Course given 2006/2007 in study period 1
Additional comments about the book:
Good, although we only learnt a part of it.
---
The book is quite good, but a little too lengthy now and then.
---
Incredibly expensive, and not really necessary to follow the
course.
---
If according to you the book is useful you perhaps should tell
the
students which chapters or pages are the most important for the course
and which one are in direct connection with such a lecture.
---
Intressant och relativt lättläst bok i ämnet.
---
I found the book very helpful throughout the course, especially
during the probability section.
---
boken var bra. Lättförståelig text med bra
förklaringar och lagom pratig. Kul med skämt då och
då ;-)
---
Boken passade föreläsningarna, eller rättare
sagt, det kändes mer
som om de tidigare föreläsningarna passade boken väldigt
bra, vilket
gjorde det lätt att se vilka delar av boken som var mer relevanta
för
den här kursen. Som vanligt innehöll den mycket mer och var
mycket
djupare än vad vi hann gå in på.
---
A too big book for this short period. Probably shortened
summaries would be better. It is also very expensive.
---
Because of the interesting contents of the course, I will buy
the book to have as a reference for future work.
---
One of the best course books I've had...
---
Not a great boook. I agree it covers a lot of topics inthe
course
and maybe that's why it was chosen, but still it uses too many words to
explain simple things and that too not-too-well.
Additional comments - please, give us constructive feedback:
Good, could've been nice with more demonstrations etc. like
in the last lectures.
---
Only attended Patrics lectures, but they were fine.
---
Sometimes there was too many slides
---
I find that there was a break between the courses given by the
both
teachers. The link between them was not enough clear according to me.
---
Danica, dina föreläsningar om HMM var jättebra,
det är svårt men jag tyckte du förklarade det bra.
---
Patric has a tendency to re-explain what was said in the book,
using
the same examples. While this is helpful to understand the book, it
would be nice to not have the concepts explained twice in the same way.
Also, we spent at least the
equivalent of a full lecture going over the problems that individual
students were having. It would have been more beneficial to take care
of those issues outside of lecture, instead of wasting the time of the
other thirty students who had different problems.
Both instructors were excellent when presenting material directly
related to their fields (robotics and vision).
---
Most was good from Patric, got nothing to comment there really.
(read about your fair-drawing robot in ny teknik, good luck with that!)
Danica
did many things good also, but should have laid up the HMM lectures
better. The subject should have had more examples and explanations of
different things HMMs are used for. Why would we want to calculate the
normalized joint probability of jointed probabilities jointed together,
i never really understood during the lectures?? Give examples but more
importantly explain why its interresting and useful.., dont throw
formulas as the forward/backward algorithm as a solution to a problem
that most (at least a few more than myself) didnt understand and/or why
it should be solved.
Also, felt like a bit of wasted time with
lots of talk about homework 3. I agree it needed more explanation
but.., felt like it took away two lectures that could have been spent
better.
---
The lectures about HMM weren't very good. Danica spent to much
time
covering the basics (markov models) and asking us if we understood -
which at least I did - and thus lost focus on the relevant material.
It
would have, IMO, been more appropriate for her just to assume we know
the basics and carry on with the material novel to us, i.e. the actual
course content.
Futhermore, I believe she spent far too much
time covering the homework (homework 3) - questions and unclarities
about it would have been much better handeled individually or posted on
the web site (personally, I hadn't started on the homework by the time
she spent more than an hour talking about it - thus any benefit for me
was forfeit and the lecture time wasted).
---
Teaching was quite good overall and easy to follow. A lot of
effort was put in by the teacher which showed.
One
comment: I felt Danika was getting too defensive about the fact that
she was teaching us a 'difficult' topic and spent a lot of lecture time
catering to rather individual problems and justifying stuff. Yes it was
slightly tough, but not THAT tough, and I felt students are lazy bums
anyway, so if you want us to learn and really do something, you have to
start by being stern and definite about it from the very start. I
really admire her for her patience, flexibility and all the help she
offered all the time, and I appreciate that since it was also needed.
just that I'd wish she was bit more stern, though equally helpful, in
the class. It helps students sit up and take things seriously from the
beginning. We are in the class to learn, and learn an important thing.
If it's tough, so what? We are responsible to put in proportional
effort.
Comments about homework 1:
The simplest homework. It was the direct application of the
course.
---
Some really ambigious questions. generally, HW1 is reasonable.
---
This was a good and not too hard introduction to agent design
and basic algorithms involved. Great as a first homework.
---
Not as interesting as hw 2 and 3
---
Easy, interesting.
---
it is okay to warm up :)
Comments about homework 2:
Nice to actually get to implement the stuff you've learnt.
---
More stronger than the first. Very interesting, above all the
last
part which demanded a real involvement for the student, but relaively
simple to solve thanks to the help that we can find on the internet,
because it was classic subjects.
---
Many ambigious questions in this one. Generally, HW2 is
considered
rather tough. Part C and D, were difficult to answer, not because I do
not know the answers, but rahter more because I cannot understand what
the question was asking.
---
I enjoyed the n-queens assignment and feel that working on it
was
not only helpful in general, but also as a preparation for the group
assignment.
---
This homework was somewhat confusing. The logic part (with the
family relations) was hard to understand.
---
Det märks verkligen att ni lägger ner tid på att
formulera hemuppgifterna. Just den här kanske var lite väl
omfattande?
---
My favorite homework because of the implementation challenge. It
was
nice to have a part of it dedicated to get an hands-on experience from
using the CSP algorithms.
---
Fun to do. A little tricky, esp the wordings of the questions.
Thought the last section was a bit too much work to do.
---
the last part (sudoku, 8queens implementation) should be more
guided.
i mean, students should know a bit more about what is required, or what
is enough to get a good mark.
Comments about homework 3:
Homework 3 was significantly more difficult and demanding
than the
first two ones. I spent more time on homework 3 than on the project.
This is one reason to question the course grading system.
---
I find that the last part was very difficult, even if we had
more
time than for the others. It was not easy to solve the problem.
---
I did this homework in less than 3 hours only because I did not
do
the final implementation part. That would have taken far too much time
considering it is end of study period and work is piling up. I
personally find this homework too tough to be finished reasonably.
---
My statistics background was not high enough for this assignment.
---
to hard!
---
fun and challenging compared to the other homeworks (sudoku
solver
was fun as well). The work spent on it and the work required in being
able to understand it were a lot more than the others so i think you
will need to tweak that a bit...
---
Implementing homework 3 was hard since the different papers
provided
different formulas and notations for the normalized algorithms. The
short paper failed to mention the n_k factor, for example. The
comprehensive one didn't even mention it and used totally different
algorithms that weren't quite as easy to implement.
Also, it
wasn't clear that we needed normalization at all (our observation
sequences were short, less than 10 observations) - the papers were
particularly fuzzy about this.
---
Vissa frågor var konstigt ställda.
Följdfrågor var baserade på att
man hade svarat på en tidigare på ett visst sätt,
annars passade de
inte riktigt in.
---
Very time consuming in a period where I didn't have so much time
to do it. So I had to skip part D without thinking about it.
---
The theoretical part of this homework was fine and could be done
with some help from the lecturers on clearing out some confusions. The
implementation part felt a bit too hard to do, and involved lots of
tricky maths and calculations that I had some hard time to tackle.
---
I spent a lot of time on this one! I enjoyed learning the
subject of
HMMs, especially since it feels very central and applicable. This was
the part of the course I enjoyed the most. I feel that I really learned
something that I can use.
---
tough job the last part. initial sections were interesting.
---
in my opinion, we lack one of the most helpful (and time-saving)
things: an example.
when
one has written a function that performs a large computation, but its
outcome doesn't seem to be logical, it is frustrating to try to guess
in which of its parts the mistake was: incorrect data preprocessing,
any of the computation steps, ...
i of course talk about the
baum-welch method, as it uses the normalized alfas and betas, as well
as the auxiliar gamma, eta, etc.. i experienced underflow problems
after a few iterations, and spotting the errors was so hard, as i
couldn't be sure about the correct operation of the sub-functions :S
such a simple thing turns a 5 hours interesting task into a 3 days
interesting, but extremely tiring, task
anyway i must say the homework helps a lot to understand how these
estimation algorithms work :)
---
the final part is a little difficult to do
---
We spent around 40 hours on homework 3.
Comments about the project:
Interesting and thought provoking.
---
Interesting to work with other people, it was a mean to better
know
who follow ths same course than us. About the subject (minesweeper for
me) it was funny to get involve in the game for the competition.
---
A little weird to make MineSweeper into a 2 player game.
---
I liked it.
---
Good project! Keept it!
---
Fun with the competition, makes you work harder.
---
det var för många med 5 i gruppen...
---
Projektet var väldigt roligt, men det hade varit bra med
lite fler
hjälptillfällen så att man kunde fått feedback
från kursledarna. Det
kändes även som att man inte hann göra det man ville med
projektet då
hemläxorna tog lång tid. Det hade också varit roligare
om man fått höra
mer om hur andra grupper löst problemet. Det kanske finns andra
sätt
att redovisa om alla har samma problem? Tvärgrupper kanske är
en idé?
Ett annat förslag är att man slumpar ut vissa grupper som
får redovisa
inför alla/mindre grupper medan andra får ha samtal med
kursledarna där
lösning osv. motiveras.
---
Mycket roligt projekt och jag hade tur med mina
projektmedarbetare, som hade liknande ambitioner med projektet.
---
Very interesting but I would have needed more time.
---
Really interesting project (minesweeper) that combined the two
major
parts of the course and ended in a fitting roundup with the competition.
---
It was really fun! I wish, however, that there would have been a
little more time for the report. We didn't start to write it until the
day before...It would really have improved the quality of the report if
we had been given an extra 2 days or something.
---
Fun!
---
nice idea! minesweeper is a pretty good project to apply the
concepts of the course.
i think time should also be graded, specially in problems that can
imply search strategies in such a big tree.
Suggested changes for minesweep:
The server should have been able to handle more parts of the
competition automatically.
---
Perhaps a little more formal information regarding rules and
behaviour of the game would be nice? It was pretty annoying to not
really know what to expect from the competition mode. Other than that,
no problems. It was great fun.
---
We should perhaps have a more scheduled timetable because we
spend a
lot of time in the room just to wait that our opponent be free to play.
---
Sätt en gräns för kvalificieringen, typ har man
inte klarat
qual-batch på 10min åker man ut. Tog för lång
tid, annars kul. Lite mer
avancerade regler skulle kunna göra det roligare för de som
vill satsa
lite mer typ passa ett drag per spel odyl.
---
Better density value and/or rules for the tournaments.
---
Document it a little better? Not everyone of us is Java expert
and
there was some difficulty in reading the codes to understand what it is
doing in such a short span of time.
---
It would have been very helpful to know how to use the
minesweeper
server and play 2-agent games before the day of the competition.
---
be able to play against last year winner
---
- better density choices, random gave many uninteresting boards.
- another way to judge those that takes too long time.
- make the start less dominating for the score.
---
It would have been better if all the groups were able to finish
the
qualification round in a shorter period of time.. Also the rules and
procuders were not really clear(at least for me.)
---
Less buggy server.
Stricter time limits - 5 minutes (?) is A LOT!
---
Kanske går att snabba upp tävlingen lite?
---
Automatisera hela tävlingen. Låt grupperna följa
ett gränssnitt för
hur man startar och stänger av programmen. Sedan kan man skicka in
sina
bidrag i början av en viss dag och sedan bara köra qualify
och större
gruppspel och finalspel och ha automatisk uppdatering av poäng
under
dagen, så slipper det bli trångt i en sal och man slipper
hantera
papper.
---
a more formal specification of the rules would be nice, as one
knows exactly what is expected from the agents performance
---
Post this years winning client (the .class files) so that one
can try out one's agent against last year's winner.
Teaches the fundamentals of AI. Got more knowledge of it,
which also developed more interest in the subject.
---
Interesting topics and ideas. HMM theory is very frequent, so
that's a good aqcuintance.
---
It was fun to get some insight into the AI field. I particularly
enjoyed the parts that brushed on machine learning and gave more focus
on the 'I' part of AI.
---
I liked the part about robotic and perception at the end,
because
I'm inteteresting with cognitive science and human machine interaction.
---
Bra exeminationsform, intressanta föreläsningar lagom
svår.
---
Teaching standard vary from class to class, it can be good on
one
day and boring on another day. For a new class with relatively young
teaching staff, it is more than acceptable.
---
Interesting programming assignments/project. Understandable
book.
Very interesting subject matter. Capable and helpful instructors.
---
Many topics covered and felt like a good overview. Most lectures
and
homeworks held a good quality. The project was very nice as well.
---
Rolig kurs som ger en bred introduktion till ämnet.
Minesweeptävlingen var ett jätteroligt initiativ.
---
Ämnena som gicks igenom var relevanta och ungefär
så djupa som man har tid att gå igenom.
Minesweeperprojektet och Homework 3 var bra för att få
implementera de delar i kursen man har lärt sig.
---
The project.
---
I felt that both teachers was really dedicated to the course and
knew a lot about the subject (or at least gave that impression =). The
times I asked for help I really felt that I got it.
---
Teachers; interesting slides/videos.
---
. it was interesting, and the project quite stimulating
. the relationship with the teachers was very good, more friendly than
one usually expects
---
the lecture is very compehensible
Nothing really, maybe some more fun videos/programs as
examples during the lectures.
---
The grading system is a mess. It should be changed to reflect
the
actual balance in the course. Information about the course was
available on the homepage, but often difficult to find. It was evident
that two people updated it. It should have been better structured.
---
The project should be given less weight in the grading. It's
really
rough to know that your grade depends on your team-mates, more than it
depends on yourself. Also, it was a bit of a bummer to find out so late
that extra attention might be paid to good homework results when
grading, when many of us have done homeworks tactically to get just the
right amount of points for the grade we want.
You should also
consider being a bit more formal and strict in your material (lectures,
notes and homeworks). Go right ahead and throw definitions, theorems
and proofs at us, we're all used to it and it will, if done correctly,
remove a lot of ambiguity in the material.
Another thing that the
course might benefit from is taking a closer look at what courses
students have taken previously. You could probably fit more material
into the course by spending less time repeating stuff that we already
know (or should know) well, such as fundamental logic and search
algorithms.
---
The part about Markov Model was sizeable and I found that it was
difficult to do the homework 3 with only the lecture. I know that there
were some further information given for that. But I was not enough
clear for me.
---
En fortsättningskurs vore intressant.
---
My main concern about this class is the way questions are asked
in
the homeworks. It makes the questions a lot tougher than they are.
---
Please be more explicit about the statistics prerequisites on
the
study handbook page and at the beginning of the course. Use words like
"Markov Models" to describe what we should know going in.
When
signing up for course, it is difficult to tell whether prerequisites
are listed because the courses are in a sequence, even though the
material is unused, or because the material is heavily relied upon, as
it was in this course.
This is especially true for the exchange students who are attempting to
decide what their equivalent course history is.
---
Homework 3 a bit and better presented HMMs.
---
See the comments on HMMs.
If a more focused approach to HMMs isn't viable, perhaps more time
should be devoted to it as it's a very interesting technique.
---
Den här kursen innehöll väldigt mycket
arbetstunga uppgifter -
kanske kan vara idé att minska lite på
arbetsbördan/utöka antalet
poäng?
---
More lectures in the beginning of the period so you have more
time
for the project. You should also probably shorten the homework a little
(less parts in each homework).
---
Personally, I prefer a written exam. You always learn more with
a
written exam. The best setup would be an exam after 70% of the time and
then the project the rest of the time.
---
Less time on non-direct topics like perception. It's good but
could be done faster. Include some more AI techniques
---
. leave more time for the project! (or, what is almost the same,
reduce the homeworks level of demand)
---
nothing