Machine Learning

Class Presentation and Project

Presentation Topic Due: April 8, 2004 (midnight)
Project Proposal Due: April 8, 2004 (midnight)
Project Due: May 11, 2004 (midnight)

Presentations

On or before the above due date you should email me (holder@cse.uta.edu) the title of the paper(s) for your class presentation. Also, include date preferences for your presentation (see Schedule for possible presentation dates). Presentations should be approximately 25 minutes. Since you will be graded on (among other things) your coverage of the material in the paper, I strongly recommend you use PowerPoint, or some other electronic form, for your presentation to expedite the delivery of information. I would suggest a more recent paper of interest to you from conference proceedings of the International Conference on Machine Learning (ICML), International Conference on Knowledge Discovery and Data Mining (KDD), National Conference on Artificial Intelligence (AAAI), or International Joint Conference on Artificial Intelligence (IJCAI). The Machine Learning Journal or Journal of Machine Learning Research are also good sources for papers. The topic of your paper presentation may coincide with your project topic. Paper selections must be approved by me.

Projects

On or before the above proposal due date, you should email me (holder@cse.uta.edu) a proposal for your machine learning class project. The project proposal should describe the problem or issue you are addressing, the relevance of the problem to machine learning, your approach or approaches to the problem, possible empirical and/or theoretical analyses of the approaches, and the expected results. Below is a list of some project ideas. The project is a significant part of this course, and I expect you to spend a significant amount of time preparing your final report. Therefore, I encourage you to turn in your project proposal as soon as possible and get started early. Project proposals must be approved by me.

The project will consist of a writeup describing, in more detail, the problem, relevance, approaches, analyses, results, and conclusions about the advantages and disadvantages of your approach. If your project involves some programming, turn in copies of all code. If your code is a modification of existing code, be sure to clearly indicate your modifications. Also turn in any data used and program output traces for all runs.

Project Ideas

  1. Applied Learning. Some of the biggest success stories of machine learning come from the application of one or a small number of learning methods to a specific problem. A project on applied learning would first identify the application domain, collect data, run appropriate learning methods, and analyze results in terms of relevance to the domain (ideally done by a domain expert). Typically, this approach goes through several iterations of modifications and tweaks to the data representation and learning method in order to achieve the highest performance possible. The project would report the data representation, modifications made at each iteration, the intermediate and final results, and an evaluation of these results and the method in terms of applicability to the domain.
  2. Improvements to Learning Methods. Several of the sections on machine learning approaches suggest problems with the approach and possible solutions (e.g., noise and missing values in inductive learning, incomplete and/or incorrect domain theories in deductive learning, or combining multiple learning methods to overcome weaknesses in the isolated methods). The only way to know if these improvements work is to implement them and test them empirically in a variety of learning tasks. A project for evaluating improvements to a learning method would implement several possible approaches to the improvement and test each to determine the effectiveness of the improvement.
  3. Empirical Comparison of Machine Learning Methods. In the homework you compared the performance of several learning algorithms. This comparison can be expanded into a project by increasing the number of learners and/or data sets (both real and artificial). The learners do not have to be limited to inductive techniques. The project would involve running several experiments on different combinations of learners and data sets, comparing results, and determining the underlying reasons behind discrepancies in the performance of different learning methods.
  4. Survey of Machine Learning Methods. This course has introduced only a subset of the available machine learning methods. Many other methods exist, and new methods are constantly being developed. A survey of methods for a particular type of learning would involve a literature search for old and new methods, descriptions of these methods, and generalizations across several methods indicating patterns in the approaches to learning. Implementations and empirical comparisons of selected methods would increase the validity of your conclusions regarding advantages and disadvantages of different approaches.