)
<br>This course covers from the fundamental concepts of deep
<br>reinforcement learning (DRL), to the state-of-the-art reinforcement
<br>learning methodologies. The target of this course is to train students to
<br>learn and implement DRL models, understanding the concepts and
<br>trade-offs of them, as well as applying them to different evaluation
<br>scenarios and environments. The course schedule is intense,
<br>containing lots of assignments and projects.
<br>Prerequisites:
<br>This course will assume some familiarity with linear algebra,
<br>probabilities, numerical optimization, and machine learning.
<br>
<br>二、指定用書(Text Books)
<br>Sutton & Barto, Reinforcement Learning: An Introduction.
<br>
<br>三、參考書籍(References)
<br>- Deep learning by Ian Goodfellow and Yoshua Bengio and Aaron
<br>Courville.
<br>- David Silver's course on reinforcement learning.
<br>- Sergey Levine’s CS294 course from UC Berkeley.
<br>
<br>四、教學方式(Teaching Method)
<br>This course will be taught in class. Each week has three hours,
<br>consisting of two hours of lectures and one hour of instructor/TA lead
<br>hands-on experiments. The course will have an online forum for Q&A
<br>and discussion. The instructor will offer papers for students to read
<br>every week. There will be 6~7 homeworks. For each homework, we
<br>will post a PDF on ILMS and starter code on Github. We will also post
<br>slides on ILMS for each lecture.
<br>
<br>五、教學進度(Syllabus)
<br>- Introduction to deep learning
<br>- Introduction to multi-armed bandits
<br>- Markov decision process
<br>- Value function approximation
<br>- TD Lambda
<br>- Deep Q-learning
<br>- Policy gradients
<br>- Actor-critic
<br>- Inverse reinforcement learning
<br>- A3C, TRPO, and PPO
<br>- Exploration techniques
<br>- Advanced topics in DRL
<br>- DRL in robotics
<br>
<br>六、成績考核(Evaluation)
<br>- 6~7 homework assignments 70%
<br>- Final project 30%
<br>
<br>七、可連結之網頁位址 相關網頁(Personal Website)
<br>Elsalab.ai
<br>
<br>八、採用下列何項 AI 使用規則
<br>有條件開放，請註明如何使用生成式AI於課程產出 Conditionally open; please specify how 
<br>generative AI will be used in course output