Deep Learning - Week 9


Course starts soon..


Quiz


We will start now with a quiz based on last week's material

You have 6 minutes to answer the quiz.

The quiz link:
Quiz Link
It will be copied in Mattermost and in the Zoom chat.

How to Upload Your Project:
1 - Choose


How to Upload Your Project:
2 - Upload


Kiel.AI Meetings

Presentation of the final project

  • Each group will make a presentation. Ideally, the presentation would be divided among all members of the group.
  • Each group should take 10-15 minutes. There will be the presentation and a round of questions.
  • The structure of the presentation is not strictly fixed, nor it is its content. However, there are some suggestions to have similar structures.

Guidelines for the presentation

  • Group: who are the people in the group.
  • Project: short description of the project and the motivation behind.
  • (*Optional) Tools: which tools you used and how you managed to work in a group.
  • Architecture: what architecture did you use, how many layers, which function (you can be technical on this part)
  • (*Optional) Story: attempts, failures, successes, whatever happened in the process. Sometimes what did not work can be as important as what worked!
  • Results: how it works? can you quantify results or show some visualizations?
  • Baselines:
    • is there a simple baseline with which you can compare? It can be as simple as the mean of the data for time series prediction, or a fixed value of 0.5 for the accuracy in a binary classification problem.
    • can you measure human performances on the task? can you personally take some samples from the test set and solve the problem yourself, and quantify the results?
  • (*Optional) Missing: is there something you missed to improve the project? Time, material, knowledge, data, computational power?
  • (*Optional) Future works: how could the project be improved or extended?

Sharing is caring

If not otherwise discussed with a single group, we will add your project to our Project Page
  • Code: Provide a link to your repository if you have one, and also some short instructions to reproduce it if needed
    Please check that the code is clean (no testing or commented code) and has comments or text fields to understand it!
  • Data: mention if the data is public (with link), or if it is not possible to share.

Peer Reviewed Project Check


Today we make the check about the project status.

Remember from last week

We create Breakout Rooms, and in each Breakout Rooms there are two groups.

The first group has 5 minutes to explain the project, and the other group 2 minutes for feedback and questions. Then group switches and the second group explains and first gives feedback. In total it should take 14-15 minutes.

Afterwards we come back to the main session and each group reports shortly what they learned from the others. So listen carefully.

QUIZ (15 mins)


1. How would you define human and optimal performances in your project? Are they the same?

2. Do you expect your results to be close to one of them?

3. We divide the project in 3 parts:
a) data preprocessing, b) building a model and getting it to work, c) fine tuning to achieve good results. Which part do you see as the easiest, and which as the most complex?

4. Can you use transfer learning on your project? What should the model know?

5. Can you think of a way to divide your objective in smaller tasks? (or the opposite)

6. If you had 3 more months on your projects, what would be your next step? Assuming no money/time limitation, would you like to have more/better data, an expert validating your data, a more complex model, a more powerful machine to iterate the hyperparameter search, some time to test and apply different regularization techniques or something else?

DISCUSSION AND ANSWERS


EXERCISE (15-20 mins)

We go through a tutorial for Time Series prediction using LSTM Networks from the Tensorflow Tutorials.

The python Notebook

Articles about RNN:
The Most Intuitive and Easiest Guide for Recurrent Neural Network
Articles about LSTM:
Illustrated Guide to LSTM’s and GRU’s: A step by step explanation
Understanding LSTM Networks

For the next week

  • Finish the last week of the last course
  • You are done! Missing only the project!