UNNATI N.
  • Home
  • Research
  • CV
  • Blog
  • Videos
  • Contact
  • Personal

Let's Learn Together

Data-driven Marketing for Theory and Practice

learning how to learn machine learning

10/26/2017

1 Comment

 
Picture
Let's face it. The hype around big data and machine learning (ML) can be intimidating for noobs. Particularly if you have little background in statistics and/or computer science, you may find this complex territory difficult to navigate. You may wonder how to go about building your toolkit in this space. Where do you even begin? 

As a marketing major with no programming background until 3 years ago, I shared these concerns before I decided to chart out my own learning path. In this post, I will share quick and helpful tips that I have found useful for getting started in ML. Specifically, I will share my experience with three online course specializations that can be taken via Coursera.org.

Exposure, exposure, exposure:

What advice what you give to an aspiring swimmer who is afraid to get into the water? Get into the water! Surrounding myself with machine learning conversations and content helped break the ice. A few things to consider:
  1. Follow ML researchers and practitioners on social media (Twitter, blogs, podcasts, etc.)
  2. Highlight as a preferred topic in your news apps, e.g., Flipboard knows that I need to know if Dubai airport is using aquariums with built-in facial recognition
  3. Set up google alerts for machine learning -- lands right into your inbox
  4. Meet the geeks -- Use apps, such as meetup to find local tech meet-ups and seminars
  5. Sit in on a machine learning class -- my first leap into ML was to audit a computer science grad course in Machine Learning. It's a great way to find out what you don't know!

Structure, structure, structure

Assuming you are now convinced that you want to be more than just an observer to exciting ML developments, it's time to add some structure to your learning. A convenient but effective way to do this is to first find resources that already have some structure. Online courses are my go-to tool. Just like any self-guided learning, they require discipline and dedication. I have been on both sides of the coin -- both an online learner and provider. Two useful tips are to (a) set aside time each week, particularly binge during weekends, and (b) get the Coursera app on your smartphone, at least to stay on top of video lessons. 

Finally, here are three sets of online courses that I found most useful and why. Together, when taken in the order below, you can get a lot out of these courses:

  • Data science specialization (Johns Hopkins University)
    • Great first exposure to R programming
    • Great introduction to Github
    • Assignments challenging at first so set aside more hours than you think necessary, but slowly gets better; you won't learn much if you don't work on the programming assignments
    • Course forums extremely useful and peer reviews help set a frame of reference
    • Good opportunity to pick up skills less talked about elsewhere, e.g., reproducible research, and coding ethic
    • Helpful dive into statistics for non-statisticians 
  • Machine learning specialization (University of Washington)
    • Very intuitive presentation of complex concepts --- BONUS points for this!
    • Engaging video content offering a unique blend of stats and computer science perspectives; you can learn quite a bit even if you don't work on the programming assignments
    • Prepare for long hours since each course is 6-7 weeks with upto 2 hours of video content, longer than a typical course
    • Good opportunity to try Amazon EC2, Dato's Graphlab machine learning packages in python and expand related toolkit
  • Deep learning specialization (Stanford University)
    • Great first introduction to neural nets and deep learning -- you will learn to recognize cats pictures pretty early on!
    • More advanced and dense than the other two specialization above, even though shorter 3-4 week courses
    • Preliminary knowledge of ML assumed (consider taking the Stanford ML course  prior to deep learning), must know matrix algebra
    • Less intuitive videos, and requires additional deep-dive
    • Programming assignments are easier to pass, since they are well-guided python notebooks. But beware of an illusion of knowledge!

Of course, here I have assumed you have all the time in the world to take these courses for a great DIY learning experience. If you are facing a time crunch, I'd recommend the Machine learning specialization from University of Washington. Final tip from a grad student: You can apply for financial aid and save a few hundred dollars via individual course pages. This option is not available if you try to sign up for the entire specialization. Good luck! 


1 Comment

    DISCOVER

    Exciting marketing research, cutting-edge methods, and highlights from marketing conferences

    Categories

    All
    Event Highlights
    Machine Learning
    Media Mentions
    PhD Resources
    Research
    R Programming
    Teaching

    Archives

    January 2025
    May 2024
    February 2024
    January 2024
    December 2023
    June 2023
    December 2022
    August 2022
    June 2022
    December 2021
    June 2021
    January 2021
    November 2020
    September 2020
    August 2020
    June 2020
    May 2020
    August 2019
    January 2019
    June 2018
    April 2018
    February 2018
    December 2017
    October 2017
    August 2017

Powered by Create your own unique website with customizable templates.
  • Home
  • Research
  • CV
  • Blog
  • Videos
  • Contact
  • Personal