The project (the practical assignment 2) will be done in teams of 4-5. The goal is to apply deep learning methods on a research-oriented problem (e.g. image classification, image segmentation, object detection, text and speech analysis, character animation etc.) It is composed of two parts:

 

Project proposal

You will submit a project proposal of 1-2 pages and make a presentation of the proposal on December 19 or December 21. The project proposal and presentation will include the following:

  • What is the goal of the project, which research topic?
  • What data will you use?
  • What are the related work?
  • What will be the neural network architecture you will use?
  • How will you evaluate the results?
  • Who will do what? When?

The project proposal presentations will take place on 19 and 21 December. Please send your 1-2 page proposal report as pdf to z.yumak@uu.nl with CC to S.T.Langerwerf@uu.nl on 21 December Friday till 23:59.

Final deliverable

You will submit a 4-6 pages double column paper using the Latex or Word template here: https://www.ieee.org/conferences/publishing/templates.html  The final report will include the following:

  • Title, Authors, Abstract
  • Introduction: Introduction to the problem and motivation
  • Related work: Relevant literature
  • Approach: Methodology, technical details and choices made
  • Results: Qualitative and quantitative analysis
  • Conclusion: Summary of your findings
  • References

The final presentations will take place on 23 and 25 January. The final report (pdf file) and the presentation file (ppt) will be sent by email to z.yumak@uu.nl with CC to S.T.Langerwerf@uu.nl on 25 January Friday till 23:59. (A github link to the code will be sent as supplementary material but it will not be graded). The grading will be based on the clarity of the document and the presentation, the technical content and critical analysis.

The suggested deep learning framework is Tensorflow: https://www.tensorflow.org/ However, it is allowed to use other frameworks.

Some useful links: