How can I get research experience outside the college setting?

How do I get a good PhD/Master's of Science admission in the field of machine learning even though I don't have a formal degree or formal research experience in the same?

• I graduated a couple of years back with a Bachelor's degree in Electronics from a top 10 university in India with an above average CGPA. I took up a pretty good job with a world-renowned MNC in the semiconductor field. But after about 3 and half years of working, I got pretty bored and wanted to move out. I came across Coursera's "Machine Learning" course and loved it. So I decided to pursue this relatively new and exciting field. As luck would have it, my division got down-sized and I was let go. Instead of taking another job, I studied for GRE/TOEFL and get a research internship in Machine Learning. I've done well in GRE and TOEFL but I'm still without a job or an internship. I'm studying Machine Learning from textbooks and papers and looking for an internship. I'd really like to hear your views on what I should do next. - The way I see it, I should concentrate on getting a research internship. I don't have any publications, just a couple of projects(unrelated to machine learning). - Also I don't have anyone relevant who would write me letters of recommendation as far as machine learning is concerned. And I need 3 such letters! - Another issue is that I haven't actually studied machine learning or computer science courses at college. Though my previous job involved a fair bit of C and assembly language coding. - I could take up a job in the numerous analytics firms that have mushroomed in India, but I wouldn't be able to do any real research work not get any academic letters of recommendation. Your views are most welcome.

• Answer:

  If you are familiar with the concepts,start implementing them, make your own codes, machine learning is also about making efficient codes. So take up a problem in the domain, try to solve it, try to get publications. Theses publications will help you to get there. :) At the beginning it will be always better to start with some known problems and known solutions. For example. I can give you a problem. It not simple. But approachable, Make a chess game which plays with itself and gain proficiency to defeat a human player. For some simple problems I may suggest something in image processing or something.