Skip to content

Latest commit

 

History

History
49 lines (41 loc) · 4.2 KB

assignment.md

File metadata and controls

49 lines (41 loc) · 4.2 KB
Raw

Assignment

Throughout the semester, you will be working on a single semestral project in which you will design and implement a deep learning model for solving a real-life problem for which deep learning has already been used. Your deep learning models can be implemented in Tensorflow or Pytorch, and you can use any technology and methods you find (the model's topology is entirely up to you). The assignments will be solved in teams of 3-4 students, but each team member will create his or her own model and implementation. At the end of the semester, you will evaluate these models and within the team, the best-performing model will get the highest number of points.

Topics

  1. image classification - you will select a dataset with images of at least eight classes and will train a neural net that can correctly categorize images into these classes (explanation)
  2. segmentation - you will select a dataset and train a neural net for segmentation (i.e. pixel-wise classification) (explanation)
  3. superresolution - you will train a neural net that can increase the resolution of input images of a given class at least 4x (explanation)
  4. image inpainting - you will train a neural net that can draw the missing parts of an image (recommended domain: human faces with eyes missing) (explanation)
  5. image coloring - you will trian a neural net that can turn grayscale images of a given class into color images (explanation)
  6. deep reinforcement learning - you will train a neural net that will learn to play a simple game using reinforcement learning (explanation)
  7. time series prediction - you will train a neural net that will be able to predict changes in the stock market (explanation)
  8. image noise reduction - you will train a model that will be able to remove noise (missing pixels) from an image (explanation)

Project time plan

  • week 3 - project plan
    • by the end of week 3, all teams must attend a short consultation regarding their assignment where the dataset used and evaluation method will be agreed upon
  • weeks 5 and 6 - each team will give a 5-7 minutes long presentation with:
    • goal specification
    • literature overview
    • possible technologies and methods
    • dataset used
    • evaluation method
  • week 10 - short report
    • each team will hand in a single A4 sheet with a short description of each team member's progress
  • weeks 12 and 13 - assignment due

Final submission

At the end of the semester, each team will submit the following documents:

  • all codes and scripts (with basic comments)
  • one trained model per team member
  • academic paper of 6-8 pages with the following structure and contents:
    1. introduction and problem specification
    2. literature overview
    3. description of dataset and methodology
    4. for each team member, the description of their model
    5. evaluation results for each model with comparison
    6. conclusion

For your final paper, please use the IEEE template.

Grading

You can get at most 30 points for the assignment, part of which will be the same for each team member, and part of it will be individual grading:

  • academic paper - 10 points (the same for all team members)
  • dataset used and overall goal - 5 points (the same for all team members)
  • topology and methodology used - 10 points (individual)
  • model accuracy and performance - up to 5 points (individual)