Firsts: writing a grant proposal

Despite my previous post about having a whole month to myself to write a journal paper, things went a little bit differently. A fellow PhD student pointed out these short-term fellowships and I decided it would not hurt to try to apply, so I could finance the second half of my visit to the Max Planck Institute in Tübingen. Because I already had a few scholarship applications lying around (such as for the Anita Borg scholarship, which I applied for twice, unsuccessfully), I thought a new application would not cost me more than 2 days. In the end, I spent around 2 weeks working on the new application and neglecting my journal paper, but I still believe it was time well-spent!

One thing that was different about this application is that it was not focused on me, but on the project. Of course, I already had an idea about what I would be working on at the MPI and how that fits together with my PhD topic. What I underestimated, is that I suddenly had to explain all of these machine learning problems to people with a different background – I assume molecular biology, as that is the core subject of the organization providing the fellowships.

What really helped me with writing, was a successful proposal, which was on a different topic, for a different fellowship, from a different organization, kindly provided to me by a colleague. It was a good example of HOW to write for a different audience, rather than WHAT exactly to write about. Here are some of my findings:

  • Don’t assume a term that you use every day is obvious to everybody else. Provide a short explanation and an example. If possible, use pictures in your explanation.
  • Provide references, even if something is common knowledge in your field.
  • Use short, clear sentences in the active voice (“We will conduct experiments…” rather than “Experiments will be conducted…”), here is a good post on how to do this.
  • Avoid words that make you sound unsure, such as “probably”.
  • Include questions which your project will address, such as “What is the cause of X?” or “Is it possible to do Y?”
  • Don’t be afraid to use bullet points for lists, this is probably easier to read than a paragraph of text that does not really fit together.
  • Be explicit about how your previous work is going to be helpful in this project, it might not be obvious to the reviewer that your list of publications is related to the research topic.
  • Ask others (especially people outside your lab) to read your proposal.
  • If possible, use examples (both successful and unsuccessful) of other proposals.
  • Don’t underestimate the time that you will need for writing 😉

I will only get the results of my application in a few months, but I hope these tips can be helpful to other PhD students that are in a similar situation.

Update: the proposal was not funded, but I received funding for my internship from another source, so it was a good experience overall

Firsts: preparing for a lecture, part 1

I’m very proud of it and very scared at the same time – next week I’m going to give a lecture for the first time. The lecture is a part of the Advanced Pattern Recognition course for PhD students and my own lecture will be about the dissimilarity representation and multiple instance learning – topics I should be familiar with 🙂

Right now I’m spending a lot of time in preparing for the lecture. I’m guessing that 30 hours will be a good estimate for how much time I will spend in total. Right now I will try to explain my progress and how many of these hours I am spending where.

I spent an hour or two searching for information on how to prepare your first lecture. I found helpful tips here and here. The main messages for me is: pick a few core topics and explain them well, rather than skipping over all the possibilities.

With that in mind, I started thinking about the actual content. Although the topic is very related to what I’m doing in my PhD, I want to talk more about the general techniques rather than the specific parts that I am doing. Therefore, I could not use the typical structure of my conference presentations. I started out with a mind map (or at least, a bunch of words with arrows between them) of both topics to see what exactly I would need to cover. In my head, I was already preparing the connections between different topics and thinking of nice examples, so in the end, this process costed me about 2 hours.

Then I looked at which topics I feel comfortable explaining (most related to my own research), and which topics I don’t have experience with / haven’t tried explaining to others. For instance, with a dissimilarity representation, there are two main possibilities to improve upon nearest neighbor classification: embedding the dissimilarities, or training a classifier in the dissimilarity space. In my research, I only do the latter, and although I understand the concept behind embedding, I don’t feel as comfortable with it. Yesterday I spent most of the day reading about it and at the same time trying to revise last year’s slides so I could actually use them in my own explanation. This turned out very time-intensive (+/- 7 hours), but also very helpful.

I still need to prepare the slides for my more “comfortable” topics, revise the whole story and practice. I’m not really into practicing the whole thing before presentations, but here I’m especially worried about the timing, because I have never talked for 1.5 hours before. Also, as my lecture is only on the fourth day of the course, I plan to attend the other lectures and see how the experienced people are doing it. So, probably I will revise a few things after that as well.

The last, somewhat more optional part, is to go over the exercises that “go” with my lecture. Because the content and slides that I’m using changed from the previous years, I have to check whether the exercises are still useful, and update them if necessary. I’m actually very looking forward to this, but I’m afraid I won’t have the time to come up with my own exercises, test the code, etc, so I might have to leave that for next time 🙂

To be continued!

Firsts: visiting a lab for an internship

Last week I visited the Machine Learning & Computational Biology group in Tuebingen. It’s difficult to summarize everything, but Tuebingen is a nice city, the institute is a great place to do research, and there are a lot of friendly people there! Therefore I am looking forward to my longer (few months) visit in the fall of 2013 🙂

At the group, I gave a presentation about my work, attended other talks, and discussed the project that I would be working on. The project is still being defined, but it is probably involve classifying brain data, and in particular, the connections in these brains. For instance, it could be the case that healthy people have different connections in their brains, than people suffering from neurological diseases. I hope to find out more about this very soon.

I also have to find out more about living in Tuebingen, and getting financial support to do so. So far, most grants seem to be for MSc students, PhD students who do not get any salary (but they are also supposed to be in the Netherlands, where PhD students DO get a salary… confusing), or more senior researchers. There are a few things I have to investigate further, so I hope something will come up :). It’s amazing (and unfortunate) how much time this search is costing me, though.

Another thing to think about is learning German. I don’t think it’s really necessary for a short visit, but I enjoy learning languages and I’m curious how quickly I could pick it up. There are no courses in Delft (they do have Chinese though, how awesome is that!), but there is a language exchange program. You pair up with somebody who can teach you a language, and who can learn a different language from you. I’m going to try that, and maybe also just start by myself. There must be an app for it!

Let me know if you have any experiences with exchange scholarships for PhD students, or with learning German 🙂

Writing papers online with ShareLateX

I’m working on a paper together with a PhD student who is technically in my lab, but geographically in Cuba. For some reason, neither SVN nor Dropbox were working, and I was afraid we would have to resort to emailing the paper to each other (the horror!). Then during lunch I thought that we could just use GoogleDocs for the LateX file, or maybe that GoogleDocs even supported LateX. It’s such a simple idea somebody had to already have thought about this!

And indeed, ShareLateX has! You can sign up and create LateX projects and invite others to collaborate with you. Then you have your main file, any other files you want to add, and a button that compiles the .tex file into .pdf (and you can even choose whether you want the latex or pdftex version).

Again, the idea might seem very simple, but I’m still somewhat in awe… You can work on the same LateX file real-time, without waiting for somebody to save, commit or upload a new version. This is very motivating because you see the paper changing so quickly. It is also much easier to decide things together, such as adding that new section, because you already see how it would change the paper. Last but not least, you are all using the same compiler, so you can’t mess up the tex file for each other 🙂

There is  a down side, of course. The free version only supports 2 collaborators and there is no version control. As soon as you want an upgrade, you get the “Collaborator” account which allows 10 collaborators per project but also costs you $15 a month. Not a lot if the only thing you do is write papers with people overseas, but too much if that only happens once or twice a year. I only hope that universities realize how service is great for the researchers’ productivity, and offer it to employees free of charge 🙂

Year in review: second year as a PhD student

This post originally appeared on my previous blog.

After my one-year evaluation, the first thing I did was sign up for a course on presentation skills. To be honest, I hesitated a bit at first, but I started hearing quite a few good things about Art of Presenting Science, so I decided to see for myself what that was all about. I really, really loved it! I can’t really summarize my experiences in a few sentences, so perhaps I will dedicate a whole post to this.  I tried to put a few of the things I learned to the test when presenting my work at the meeting of the NVPHBV (Dutch Society for Pattern Recognition and Image Processing), and at the Benelearn conference in Gent, Belgium.

As for research, I was pursuing several directions that are relevant to Multiple Instance Learning, learning with dissimilarities, or both.  A few of these ideas resulted in submissions to conferences, while others were abandoned after a while. Perhaps “abandoned” is not the correct word to describe the situation, and “on the shelf” would be better.  The ideas are still very interesting, but at the time I did not have enough insight to turn them into something that could be published. I hope that I will have more luck with this in 2013 🙂

In the summer it was time to take a break from my own research, and learn a lot about what others are doing at the Machine Learning Summer School in Santa Cruz, California in the US. The summer school consisted of two weeks of lectures from people from academia and from industry. The proximity to Silicon Valley ensured a lot of interesting talks by Google, Facebook and other companies that have a lot of data and therefore do a lot of machine learning. Next to all the talks, it was a great experience to meet other researchers from all of the world and compare notes on everything from doing a PhD to making tacos. I hope we will meet again!

mlss

As if that wasn’t enough traveling, I received the decisions on two papers that were submitted a few months ago. Both were accepted as poster presentations: “Does one rotten apple spoil the whole barrel” at the International Conference on Pattern Recognition (ICPR), and “Class-Dependent Bag Dissimilarities for Multiple Instance Learning” at the Structural, Syntactic and Statistical Pattern Recognition (S+SSPR) workshop.  S+SSPR was held in Hiroshima, and ICPR was held in Tsukuba, close to Tokyo. With an extra day to recover from jetlag and a few days (because three weeks in the US aren’t cheap) for sightseeing, this meant a two-week trip to Japan. One of my posters in action:

dscf7148

I enjoyed both conferences although they were completely different from each other. S+SSPR was very small, which allowed informal discussions more often and attending a lot of the talks. I also noticed the same type of close community (but with a different subset of researchers) that I saw at Multiple Classifier Systems a year earlier. ICPR was very big, which was a new experience for me. The program booklet was as large as the proceedings of some conferences! Therefore it was quite difficult to choose which talks to attend. I found out that often, my first impression (such as “this is relevant to my research” or “I won’t understand this at all”) was wrong.  In fact, I was pleasantly surprised by a few talks about unrelated topics, but given by a great presenter. There is still a lot to learn for me there.

Although I thought I could deal with time differences pretty well, I did have a few sleepless nights in Japan. On one of these nights, I had an idea about the relationship of my own work to a quite successful MIL classifier. Who wouldn’t get inspired if you are surrounded by wonderful things like this:

dscf7051

dscf7115

 

Instead of scribbling the idea down with a few words, arrows etc, which is what I usually do, I actually started writing the paper. I didn’t get very far while I was in Japan, but I did discuss the idea and let it develop. After the decision to made to submit a paper to Multiple Classifier Systems 2013 (I admit, I loved MCS 2011 so much I just couldn’t resist), the whole process of writing and submitting the paper cost me about a month. Perhaps that might be long for some people, but for me it was definitely a record. I’m also very happy with the process, so perhaps I will try this more often (starting to write a paper as soon as the idea is there).

 

Year in review: first year as a PhD student and before

This post was originally appeared on my previous blog.

Before I write an overview of 2012, I thought it would be nice to write about some highlights of the previous year.

I graduated in November 2010 with a thesis titled “Random Subspace Method for One-Class Classifiers“. This is me with my diploma:

imgp0905

After going on vacation and celebrating the start of 2011, my first official working day was on the 3rd of January 2011. Now my job was to work on a different pattern recognition topic: Dissimilarity-based Multiple Instance Learning. 

However, for a while I still continued working on my MSc subject in order to write a paper for Multiple Classifier Systems 2011. The paper “Pruned Random Subspace Method for One-Class Classifiers” (as you can guess, the regular method was not good enough) was accepted for presentation!

In June we went to Naples, Italy for the MCS 2011 conference. Here is an impression of me presenting:

dsc_3895

Here is an impression of the conference lunch:

dscf4443

Besides the great food, wine and weather I really really enjoyed the conference. I finally got to meet the people behind all the papers I have been citing and I got a feel for the type of community that all these people formed. After the conference, we had a few days to see more of Naples, the Vesuvius volcano and the ruins of Pompei.

As my next project, I helped with a journal paper that a colleague of mine was writing, “Bridging Feature and Structure Representations in Graph Matching”. Without getting into too many details, we wanted to classify objects that are represented as attributed graphs, while varying the importance of the attributes (features) or the actual graph structure. We investigated two ways to do this: using a graph edit distance and using graph kernels, which was my part. Besides learning a lot about graph kernels, I really enjoyed this project because of the regular meetings and discussions and my responsibility to the other people involved.

In September, it was time to go to Italy again, now for the Similarity-based Pattern Recognition Workshop (SIMBAD) 2011 in Venice. My supervisor presented the paper “Bag Dissimilarities for Multiple Instance Learning”, which is also what most of my current work is about. Unfortunately, I was very ill during the conference, so I didn’t have such a good conference and sightseeing experience as in Naples.

Next to research, some of my time was spent on education. I followed courses on topics related to image processing and bioinformatics, and also did the online Machine Learning course, which was very helpful. I also got to experience education from a different side a little bit while assisting in Pattern Recognition courses for PhD students and for people from industry.

In the end of the year, I had my go/no-go presentation and I received a go, together with a lot of helpful advice on how to improve my work. The main points were to become more comfortable with mathematics, be more precise in why I’m pursuing a certain direction in research, and to improve my presentation skills.

%d bloggers like this: