Welcome to AI primer course#
Main aim of this course is to give you enough information so that you can start exploring field of AI on your own and maybe even start searching for DS role. We have only 5 main chapters and one bonus lecture to cover.
Unsupervised learning
SVD (Singular Value Decomposition) - it’s a good tool to introduce both technical tools we will be working with as well as giving us a glimpse at unsupervised learning.
Supervised learning
RF (Random Forests) - one of the first “silver bullets” out there. Our discussion will also cover Shannon’s work on entropy as it’s one of the key ingredients.
Deep learning
DNN (Deep Neural Networks) - we will build our own Perceptron from scratch, thus focusing on gradient descent and backprop on the way. By changing activation function logistic regression will be introduced and finally we will explore what a stack of layers (deep NN) can offer.
CNN (Convolutional Neural Networks) - even though different techniques come and go in deep learning world I strongly believe that CNN’s will be around for quite some time to come. We will use them not only for images, but also for time series prediction.
Attention - powerful idea that stands behind Transformers and one of the enablers for GPT-3, DALL-E 2 and others.
Reinforcement Learning (bonus lecture)
TD (Temporal Difference) - one of the core principles in reinforcement learning. We will apply it to play tic-tac-toe.
Also we will cover following toolset, which hopefully will be useful for your future projects:
numpy (mainly in SVD and FCN lectures) - will help us store vectors, matrices and perform operations on them.
matplotlib (in all lectures) - nice and simple plotting lib.
scikit-learn - ML library.
pandas (mainly in RF lecture) - structured way of looking at tabular data.
PyTorch (FCN and CNN lectures) - simple deep learning library based on tensorflow.
git (final project) - version control tool.
Toolset will be presented only in lectures, thus it’s up to you to learn them on your own if you do not plan to attend. There are a lot of resources, but I highly suggest to read intros in corresponding docs.
What to expect from a single lecture?
There will be no clear distinction between theory and practice, thus you should have your PC ready for small assignments that you will encounter on the way.
Most important material will be listed here, but during lectures you will hear and see a lot of complementary material.
Each lecture will end with a list of resources (some of them mandatory). We will start a new lecture with a recap of what was done last time and discussion regarding mentioned resources in the hope to deepen understanding in the subject and inspire you to search for sources and publications yourself.
Launching notebooks#
You can launch notebooks while in interactive book by simply pressing the rocket logo and choosing Colab. To get faster run times click Runtime and Change runtime type, then select GPU or TPU. If necessary you can install missing packages by running !pip install [package name] directly in the notebook.
NOTE: Colab will not save your changes between sessions! Download the notebook or save a copy in Google Drive before closing the browser. If you want to open notebooks locally (for a quick preview) you might find nteract useful.
As an alternative you can use non free, but cheap options like Jarvislabs or Paperspace. Actually Paperspace has free GPU option, but often it is not available.
(re)Sources#
Each chapter will have a list of resources, but for now I highly recommend to start listening/watching following resources on your spare time:
If I had to recommend a single book for beginner it will be this one - Hands-On Machine Learning with Scikit-Learn, Keras, and TensorFlow, 2nd Edition.