(Winter 2021) CS 510/610 - Topic on probabilistic graphical models

Project

Instructions

2 people group projects. You may choose one of the options below.

  • Research project. Take any route to gain some research experience.
    • literature review. Pick a topic of interest, and summarize the state of art of it. The final outcome would be a survey paper including open problems.
    • original research. Challenge yourself with some open question in the field. You are encouraged to explore the potential of quantum computing in your own research area. Be creative and keep a critical mind.
  • Programming project. Sure, why not get your hands dirty? You may demonstrate some typical models, especially in real-world applications, compare different models, and test inference/learning algorithms.

  • Societal-impact project. How is PGM shaping the daily life? What is the distribution of research and education efforts across the country? Are resources easily available? Is the research in this direction influencing diversity and equity, and how?

Milestones

  • Proposal: 1-2 pages consisting of 1) the topic, background, context, and motivation; 2) identify a few core references; and 3) a goal you intend to achieve and a plan. (10%)
  • Oral presentation: Each group will have about 30mins to present your project including Q&A. Your need to demonstrate both breath and depth. Aim for a clear introduction that would engage the audience, and then explain one or two key technical ideas in some detail. Every group member needs to participate, and your group will be graded by other fellow students. (20%)
  • Final report: ~10 pages. This should resemble a research paper: 1) a short abstract; 2) an introduction that motivates the topic and offers an overview of the entire report; 3) details including proper preliminary materials (e.g., notations & definitions), explaining some main technical results; and finally 4) further discussion and open questions. (10%)
  • Report format: I recommend you to typeset your reports in LaTeX, and manage your bibliography using BibTeX (your will earn extra credit if you do so). You should use single-column, single-space (between lines) format on letter-size papers. You may adopt the style of popular conference proceedings (e.g., NeurIPS).

Timeline (Tentative)

  • Week 1 - 3: forming groups.
  • Week 4 - 5: discussing project ideas.
  • Feb. 4: proposal due, 11:59pm anytime on Earth.
  • Week 10: in-class presentations.
  • Mar. 18: final report due, 11:59pm anytime on Earth.

Suggested topics (updating)

The list below is merely a kindler, and you are welcome to choose a topic not on the list.

  • [GB12] Samuel J. Gershman, David M. Blei. A Tutorial on Bayesian Nonparametric Models. arxiv

  • [YJ09] Chun-Nam John Yu, Thorsten Joachims. Learning Structural SVMs with Latent Variables. PDF

  • [GYNL19] Lingrui Gan, Xinming Yang, Naveen Narisetty, Feng Liang. Bayesian Joint Estimation of Multiple Graphical Models. Link

  • [ZARX18] Xun Zheng, Bryon Aragam, Pradeep Ravikumar, Eric P. Xing. DAGs with NO TEARS: Continuous Optimization for Structure Learning. arxiv

  • [HYL17] William L. Hamilton, Rex Ying, Jure Leskovec. Representation Learning on Graphs: Methods and Applications. arxiv

  • [KW17] Thomas N. Kipf, Max Welling. Semi-Supervised Classification with Graph Convolutional Networks. arxiv