Ten things I wish I knew before starting on a PhD

Taking a PhD is about defining yourself as an independent researcher, becoming a field expert and evolving as a person. A journey, with lots of pitfalls and setbacks, sometimes exciting and pleasant, sometimes exhausting and seemly impossible, mostly unplanned and completely unknown beforehand, but hopefully bringing you to the right place at the right time (that is your defense). At least, this is what my PhD was like. Below, I have written down ten things that I have learned mostly by trying and failing and then doing the opposite and nailing it. I hope some of these tips will be useful to the PhD students out there and perhaps make their journeys a little bit more straightforward.

So here it goes:

  1. Taking a PhD is not the same as being a student

    Getting a PhD is not about learning, but about finding things out on your own. Therefore, it does not really help to take any courses, even the mandatory ones (apart from just getting them off your requirements list). There are also no curriculum and no final exam. And most importantly, every minute counts, so spend it by doing actual research.

  2. Prepare to fail a lot and even to fail completely

    Good conferences have acceptance rates below 20%, which means that your paper is very likely to be rejected. