IE532: Course Project Information

Sewoong Oh, University of Illinois Urbana-Champaign

Project Schedule

Important dates

  1. Friday November 3rd: Email (to the TA and the instructor) a ranked list of 3 projects you want to work on.

  2. Thursday November 9th: Submit a one page description of your proposed work. It should include

    1. Problem statement in a few sentences

    2. Input to the training

    3. Output of the training

    4. Suggestions for probabilistic method for generating the data for training and testing

    5. Suggestions for the result to be tested and reported (Write as many as you can think of, you do not have to actually do everything you write in this one page description), for example

  • ”We plan to plot the testing error as a function of the size of the testing graph, while fixing the training graph to be of size 100, and report how the testing error degrades as the testing graph size increases”

  • ”We plan to compare the testing error against another heuristic algorithm called XXX that is not graph neural network”

  • ”We plan to train on Erdos-Renyi random graphs and test it on Stochastic Block Model random graphs, and report how much the performance degrades.”

    1. Suggested improvements to be attempted, for example

  • ”We plan to try out an architecture with more layers to improve the testing error when testing graph is large”

  • ”We plan to try out an architecture by concatenating multiple graph neural networks, to XXX”

  • ”We want to try a larger architecture, but add regularization by doing a special stochastic gradient descent to overcome overfitting” … ”Any novel ideas here will be highly regarded, even if it does not work.”

  1. December 10,11,12 (The actual dates to be determined later) : Final presentation 15 minutes for each team, 12 teams, and 3 hours in total.