Homework 3: Instructions

• Find related work, either similar problems and underlying data, or prior work on the same data set

  • what have these other studies done to analyze the data?
  • how have the succeeded? How have they fallen short?
  • how did they validate their performance?
  • how do you expect to improve on what has been done before (either in methods or in performance)?

• What methods will you use to analyze these data?

  • how will features be represented?
  • what is the probabilistic model used to capture important signal in these data? *
  • what assumptions are made about the data in the model (explicit and implicit)?
  • what will you do if those violated assumptions hurt performance?
  • how will you fit the model to the data? will this be computationally tractable?
  • how will you validate performance?
  • feasible to compare method against other methods? approaches to compare models?
  • more details: does a reasonable implementation of the methods exit?