Saturday 23 February 2013

Hacking my education- Machine learning Endeavour



Originally posted on posterous (Oct 9 2012) which is shutting down.
I have had an interest in machine learning since a few years. In spite of the numerous resources available online, I had never really gotten to a system of organized study due to various reasons.
With the advent of Coursera and other MOOCs, it has become relatively easy to gain knowledge in a particular field of study, with all the advantages of a classroom-like environment (great teachers, a well-rounded and structured syllabus and motivated fellow students) without the attendant disadvantages (cost, distance and time).
When I started on this endeavour in Dec 2011, I did not think I would get this far. Having left college 12 years ago, it was quite an effort to get starting studying ML, and I was told that my getting into machine learning would not be easy because of the math prerequisites. However, close to 10 months down the line, I realize that nothing is impossible.
So this is my attempt to “enlist the goals” in my machine learning endeavour and chart my progress so far.  I’ve checked the syllabi of a few master’s courses in Machine Learning, and here are some of the courses that are required to be taken.  
I plan to finish at least 5 courses related to ML before May 31st 2013, from the ones in this list:
  1. Probabilistic  Graphical Models @ Coursera
  2. Machine Learning @ Coursera
  3. Neural Networks @ Coursera
  4. Computing for Data Analysis (a course on the R language) @ Coursera
  5. Image and Video processing @ Coursera
  6. Natural Language processing @ Coursera
  7. Optimization @ Coursera
  8. Statistics 101 @ Udacity.
Real world machine learning is more than just taking courses, it should involve working on some projects. Since I’m mostly self studying, it would be difficult to find a proctored project to work on. However, there’s Kaggle(a site that hosts Machine learning competitions)  to the rescue. My goal is to compete and finish at least 4 competitions in the top 50% of the field.
My progress so far:
  1. A course in Linear Algebra @ GuruPrevails-completed Jan 2012
  2. Probability & Statistics @ GuruPrevails-completed Mar 2012.
  3. Probabilistic  Graphical Models @ Coursera – completed successfully in June 2012
  4. Machine Learning @ Coursera – in progress, ends Oct 2012.
  5. Neural Networks @ Coursera – in progress, ends Dec 2012
  6. Computing for Data Analysis (a course on the R language) @ Coursera – in progress, ends Oct 2012.
  7. Algorithms – Design and Analysis @ Coursera – didn’t complete beyond 4thassignment.
  8. Kaggle digit recognizer “learning” competition: currently 23rd/400 teams (Oct 2012).
Along the way, I’ve realized that my learning methods are sub-optimal, there are other ways to learn than just ‘being motivated and working hard’. I signed up for the “learn more study less” course by Scott H Young, and although it is too early to see any results, the course material is certainly worth it.
I’m off to a much needed break to Roopkund (it’s been 4 years since I hiked in the Himalayas L) , and its back to the Machine Learning Endeavour after that!

No comments: