Comments on the difficulty level of the course with respect to prerequisites:
I think the problem lies more in Machine Learning, Base Course. There were not much in that course that were in use in this course.
---
It is a MUST to inform in advance that this course requires a lot of confidence in Bayesian statistics. A small course or practical handouts in form of theory and exercises would be beneficial for this matter. It feels like the basic course could have done a better job with preparing us for the heavy Bayesian statistical framework that the advanced course encapsulates.
---
Was a long time since i used linear algebra so was hard to understand the derivations
---
Probably the most work intense course I have taken, had to study a fair amount to understand all material but it was definitely doable.
---
I could understand but not replicate in detail every mathematical "trick" in some of the methods. I believe this is due to my own lack of knowledge.
---
People from CS have all the needed prerequisite math courses in year 1 and 2 but after that, that math is mostly unused until courses like this which makes it take a bit longer to get up to speed again.
---
I think it had the appropriate level of difficulty.
---
The topics covered in the course are indeed advanced and demand prerequisites which are not gained in the basic course, So there is a mismatch between the basic and advanced course in that sense.
---
I feel that there was a big difference between the course DD2431 Machine Learning and this one.
Comments on the focus of the course, e.g., the focus on theory vs practice.
I had a lot of fun doing assignment and projects. Maybe a bit too much for 7.5 credits?
---
The course was so abstract so that it was hard to imagine what you would use things for.
---
Good balance
---
I think theory and practice was nicely balanced in the assignments and lectures
---
One of the best courses I've taken
---
This course had equal flavors of both theory and practice.
---
I think that the course covers too many topics. The course is very demanding, and there is nothing wrong with that. However, after having taking it I personally have a hard time pointing out what exactly I learned and where I can apply the techniques presented in the course.
---
We have seen some quite abstract notions and it was difficult to grasp sometimes without seeing a practical implementation.
Comments on the book, e.g., suggestions of alternative books:
I didn't not like the book too much. Not sure why but maybe it's not that well structured and sometimes with too many details.
---
I think Murphys book is better. Although I have heard that some version of that book has a lot of typos etc in it.
---
I didn't spend too much time reading the book.
---
Some examples felt a little contrived, otherwise solid book.
---
The book is fine. I just didn't have much time to read it.
I think a learned a lot with both the assigments and the project. Not sure it would be the same with an exam. With these tasks it's hard to learn without getting hands dirty.
---
I loved the assignments, even though they were difficult. I learned a lot from them. The project is a great way to end the course
---
The course design is good, but the workload is brutal. :)
---
It was hard to implement the algorithms in the assignments because it was hard to see the connections to the abstract lectures. Jens part was fairly easy to understand since it was not as abstract.
---
I like the way it designed, especially that no written exam.
---
It is a nice design and the courses like this will need to have assignments and projects instead of exams.
---
Keep this structure, good way of putting theory into practice
---
It was fine! A pretty high workload compared to some other courses.
---
Just perfect.
---
I liked it a lot
---
I think it should be 4.5 to assignments and 3 credits to the project. Project should be divided into two separate deadlines, one each for theory and implementation.
---
I think a combination of a written exam and projects would be more interesting, as to cover both theoretical and practical issues in depth.
---
No exam and practical assignments is great. I would prefer having more time to work on the assignments, and removing the project.
---
I feel that it is a design that suits very well this course. Taking the advanced course means that we want to learn more in this area and the project gave us an opportunity to go deeper in a certain subject as it is impossible to cover all. However, the assignments were really challenging and I think that giving the chance to work in pairs for the assignments would ease the work without affecting at all the gained knowledge.
Comments on Lectures 1 and 13:
I didn't find them much useful. They can rather be used for the content of the course. Since the first half of the lectures by Carl were a little quick.
---
It was interesting to hear what the main trends in Machine Learning. At this moment I really felt that I acquired important knowledge as I was capable to grasp the trends discussed at NIPS.
Comments on Lectures 2-5:
Carl is a great teacher and he knows how to keep high the attention.
---
I love the entusiasm of Carl Henrik.
---
Pretty much understood what Carl meant when he talked but the slides did not much to improve the understanding. When i went back and looked through the slides in order help with my understanding i could not understand alot of the slides. One of the reasons for this was that alot of the figures/pictures missed titles and axis labels making it hard to interpret them. Especially when something changed in a figure between slides but it didnt say what changed. It also didnt help that there often was over 150 slides each lecture.
---
The intuition of the methods were given real nice, the efforts of Carl Henrik on trying to explain the math behind it was admirable.
CHEK FOR PRESIDENT
---
Carl-Henrik is a really good lecturer.
---
Undoubtedly, one of the best lectures that I have attended.
Comments on Assignment 1:
In the lectures we went through it all on the surface and in the assignments we went hard into the details which became hard since you only knew briefly what it was supposed to do.
