Sessions 1@5

Session 1

Course content

• Explain the objectives and requirements of the course in terms of:
  • Lectures structure and homework load
  • Time and dedication
  • Goals and expectations from course and participants
• Explain the environment selected initially, this will be Google Colab
  • Demonstrate and guide on the initial usage
  • Ask related questions

Homework

• Check understanding and capacity to:
  • Use Google Colab , tutorial
    • Connect GDrive with Google Colab, tutorial
• Read some theory about neural networks, it will be explained, though try to start understanding
  • Basic:
    • https://medium.freecodecamp.org/want-to-know-how-deep-learning-works-heres-a-quick-guide-for-everyone-1aedeca88076
    • https://www.analyticsvidhya.com/blog/2018/10/introduction-neural-networks-deep-learning/
  • Recommended:
    • http://introtodeeplearning.com/materials/2019_6S191_L1.pdf
    • https://cs231n.github.io/neural-networks-1/
    • https://cs231n.github.io/optimization-2/

---

Session 2

Course content

• Thoroughly explain the basic theory of neural networks, using the following material:
  1. Perceptron
  2. Neural Network on its MLP (Multilayer Perceptron conception)
  3. Gradient descend
  4. Training procedure:
     • Forward pass
     • Loss calculation
     • Backpropagation

Homework

As a first vanilla training set, MNIST dataset from Kaggle, to be solved:

• Using which ever machine learning algorithm
• Using deep learning, as explained in the tutorial To test the models there are two options:

1. Split a test set, approximatively 20-25%, and test your models accuracy on it.
2. Use the leaderboard from Kaggle (recommended).

---

Session 3

Course content

This session was a practical session focused on the solution of different doubts that arose of the MNIST realization, specifically of the deep learning part; special focus was introduced in the necessity to start looking into the official documentation (for Keras, see here). A brief review on key deep learning concepts was done using this playground.

Homework

No extra homework will be given, due to lack of advancement from participants. A couple of days more to try to "improve" your homework, by:

• Trying to solve the problem at hand
• Trying new architectures

---

Session 4

Course content

During this session, the correction of the MNIST homework was done.

Homework

As on next session there will be the initial explanation of the convolutional neural networks, CNN, a previous lookout to theory will help during the explanation:

• https://towardsdatascience.com/demystifying-convolutional-neural-networks-384785791596
• https://ujjwalkarn.me/2016/08/11/intuitive-explanation-convnets/

---

Session 5

Course content

Session devoted to the theoretical explanation of CNN networks, different material used during the session for explanation:

• Explained the CNN structure:
  • https://ujjwalkarn.me/2016/08/11/intuitive-explanation-convnets/
  • https://towardsdatascience.com/applied-deep-learning-part-4-convolutional-neural-networks-584bc134c1e2
  • https://towardsdatascience.com/demystifying-convolutional-neural-networks-384785791596
• Extra: More advanced types of convolutions:
  • https://towardsdatascience.com/a-comprehensive-introduction-to-different-types-of-convolutions-in-deep-learning-669281e58215

Homework

We start to play with a Kaggle challenge; the Plant Seedings Classification, no more MNIST! The goal is to start exploring this challenge by:

• Understand the challenge
• EDA (Exploratory Data Analysis)
• Prepare neural networks ingestion pipelines

The expected time allocation will be 3 sessions.