Machine Learning for Structured Data

10-418 + 10-618, Fall 2019
School of Computer Science
Carnegie Mellon University


There will be 5 homework assignments during the semester in addition to the exams. The assignments will consist of both theoretical and programming problems. Homework assignments will be released via a Piazza announcement explaining where to find the handout, starter code, LaTeX template, etc.

  • Homework 1: Learning to Search for RNNs [speech recognition] (written / programming)
  • Homework 2: General-graph CRF Module for PyTorch [natural language processing] (written / programming)
  • Homework 3: Structured SVM [computer vision] (written / programming)
  • Homework 4: Gibbs Sampling [topic modeling] (written / programming)
  • Homework 5: Variational Inference (written / maybe programming)

Tentative release dates and due dates are listed on the Schedule page.

Project (10-618 only)

Students in 10-618 (but not 10-418) will submit several deliverables throughout the semester. The (tentative) deadlines for each component of the project are listed on the Schedule page.

  1. Team Formation: Each team will consist of 2-3 people. Teams must be specified in advance of the proposal deadline.
  2. Proposal: (1.5-2 pages) The proposal will describe the task, dataset, and methods. Crucially the project must compare multiple methods. The ideal project will start with the approach of an existing paper and then change either the model, inference, or learning in order to paint some contrast. The result will be an empirical study of contrasting approaches.
  3. Midway Poster: The midway poster offers each group a chance to present their progress halfway through the project’s duration. The methods section should be (nearly) in its final form. The experiments section should minimally include a description of the experiments that will be run. Skeleton tables and plots without actual numbers are encouraged. Any results that are ready may also be reported. Students are required to attend the midway poster session.
  4. Final Poster: The final poster should describe the methods that were used and the present experimental results that illustrate a contrast between competing methods. Students are required to attend the final poster session.