Feedback on the course
0 views
Skip to first unread message
Tim Josling
Dec 19, 2011, 8:12:26 PM12/19/11
to feed...@ai-class.com, aiclass-...@googlegroups.com
What I liked
Having tests and exams to provide feedback and to ensure you really learn the material.
The office hours.
Course covered a lot of good ground. Feel I learned a lot.
Made me very enthusiastic about AI - I will probably start a Masters in AI next year and I will certainly do something with it.
Liked the videos and descriptions of real AI projects.
Opportunities for Improvement
In the homework and exams there was no indication of what each question was worth. I have never ever sat an exam before where they did not tell you the weightings! And the weightings of the questions seemed arbitrary and capricious.
Server and software stability were very poor. Why did you have to reinvent the wheel when Andrew Ng had a full setup that worked like a charm?
Lack of programming assignments. You can't have an AI course without programming. Again Andrew Ng had it all set up and working and it added immensely to the value of the ML course.
Wild changes in the schedule messed things up for me and many others. Partly a by-product of the server issues but they could have been handled much better. You cannot just add 2 hours and think that people can use that extra 2 hours. People have other commitments and plans. I learned my lesson and did the homeworks early after the first few weeks.
Ambiguous questions. I wasted hours and hours on this - trying to work out what they were asking and what they were assuming! So many questions had hidden assumptions and gotchas. Most of the questions I got wrong were due to this (the only other one was a transcription error). Q3 in the final is a prime example. It was clear from some of the clarifications that the questions had not been thought through before being posed (eg the one where the car might go off the road and in the end PN just said "If you find yourself in the top right corner, just go Up".
Poor communication. Clarificiations were all over the place eg some random discussion on facebook for the top right corner issue above. One had to poll all over the place - though this got better during the course. There was a last minute change on what was in scope for the midterm.
It turned out for some parts of the course the textbook was essential (eg precise definitions of first order logic) even though it was not supposed to be essential.
++GOFAI. I felt the course over-emphasized old fashioned AI and lacked material on newer developments.
Course materials out late. From the time the course materials were available to when the homeworks were due was often quite a short, uncertain and variable window.
To be honest while I have gotten a lot out of the course and I appreciate the opportunity to participate and how hard they worked, I would need some reassurance that the above issues will not be repeated before I do any course by PN and/or ST in the future.
Tim Josling
Having tests and exams to provide feedback and to ensure you really learn the material.
The office hours.
Course covered a lot of good ground. Feel I learned a lot.
Made me very enthusiastic about AI - I will probably start a Masters in AI next year and I will certainly do something with it.
Liked the videos and descriptions of real AI projects.
Opportunities for Improvement
In the homework and exams there was no indication of what each question was worth. I have never ever sat an exam before where they did not tell you the weightings! And the weightings of the questions seemed arbitrary and capricious.
Server and software stability were very poor. Why did you have to reinvent the wheel when Andrew Ng had a full setup that worked like a charm?
Lack of programming assignments. You can't have an AI course without programming. Again Andrew Ng had it all set up and working and it added immensely to the value of the ML course.
Wild changes in the schedule messed things up for me and many others. Partly a by-product of the server issues but they could have been handled much better. You cannot just add 2 hours and think that people can use that extra 2 hours. People have other commitments and plans. I learned my lesson and did the homeworks early after the first few weeks.
Ambiguous questions. I wasted hours and hours on this - trying to work out what they were asking and what they were assuming! So many questions had hidden assumptions and gotchas. Most of the questions I got wrong were due to this (the only other one was a transcription error). Q3 in the final is a prime example. It was clear from some of the clarifications that the questions had not been thought through before being posed (eg the one where the car might go off the road and in the end PN just said "If you find yourself in the top right corner, just go Up".
Poor communication. Clarificiations were all over the place eg some random discussion on facebook for the top right corner issue above. One had to poll all over the place - though this got better during the course. There was a last minute change on what was in scope for the midterm.
It turned out for some parts of the course the textbook was essential (eg precise definitions of first order logic) even though it was not supposed to be essential.
++GOFAI. I felt the course over-emphasized old fashioned AI and lacked material on newer developments.
Course materials out late. From the time the course materials were available to when the homeworks were due was often quite a short, uncertain and variable window.
To be honest while I have gotten a lot out of the course and I appreciate the opportunity to participate and how hard they worked, I would need some reassurance that the above issues will not be repeated before I do any course by PN and/or ST in the future.
Tim Josling
Reply all
Reply to author
Forward
0 new messages