Research Methodology for Machine Learning Projects

Yesterday, I delivered a talk on "State-of-the-art Research Methodology for Machine Learning Projects". Here is a gist of important points:

1. Exploratory research in ML and DL is extremely valuable.

2. Focus of current research is on interpretability, generalizability and ability to solve non-convex and non-smooth functions.

3. A good mathematical foundation is extremely valuable.

4. Delineating the scope of your project during initial phases is very important

5. ML/DL projects must be metrics driven.

6. Life cycle of a research project in ML/DL : Exploratory research, initial solutions, refinements and evaluation, deployment.

7. Skills like ability to read and assimilate the literature, analysis, theorem and proof formulation, algorithm design, coding, communication skills (writing research articles, giving talks, presenting posters) are extremely valuable.

8. Cross validate your results, compare with baseline measurements, prevent data leakage.

This document from Tom Dietterich is an excellent guide.

https://web.engr.oregonstate.edu/~tgd/talks/new-in-ml-2019.pdf

#ml #projects #researchmethodology #deeplearningai