Student Comments for Machine Learning (50.007), SUTD, Sep-Dec 2014
The instructor is very knowledgeable in the subject and very dedicated in his
teaching. It has been an honour to be in his class.
I am impressed at how flexible Stanley was in responding to our feedback. He
chose to switch teaching methods halfway through the course - from
teaching from slides to teaching from the board. Learning from the
board was much better, as the pace of the class was slower and more
reasonable. Stanley is a very patient teacher who is always willing
to be interrupted during class to answer questions - yet he is able
to stay on track. He is good at breaking down complex concepts, and
he's able to explain them in good English. He's an excellent teacher
who remembers all of us by name and probably the only one of my
teachers who can get all of us talking and asking questions in
class. Sorry I should also have said this earlier in the course
feedback: Something to improve is to separate the coding component
and written component from homework, and to move the coding to mini
projects. I think it is easier to tackle learning overall in this
manner - homework is for consolidation, while projects is for active
application unto real life. I also think that it is much better to
have consistent quizzes (perhaps bi-weekly?) instead of two huge
exams with such huge weightage. It takes the stress off from taking
one-off exams, and also helps us consolidate the material more
consistently over time.
Generally much improved pace in the 2nd half. Liked slower teaching, and the HMM
materials are quite good quality.But honestly, I still prefer
slides. Perhaps something like Caltech's Machine
Learning. Suggestion: for future ML courses... all lectures should
be recorded... sometimes I really miss out certain key points when
Stanley teaches slightly too fast. A video can allow me to rewind
and revise better.E-dimension notes can be clearer.
I preferred his earlier method of teaching with slides, as it gives us something
concrete to refer to after class (based on what was covered in
class, not the notes). Otherwise, he is still generally a good
powerpoint > whiteboard
Probably the most challenging course in my time here at SUTD, because of the
nature of the subject. But worthwhile! If possible, more time could
have been given for us to digest each topic, but we are constrained
by the short term time.
Some of the homework questions are really very difficult. =(
it's too hard i can't copy and listen at the same time i do not know what is going on
Very proficient in teaching, Stanley have been an awesome teacher in class. He
explains theories very clear and he doesnt hesitate to always ask us
to ask more questions. Great Job!
Can there be more concrete examples during lecture
Sometimes the explanation feels rushed. It is hard to ask questions and clarify
when we don't know what we don't understand.
Instructor is dedicated and knows the subject well.
Slides are comprehensive, Lessons are made interesting with various analogies.
Should explain more on the algorithms used for the homework. The problem with
Machine Learning is that every different paper have their different
set of conventions, and that algorithms are often skimped over from
a mathematical perspective without proper considerations to the
pratical performance (e.g. how to make the algorithm run fast using
proper data structures, etc.)
Stanley tries to teach the material the best he can, which I appreciate since
the content is rather difficult in itself. The use of slides is
definitely appreciated, even small touches like printed handouts,
and asking students repeatedly to clear their doubts. The slides can
probably still be improved though - e.g. more
informative/explanatory as opposed to just brief points. The slides
also become quite small when printed 6 per page - perhaps 4 per page
would be a better size. In all, good effort in teaching, do
The content is well structured and understandable
too much content maybe? feel as if we're rushing onto the next topic without
understanding what we've just learntI hope that they will teach or
help a bit more on the programming portion for ESD students.
More application would be great. A project would be time consuming but it will
really help us know what we need to know if we were to implement
some learning system in the future.
I feel like doing more programming assignments. But argue questions aslo helped
me understand the problem.
it is too hard
Very subjective. 2hard4me
This course alone is excellent. However, I still feel that the content of the
course is very intense. Especially since there is alot of math to
substantiate the theory. Being in the ISTD pillar, I know myself to
be very weak at math. This course, coupled with other heavy course
from other courses will make a student's workload difficult to
Perhaps an additional reference material (if the topic is outside of the lecture
that we did not already know) would be beneficial, especially for
students who need to derive and verify the math to the very core
It's too difficult.
Make big programming questions (Like ID3 Decision Tree) Pair Work :) It will
make the course load a lot more manageable while still ensuring we
learn from implementing the code.
It's really interesting, but the programming parts are harder to grasps as when
trying to program it, we realise there is so many things to tune and
unknowns to calculate.
I feel that they are trying to teach us many Machine Learning Models in a very
short period of time. It can be better if they choose to teach
lesser models, but with deeper and more thorough explanation of each
I appreciate how Stanley always tries to engage all of us in class. I think that
his approachability and willingness to entertain questions has led
to more of us speaking up and asking questions. I do think that
sometimes Stanley speeds through certain concepts; he should
elaborate on them more. One example is the lesson where he explained
the backpropagation algorithm - more time should have been spent on
working out the math instead of showing us videos on the ALVIN cars.
It also feels like we're going at a pretty fast pace. Before we
fully digest a new machine learning method, we are moving on to the
next already. There is no way I can keep up with the course without
going back to read on my own. I do hope the pace will slow down a
little more and more time can be spent on explanations for each
machine learning methods.
Course material is rather difficult. Perhaps it could be good, if possible, to
provide a greater intuition/more detailed description of the math,
easing more into it as opposed to jumping into it straight
away. This can help in the understanding of the material.
Machine learning has been an interesting lesson! It has really broadened my
perspective on how machines and code are used in classifying huge
amount of information and how applicable this can be in all parts of
i think this course is very interesting and useful. however, as i came from ESD
background, i would like to suggest the contents can be more
considerable designed. currently this course ( also homeworks) are
more likely to be designed for ISTD students. i had a chance taking
MOOC machine learning from Stanford university and their course is
designed to students from varies backgrounds. their course are not
so technical and the course works are more theoretically
based. since machine learning in our school is also open to students
from EPD and ESD, it would be nice if the course can be more
As a student who did not have long exposure to computing, programming the ID3
algorithm took me at least 10 hours without any results
It seems to be too difficult for ESD students (especially for the programming
part) Hopefully if we put some efforts in it, we will get A.
The coding is abit difficult
Hugely disappointed by this course. Steep learning curve, no consideration for
students' abilities given the other workload of other courses and
programming skills. Concepts are interesting but more effort should
be made for students to UNDERSTAND the course materials with DIRECT
APPLICATION on the homeworks sets. This course is also extremely
unfair and biased.
SUTD seems to be trying to fit a lot of content into just 14 weeks, unlike other machine learning courses I have seen online. It's not good for learning in my opinion.There seems to be no overview, rather like a meandering through different topics. Probably should start from what we already knew, like regression, from the first week, then progress on from there
Although this is a very difficult course, this is a very interesting course.
Last modified: Sun Jan 4 23:11:56 SGT 2015