---
Assignment was quite a challenge but I learnt a lot from it.
---
I liked Assignment 1 because a) it covers one single (altough very broad) topic, b) each part of it builds on the previous part, so as you proceed with it you gain better understanding of the subject
---
I didn't receive a lot of notes on my assignment. However, I understand that the time was limited and it was not possible to give a full feedback at each student.
Comments on Lectures 6-9:
I though your slides was clear in what they meant and the examples you used and went through was very informative and easy to relate to. The only thing i though was a little unnecessary was that you did the derivations for almost all the methods.
---
Some of the answers and explanations were a little confusing at times.
---
This part of lectures were kind of repetition of stuff done in AI course with different terminology. Lectures would have more worth, if the content is discussed with AI course faculty.
Comments on Assignment 2 part 1:
The casino model task was not so clear. We had no introduction to qualitative belief networks during the lectures.
---
I think there was limited course material for Assignment 2 part 1. The ones available did not help me understand the questions.
---
It was only the first part where you had to say which one that was bigger that didnt really reflect on what we had done on the lectures.
---
I think the first assignment of Assignment 2 was one of the most discussed and asked about on KTH social since the lectures didn't really cover everything needed for it and it was hard to find information about the subject
Comments on Lectures 10-12:
It often feelt like you had not rehearsed what you where going to say on the lectures and alot of time went to finding it. Once found the lecture was good.
---
Most of the felt like the lecturer was not prepared and/or stressed out. So the structure suffered heavily from this.
---
Even though some of the explanations used the advanced technique of rigorous hand-waving, it was still clear how the methods worked in the end.
---
Although I couldn't attend all the lectures, but I didn't find the lectures I attended much convincing.
---
Even though the lectures were quite chaotic sometimes, they really helped me to understand Hegvig's part in HA2.
Comments on Assignment 2 part 2:
It was what had been on the lecture on a fair level.
---
I liked Hedvig's part because she made us read papers rather then the text book. Also I loved implementing LDA. I feel like Hedvig's part gave me a broader insight into what ML is. The other two parts presented classical techniques that I have studied in other courses.
Comments on the project:
I had fun to be honest. I learned new methodologies, and going through the problematics of reading a scientific paper is really educative!
---
5 people are maybe to large project groups. It was difficult to properly assign tasks for each person.
---
Was a little hard to understand what you where supposed to research, make a little clearer explanation of that and it will be good.
---
Project was good and the team assignment by the teachers have helped to get to know others. Implementation, report writing and the corresponding oral presentation is good.
---
We got a really interesting paper to work on so that meant a lot
---
It was fun to really dig into a specific algorithm and learn every detail of it
---
Maybe mixing the groups a little in terms of their first assignment grades would have been than bunching up similar graded people. So the more enthusiastic people could maybe motivate the others.
---
Project should have multiple deadlines to ensure better results from all groups. Almost all the groups ended up doing all the work in the last week. Due to this reason, few groups could not complete the task.
---
The schedule of the project was not very adequeate, because during christmas it is always difficult to do group work, even more for groups of around 5 people.
---
Although I think that the project was too much and that I would have preferred more time on the assignments, I did learn a lot from working on it. I enjoyed my group a lot so it turned out to be one of the best project works I've ever had throughout my studies. In the end, I learned more about both practical and theoretical parts of working with the material in the paper. I like that it was based on a research paper.
I love how the assignments are formed to really encourage the students to fully understand the methods. I have learned a lot from this course.
---
That the course is brutally hard. It challenged my intellect in a whole new way.
---
The extra help sessions for the assignments, would even like to see some more. The recap that Hedvig did on the last lecture from the conferance. Jens part of the course I really liked. The explanation on assignment 2 on ICA both had generated data and figures of it before ICA and after making it alot easier to see when you were on the right track, keep it.
---
I think all parts should be retained pretty much as they were.
---
Good with no exam and large assignments.
While I love the assignments, they were very time consuming. I think I spent around 150 hours only on the assignments, and I did not even answer every question.
---
Better preparations for Bayesian statistics.
---
One thing that would have helped me with this course is a list of help functions that can help when doing the plots for the assignments. For example: "These are some commands that can help you along the way:
numpy.random.multivariate_normal(mean, cov[, size]) % create a multivariate distribution of size n
plt.plot(x,y) % plots a line ........" and so on. Because it took alot of time to just find the tools that created what you wanted and I spent alot of time on these assignments.
---
Perhaps deep learning should be covered since it has gained a lot of traction recently
---
I think that it should be stated clearly how deep the answers to the reports have to be, because even though I performed the experiments and answered questions, some were marked as "not complete" or "too short". If I had known that before, I would have extended my answers.
---
The course covers too many topics. I worked very hard but at the end I am not sure if I can summarize what it is I learned.
Denna sammanställning har genererats med ACE.