Signal Processing Reading Group (Spring 2018)

Time and Venue

• Wednesday 4:00 pm - 6:00 pm. KEC 1005 (tentative)

• sign up: send emails to xiao.fu@oregonstate.edu

• Topic of Spring 2018: Approximate Dynamic Programming and Reinforcement Learning

• Note: We also welcome undergraduate students who are interested in the broad area of signal and data analytics to join our reading group. Please send me emails and I can add you to the mailing list.

• Location changes: BEXL 323 for April 25; then we move to MLM 318 for the rest of the term.

Reading Materials

• From Prof. Dimitri Berstekas's lectures (MIT): http://www.mit.edu/~dimitrib/Dynamic_Prog_Videos.html

• From Prof. David Silver's website (UCL): http://www0.cs.ucl.ac.uk/staff/d.silver/web/Teaching.html

Lectures (Tentative; following Prof. Silver's slides)

• Introduction, Markov Decision Process (Ahmad Zoubi, April 11)

• Planning by Dynamic Programming (Zeyu You, April 25)

• Model-free prediction and control (Falah Alanazi, May 2)

• Value function approximation (Sharmin Kibria, May 9)

• Policy gradient (Trung Viet Vu, May 16)

• Integrating Learning and Planning (Henrique Dantas, May 23)

• Exploration and Exploitation (Leonardo Cavalcanti, May 30)

• Case Study: RL in Classic Games (June 6)

• TBD (KEC 1005, June 13)

More discussion

• April 11: Thanks to Ahmad, the first lecture was very nice!

  • Here is a webpage that gives a very simple proof of ‘‘strictly diagonally dominant matrices are invertible’’ check out.

  • If someone found the answer to Ahmad's question towards the end of the lecture, shoot me an email and I'll post it here.

• April 25: about contraction mapping see here

• May 2: Convergence of Q-learning. Read this paper here.

• May 16: Critic-Actor Pair and the Fisher Matrix: Seehere