Introduction

A few months ago, I began teaching myself machine learning. While I was feeling confused and didn’t know how to start, I happened to see a Youtube recommended video about a lecture of ML career advice by Andrew Ng (great job again, Youtube!). As a beginner, I found Andrew’s advice very helpful. Whenever I feel a bit lost on my journey of studying ML, I would come back and read my lecture notes. I can get new inspirations every time! Therefore, I moved my notes from my personal notebook to here. If you happened to see this blog, I hope you’re benefiting from it as well.

How To Read ML Papers

Multiple Passes:

1. Title/abstract: big picture
2. Intro + conclusions +figures + skim rest (skip related work)
3. Read but skip/skim math (or read it later)
4. Whole thing, but skip parts that don’t make sense

Questions to Ask Yourself:

• what did authors try to accomplish?
• What were the key elements of the approach?
• What can you use yourself?
• What other references do you want to follow?

Sources of Papers:

• Twitter
• ML subreddit
• NeurIPS/ICML, ICLR
• Friends

Nice Conferences in AI:

• ICML

• NeurIPS

• UAI

• KDD

• AAAI

• IJCAI

• COLT

• AISTATS

• ECML
• ACM on RecSys
• MSML

Math

• Read though it, re-derive from scratch

Code

• Download the open source code
• Reimplement algorithms from scratch

Steady reading, do not burst, long-term learning, learn a bit every day

---

Career Advice

Goal: (Do Important Work)

• Job (big company or start up)
• PhD

Two Steps:

1. How to get a position
2. Set up a position

(Job Senario) Recruiters are looking for:

• Skills
  • ML quizzes
  • coding ability
• Meaningful work
  • Not just know the theory, but can show how to use it

Area:

Wide knowledge and deep understanding (in at least one area)

What to do:

• Pick one of these area to work on a project: NN, BN, CNN, RNN

• Know how to make it work

Bad way to navigate career:

• Take tons of courses but not go into depth or go into only one deeply (jump into research papers) without many experience – not very productive until take classes and master basics to understand advanced projects.
• Do many (easy) projects in different areas – volume of projects is not important, 10 lame projects is not better than 1 great project. Recruters are trying to understand how good you are through the projects.

Building Foundation Skills:

Saturday morning problem:

• Read paper (do not work too hard, do this consistantly)
  • Read 1 paper a week –> 50 paper a year
• Work on projects
• Open source
• TV

---

Selecting A Job

• Working with great people and projects, they will influence you
• Focus on the team (10-30 people) you’ll work with
• Manager
• Do not focus on “brand”

Ideal: Giant company with 50000 employees, a 300 people AI team or 30 people team is a great offer, because you know what exactly you’ll do and talk to manager, rather than being assigned to a boring job.

Rotation programs (NO!):

Be careful of the job offer that refuses to tell you that team you’ll join and what job you’ll do, because it might be boring and they don’t want you to know.

What is OK: A company with 10000 employees that is not in AI world but has a 10 elite AI team. If you have a job offer from this team, take it.

• Learn the most
• Do important work