Projects

Requirements to the problems

A student research project must solve one particular well-formulated problem and must carry one clear message. The suggested research problems should meet the following requirements:

1. A problem should have a novel solution, suggested by the expert or the course advisor (novelty).
2. Solution of the problem should have a practical application, anticipated by a number of experts (importance).
3. Ideally, the solution should influence the problem statements and research in the research area (significance and citations).
4. The topic of the problem should correspond to some narrow field of Machine Learning and Data Analisys.
5. Completion of the tasks, associated with solving the problem, performing computational experiments and the paper should take the planned amount of hours.

Evaluation criteria and key features of a nice paper

1. The paper regards one particular problem, proposing new and clear solution.
2. The solution is theoretically justified. The properties of proposed method are studied.
3. The theoretical problem is illustrated with practical examples. The paper contains a comprehensive description of computational experiments with error analysis.
4. The author understands to whom the paper is addressed. The author is able, after some publication analysis, to name a journal for submission.

Some indicators of a weakly stated problem

1. An extensively researched problem is solved by a well-known method, though such combination may have not have been widely used (lack of novelty).
2. A new solution to the problem is proposed, but the algorithm is tested on the UCI data (lack of importance).
3. The solution is essentially a number of heuristics (lack of justification).

The project description template

• Title: a title of the paper to be submitted in a scientific journal.
• Problem: a short informal problem statement. An optimization (\(\text{arg}\min\) style) problem statement is welcome. Put a link to a classic problem statement, if any.
• Data: a brief description of your data set, a link to the data.
• References:List of the papers and supplementary materials, which includes a comprehensive problem statement, links to new results, comprehensive reviews in the field.
• Basic solution: a nearest simple algorithm, which solves he problem to start from.
• Method: a proposed solution of the problem with details about quality criteria and error analysis.
• Novelty: the main message convey to the reader and to defend.
• Authors: the expert and the consultant. Link to the project, is it has been started.