Repository for Term Project Material in 2019
| Time Slot | Title | Presenter |
|---|---|---|
| 10h00 | Reinforcement Learning for Active Appearance Model Dataset Selection | Andre Antonitsch |
| 10h15 | Pruning Neural Networks with Lottery Tickets in a MDP Approach | Andrey Salvi |
| 10h30 | Neural Network Architecture Search Using Automated Planning | Felipe Tasoniero |
| 10h45 | Public Transportation Modelling: Planning Bus Lines Routes | Gabriel Figlarz |
| 11h00 | Learning Heuristics with Graph Convolutional Networks | Matheus Marcon |
| 11h15 | Behavioral Cloning from Image Observation | Nathan Gavenski |
The first assessment grading criteria is as follows (you should follow the proposed structure).
- Application Domain Complexity (30%) - How complex the application domain you selected is difficult to model, yet realistically achievable within the course.
- Paper readability (40%) - How well written the 2-page paper you wrote is, we break this down into the following criteria
- 20% - Introduction clarity: how well does the introduction answers these questions: what is the problem? why is it an important problem? how do aim to solve it? and what follows from your proposed solution?
- 10% - How well do you refer to background material and relate it your proposed application area?
- 10% - How detailed and realistically you plan the work for the rest of the semester?
- Presentation clarity (30%) - How well you presented your project proposal, which we break down into three criteria
- 10% - Use of time during the presentation
- 10% - Slide quality (conciseness, use of figures, etc)
- 10% - Presentation organization
The second assessment grading criteria uses two main criteria First, the technical form of the project
- Application Domain Complexity (15%) - How complex the application domain you selected is difficult to model, yet realistically achievable within the course.
- Domain Modelling (15%) - How close to the underlying domain is the planning model developed in the project? Are the proposed simplifications justified? What is the tradeoff of these simplifications?
- Problem Complexity (15%) - How complex are the problem instances used in the experimentation? Are these instances computationally challenging or are they just toy problems?
- Formalism Appropriateness (15%) - Is the selected formalism (Classical Planning of various types, HTN planning, reinforcement learning) appropriate for the selected domain? Is this selection justified?
Second, the report describing the project and its results
- Report Clarity (10%): Is the report clearly written, following the guidelines for part 1
- Report Problem Description (15%): Does the report describe the problem being addressed with enough detail that it can be replicated?
- Report Implementation (15%): Does the report describe the solution both technically and theoretically in a way that allows others to replicate it?
The same criteria for the project presentation applies to the final presentation, with the following weights
- 30% - Use of time during the presentation
- 30% - Slide quality (conciseness, use of figures, etc)
- 40% - Presentation organization