A Caltech Grad in a Caltech MOOC

picture of Harry KellerBy 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, has been sharing his first MOOC experience as comments to Jim’s “Technology in Higher Ed: We Ain’t Seen Nothin’ Yet” (10/3/13), but they’ve quickly grown into a series that we’ll be publishing on a loose schedule. See part 2, 3, 4 and 5. -Editor]

October 3, 2013

Future MOOCs may or may not be termed “MOOCs.” Things are reaching the point where small operations may change the world. Just look at the impact that Salman Khan had on education.

The Kepler project in Rwanda is an example of being a bit more creative by taking MOOCs and combining them with in-person teachers to deliver high-quality education. After all, they’re free!

How this resource comes to be used will affect how it evolves.

Another factor will be adaptive learning options and more interactivity.

Opening screen for "Lecture 1: The Learning Problem Free," from Caltech Professor Yaser Abu-Mostaf's free introductory Machine Learning online course (MOOC).

Opening screen for “Lecture 1: The Learning Problem,” from Caltech Professor Yaser Abu-Mostaf’s free introductory Machine Learning online course (MOOC).

I’m currently taking my first MOOC, given by my alma mater, just to learn something new (machine learning — haha) and learn about MOOCs first-hand. So far, it’s nothing very exciting, but I haven’t bothered with any of the discussion group stuff because I just don’t have time for it. I may not have time to complete the course, but at least I’ll have learned SOMETHING and experienced it.

October 6, 2013

Follow up on my MOOC — I handed in the first homework assignment. I tried to do the last problem (requiring writing software) the hard way (by quantitative analysis) and decided that it would just take too long and settled for an alternative approach (Monte Carlo method), which only took a few minutes to program and run.

Professor Yaser Abu-Mostafa

Professor Yaser Abu-Mostafa

My first homework grade = 10/10. I took a look at the discussion group after I finished my homework to see what sort of questions and answers were being posted. I guess I’m rather biased from having been a Caltech student and having done essentially all homework solo. I think figuring out how to do it is as important (maybe more so) than doing it. 

In any event, I was disappointed by the questions because the people asking them clearly hadn’t spent much time or mental effort (or both) on the problem before running for help.

Nevertheless, I am not being judgmental here, but noting a fact. Many people come to MOOCs for many reasons. Some of those reasons may well lead to such questions.

The professor is also monitoring the discussions (maybe by proxy — don’t know) and will help out if need be. Still, a question came up in my mind in one of the two lectures so far that I would have liked to have asked of the professor. So, I’m left with an unanswered question. That’s not very satisfactory to me but hardly a course killer.

My purpose here is to understand machine learning well enough to decide if it can play a role in my company’s work (don’t know yet) and, if so, exactly how to apply it.

I also like the relaxation that comes from doing something engaging and completely different for a change.

October 13, 2013

The discussion group is becoming more interesting. I’m guessing that some students are dropping out and leaving behind the ones who “get it” better. Questions have ranged from clueless to banal to intelligent. The second week had fewer of the clueless questions.

This MOOC is very technical — machine learning. It may sound prosaic, but it requires knowledge of matrix algebra and calculus for the theory and computer programming skills for the homework.

I am confident that those who complete [this MOOC] will have a good grounding in the principles and application of machine learning.

The homework questions are quite penetrating. They have been cleverly constructed to require understanding of the material rather than just parroting the lectures and notes. I was used to this sort of thing as an undergraduate but haven’t seen it for many decades. Well, it’s a Caltech course. What else should you expect?

I’ve been scanning the discussions but not yet participating. The questions in the homework are generally terse and assume quite a bit of knowledge as well as paying attention to the lectures. There is no textbook. It’s not really surprising that quite a bit of discussion is generated.

This is a ten-week course and is moving along rather rapidly. Given the pace, I am confident that those who complete it will have a good grounding in the principles and application of machine learning. The professor knows how to pace his lectures and explain more difficult concepts thoroughly. The Q&A after each lecture shows that he knows this subject very deeply.

Given my reasons for taking this course and the nature of the material, this format works for me — so far. It mirrors my undergraduate experience.

October 14, 2013

I’ll primarily comment on [Bett’s] first . . . paragraph.

Michael Bett on October 10, 2013: MOOCs are great in that cheaply make knowledge widely available to those who may not have the inclination or desire to study otherwise. Discussion groups attempt to fill the gap for personalized learning, in a very inefficient way. In my opinion, they are useful if accompanied by localized or small group recitations, but the knowing what the students are learning is difficult.

Those without the “inclination or desire to study” will be the worst MOOC students. Taking a MOOC requires some real self-discipline.

Discussion groups succeed or fail in this venue based on many factors, including instructor participation and quality of students.

Small groups led by an expert, whether F2F or remote synchronous, can raise a MOOC to the standard of a “real” course.

The lectures are paced so that I can think that I understand everything. Then, I go to the homework and find out that my understanding is shallow.

As a general rule, improving the personalization of a course can lead to greater learning gains. We don’t require studies to prove that.

Whether it’s expert-led small groups or review of student work to tailor material to the student, a truly good learning experience costs more than just a server and bandwidth.

The MOOC I’m taking now may be “Introduction to,” but it’s not an entry-level course. You have to be ready to learn, both in attitude and in prior knowledge. The discussion group material shows that some are not. They’re having difficulty with concepts that are prerequisites for this course.

I’ve been through four lectures and two homework assignments so far (out of twenty total lectures). I’m learning a great deal. The lectures are paced so that I can think that I understand everything. Then, I go to the homework and find out that my understanding is shallow. This course moves along at a rapid but not supersonic speed, but if you miss one week, you’ll have trouble making it up. Right now, my goal is making it to the mid-term — through week 5 and lecture 10. Then, I’ll decide whether to complete the course, which I am only auditing.

I already have enough degrees and have taken enough courses for credit for any single person. I’m interested in having a real working knowledge of machine learning. The next two weeks are theory. Hope I survive them.

4 Responses

  1. […] Future MOOCs may or may not be termed “MOOCs.” Things are reaching the point where small operations may change the world. Just look at the impact that Salman Khan had on education.  […]

  2. The third week of this MOOC unveiled to me a problem with MOOCs. Not every student is a “student.” I am not really a student. I have conflicting priorities that few students have. Sure, some have jobs and family responsibilities. Generally speaking, they can leave their jobs behind when they leave their jobs.

    As an entrepreneur, I am “on call” 24/7. I have to “steal” a few hours a week for this MOOC. I have a goal that may help my business — or may not. The MOOC has to be a low priority. I have no way of knowing when I can watch the lectures or do the homework beforehand.

    This has been a very busy week. I already have two full days committed next week. By full, I mean from arising to falling asleep, not a mere 8 or 9 or 10 hours. There will be not a minute for MOOCs.

    Last week, I managed 8/10 on my homework due to not paying attention when I encoded my answers as (a) through (e). This week I paid attention to that but did not have time to analyze my answers as fully as I would have liked. Instead of the recommended ten hours spent on homework, I spent about one. My 7/10 on the homework reflected this less deep thinking.

    One problem I missed had me determine the largest number of points that can be shattered by a planar triangle learning model. I really thought that i understood this problem but clearly did not. I don’t really have the time to figure out why.

    The last problem in the set required finding the growth function for a planar annulus training model. I was able to find the answer quite nicely and got it right.

    The good part about this homework is that I did not have to write any software. The two previous homework sets required considerable amounts of programming. You could make a mistake in understanding or in coding. Having two modes of error made the exercises more stressful than usual.

    Also, I was able to spend time reviewing the student discussions on the last homework but not on this one.

    If you cannot set aside a definite and adequate number of hours every week, you may not be able to achieve all you wish with your MOOC. I may struggle through to the end or not. I’ve made it 33% of the way with a grade average on homework of 83, which gives me an A- so far.

    I will do my best to make it through two more weeks and then reevaluate my commitment to completion at the mid-point.

  3. Week #4 = lots more theory. This week I had meetings and deadlines associated with my business. I did my best to look at every homework problem and at least understand what it was asking. I didn’t have time to analyze every question and didn’t answer one at all.

    If you are taking a MOOC with time constraints, make sure that you’ll always have the time. This week, I barely was able to finish watching the lectures.

    Some of the questions required some deep thinking. That’s good — if you have time for it. It’s also typical of Caltech.

    The problem all of the question sets is that you’ll never know how to solve the questions because you’re only told the correct answer. If you have lots of time, you can keep working on last week’s problem set until your answer matches the correct one. I wish I had that time.

    The discussion group seemed oddly quiet this week. I’m guessing that last week’s theory-heavy material caused many students to drop out. This week was even more abstract and theoretical with even more terminology to master.

    Depending on online, typed discussion groups to help out with understanding is not a winning strategy in this course. The professor clearly is an expert with much more than the usual expertise. He lectures carefully and paces his lectures well. His slides are nicely done and available throughout the course.

    What’s missing in a course that’s this challenging is direct interaction with someone who understands the material or, in my case, about a dozen more hours to review and digest it.

    This week was the worst so far for me. I get to drop two homework grades if I’m taking the course for credit, which I am not. This week would be one of those grades. I’m more concerned with my level of understanding. Maybe I’ll catch up in future weeks. Maybe not. Clearly, I must have more than four hours a week for this course. Next week looks more open, and hopefully I’ll not lose too much time due to loss of time this week.

    I promised myself (and you) that I’d finish five weeks. One more week to go before a serious evaluation of whether to continue. The material is interesting in the abstract, but the applications are missing these two weeks. That’s supposed to change next week. We’ll see.

  4. At lecture 9, things are getting more practical and less theoretical but still heavily mathematical. I have to wonder how the homework will be this time.

    There may be many styles of MOOC. This particular one is two lectures, with audience questions after each, followed by a ten-question homework set. Each question has five possible answers. You choose one. As soon as you submit the answer to one question, you get scored. Answer keys are supplied after the due date. Then, the next two lectures and homework set are provided — on a weekly basis.

    Discussion groups may help students with problems if they can get answers.

    The volume of discussion quickly diminished as the course progressed. I have to assume that, as the course was getting more difficult, the number of attendees declined rapidly.

    To my mind, this is the minimum that an online course can be. It’s just the F2F course without the help from professor office hours or asking questions yourself but with the benefit of flexible time and low cost if you aren’t seeking a degree.

    I had forgotten just how passive lectures are. You sit. You listen. You may take notes but don’t have to here because you can always review the slides and play any part of the lectures.

    Engagement comes only through the homework and discussion groups. The dwindling discussion groups provide less and less interaction. The homework interaction is mostly in my own mind.

    In short, this is a great course in a poor medium.

    The above does not condemn MOOCs. Rather it suggests that old-fashioned modes of presenting courses converted to online delivery are not the best way to go. IMO, they’re worse than the original.

    The Internet is being used just to stream lectures, provide background materials, and deliver multiple-choice, automatically scored tests. It can do so much more that this sort of thing borders on criminal. Caltech, the home of some of the smartest minds (professors, graduate students, and undergraduate students) in the world, could do much more.

    This could be a “learning to mastery” course, for example. That would mean a great deal more effort on the course developers but also a much higher completion and success rate. Instead, it reverts to the Caltech norm, which I had experienced first-hand in the ancient days (middle of last century) when I was a student there.

    Students rarely mastered any course taken at Caltech while taking it. Sometime later, it finally sank in — if they were fortunate enough. The amount of material, the complexity and depth of the material, and the pace made mastery all but impossible. Some students managed it but most did not.

    I remember my freshman physics course was difficult. I did get an A grade, but was fuzzy on much of the material. My mathematics aptitude saw me through. Next year, as I was taking second-year physics, some freshmen asked for help on an assignment. I could see that it was one that I had had trouble with the previous year. Yet, suddenly everything was clear. It was easy for me, the same stuff that was so hard before. This is the norm at schools such as Caltech and MIT (where my daughter went to school).

    It’s past time to evolve away from the “Khan Academy” model of MOOCs and build something much greater. In some subjects, you may be able to rely on student interaction. In others, such as this machine learning course, that is just not the case.

    This is a graduate-level course using rather high-level mathematics and all of the symbology you can imagine: set intersections, sums, expectation values, integrals, probability expressions, and so on.

    Many have suggested, even at the recent conference I attended, that online courses are harder than their F2F counterparts. That’s generally true for a number of reasons. However, it does not have to be so, IMO. It’s just easier that way.

    It’s the same thing with testing. This course has test questions that cannot be answered quickly and that require true understanding to answer correctly, even though they are multiple-choice. Most test questions lie at the opposite end of the scale. You only have to memorize and may have to perform some perfunctory analysis (in science, that’s ‘plug-and-chug’) to succeed.

    The simple test questions take less time and cost less to create, which is why they’re so prevalent. Similarly, the lecture-and-quiz MOOC is the cheapest to create. Let’s hope that tools will be made that can break down the barrier to exciting, engaging MOOC-style courses.

    When a course is over, you should know that you have learned to do something, and you learn by practicing, not by listening. Even Aristotle understood this: “For the things we have to learn before we can do them, we learn by doing them.”

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s