By Harry Keller
Editor, Science Education
[Note: Harry, who holds a BS in chemistry from the California Institute of Techology and a PhD in analytical chemistry from Columbia University, is sharing his first MOOC experience in this series. See part 1, 2, 4 and 5. -Editor]
November 2, 2013
The discussions should play a larger role, but I have found them to be unsatisfying. Either they’re about esoterica or trivia. Perhaps, I’m too pragmatic for this course. It’s not just about the course; it’s also about the students. It’s clear that not all students and courses go together.
I think that I’ve already gained the most important insights into machine learning from this course, how to know whether a given situation lends itself to this valuable tool. Completing the course will expand the machine learning options and my depth of understanding of how to use them.
Last week, I didn’t have the time to visit the discussion groups. This week, I don’t feel the necessity but may do so just to see what’s going on.
As I work on the fifth homework assignment, I’m not sure whether I’ve become smarter or the homework has become easier. Last week was very hurried, and I stumbled badly. This week went along nicely. There’s still plenty of mathematics, more than you might imagine, but the concepts seem more manageable.
When I began, I only knew a bit about this field from my days as a university professor when a colleague published papers about machine learning used to identify compounds in gas chromatograph tracings and spectroscopy. It was a promising area then.
Today, machine learning is used across many areas. For example, Netflix uses machine learning to predict which movies you’ll like and make suggestions. Banks use machine learning to help process credit applications. In principle, machine learning could help to identify the next lesson a student should take based on the past work done instead of relying on human judgment alone.
In this course, I am back to having to write software to complete the homework. I still find this approach a problem because your software skills become as important as your understanding of the subject. I am fortunate in having been working in software, including very advanced programming, for over half a century. So, creating programs is easier for me than for many (but not for those Caltech students taking this course). Yet, it does take time to write, debug, and validate the programs.
The big benefit of writing software comes from being forced to understand the material very well. You cannot program what you do not understand. Right now, I’m faced with having to write a program to process stochastic gradient descent and am discovering that I don’t understand it as well as I thought I did.
I wasn’t able to write, test, and debug the second program required for the assignment in time and so had to skip two questions. I missed one other question. Not too bad. Actually, I was in such a hurry that I entered one answer in the wrong place. To make up for that, I guessed on the two questions I was unable to finish the software for. With five answers, my odds of getting one guess right were 46%, and I did get one guess right, which made up for my incorrect entry. Note to self: take more time when entering answers even when rushed!
I still miss being able to ask questions during the lecture — in person. I didn’t use office hours much as a student because they were generally few and crowded. I’d really like to have a discussion of just a few minutes about some aspects of this subject that I’m curious about.
I plan to go on for one more week and take it one week at a time. Who knows? I may find myself at the end of the course without even realizing that I made it there. I certainly hope so because last week was a killer. Another such week would probably result in me dropping out.
Filed under: MOOC |