#What exactly are the things I should learn from any AI course?

4 messages · Page 1 of 1 (latest)

stoic pier
#

High Level:

  • How to compute gradients/how gradient descent works
  • How foundation ML models (MLP, trees/forests, boosting/bagging, KNN) work
  • How does back propagation work
  • The difference between core neural network layers (MLP, CNN, RNN, Transformer/Attention)
  • How to perform data processing (balance labeled data)
  • How to train models (avoid overfitting/underfitting)
  • DSA (data structures & algorithms)

Application:

  • How to use key modules like sklearn, pytorch/tensorflow, etc
  • How to build neural networks
  • When to use each technique (classical ML vs deep learning vs algorithms)
  • Intermediate comfort with using Python (and programming in general)
dense mural
#

memorizing the equations is not really useful for RL, You need to understand the functions. Example: without full understanding of Bellman, it will be hard to move pass the K-armed bandit problem. And you need to understand the the idea of "looking forward" which David Sliver explains really well in the deepmind course.

For "self judge" metrics, i reckon able to build a working RL model from scratch is probably a good indication you are learning. Many of the questions in Sutton's book are about understanding rather than memorizing, hence, able to solve them is also a good metrics on your learning progress.

scarlet cargo
#

I'm working on NBA prediction module and I want to team up with someone

hasty quiver