Subfield of Machine Learning

Deep Learning

The simplest Neural Network is called a Perceptron.

Holy crap this is good: https://nn.labml.ai/cfr/index.html

The thing that I really hate about machine learning is that these models take so much compute and are not very lightweight. I want the code to be more lightweight. it’s so annoying to have to train something, and fine-tune it. Sure, the results are nice, but, it’s like, my passion is probably not entirely writing code every time?

Two courses you must take from Krish: https://github.com/parmarkrish/Deep-Learning

https://mithi.github.io/deep-blueberry/ch3-deep-sequence-models.html

• https://course.fast.ai/ → seems really good

Deep Learning is a paradigm shift, where instead of classical methods (where features are hand-engineered), you now have learning-based methods. Features are learned by themselves → YES, found where the original idea came from, it was from this set of slides by F1TENTH.

• Classical Methods
• DL methods

Classes of ANNs

Transformer

Supervised Learning

• Neural Network
• Recurrent Neural Network
  • Long Short-Term Memory
• Convolutional Neural Network

Unsupervised Learning

• Self-Organizing Map
• Boltzmann Machine
• Autoencoder

Topics

• Neural Networks
• Gradient Descent
• Activation Function