Welcome to ML Cookbook

Frequently Asked Questions

Find quick answers to common questions

• How do I start AI/ML/DL?

  There are steps that we think are wise to take:

  • Step 1: Learn Math (incl. Lineal ALgebra), Inferential Statistics, and basic of Programming (Python).
  • Step 2: Take some beginner classes, e.g. try out Ng’s Machine Learning on Coursera.
  • Step 3: Find some problem/projects to play with. Kaggle has tons of such tasks. We do too.

  Iterate the above 3 steps until you become bored. From time to time you can share what you learn :)

• What is your recommended first class for ML?
  • Ng’s Introduction to Machine Learning on Coursera
  • the CalTech edX class
  • the UW Coursera class
• What is the prerequisites of Machine Learning/Deep Learning?

  Mostly Linear Algebra and Calculus I-III. In Linear Algebra, you should be good at eigenvectors and matrix operation. In Calculus, you should be quite comfortable with differentiation. You might also want to have a primer on matrix differentiation before you start because it’s a topic which is seldom touched in an undergraduate curriculum.

  Some people will also argue Topology as important and having a Physics and Biology background could help. But they are not crucial to start.

• What is the best/most recommended language in Deep Learning/AI?

  This is one of the most frequently asked questions and it re-emerged from time to time. So here are few couple of useful point of view:

  In general, if you are working on computer, you will need to switch between languages from time to time. So arguing about language and which is better is not a wise thing.

  If we want to give the best advice, perhaps the answer is you should just learn as many of them as possible. Of course, your time is limited. So we usually recommend python as your first choice. Python is cited as a good language for AI/DL because it has the best support of libraries. Most ML libraries from python links with C/C++. So you get the best of both flexibility and speed.

  Other also cites Java (deeplearning4j), Lua (Torch), Lisp, Golang, R. It really depends on your purpose. Practical concerns such as code integration, your familiarity with a language usually dictates your choice. R deserves special mention because it was widely used in some brother fields such as data science and it is gaining popularity. Finally, there is Matlab/Octave, it is also useful to learn because many Coursera ML’s class is Matlab-based. e.g. Ng’s class and Hinton’s class. Sure, you can finish them in python, but there are much re-implementation in the process.

• What is the difference between Deep Learning and Machine Learning?

  (source: Arthur Chan)

  Deep Learning is Just a Subset of Machine Learning

  First of all, deep learning is just a subset of techniques of machine learning. You may heard from many “Deep Learning Consultants”-type: “deep learning is completely different from from Machine Learning”. But then when we are talking about “deep learning” these days, we are really talking about “neural networks which has more than one layer”. Since neural network is just one type of ML techniques, it doesn’t make any sense to call DL as “different” from ML. It might work for marketing purpose, but the thought was clearly misleading.

  Deep Learning is a Representation Learning

  So now we know that deep learning is a kind of machine learning. We still can’t quite answer why it is special. So let’s be more specific, deep learning is a kind of representation learning. What is representation learning? Representation learning is an opposite of another school of thought/practice: feature engineering. In feature engineering, humans are supposed to hand-craft features to make machine works better. If you Kaggle before, this should be obvious to you, sometimes you just want to manipulate the raw inputs and create new feature to represent your data.

  Yet in some domains which involve high-dimensional data such as images, speech or text, hand-crafting feature was found to be very difficult. e.g. Using HOG type of approaches to do computer vision usually takes a 4-5 years of a PhD student. So here we come back to representation learning - can computer automatically learn good features? Spoiler alert: yes.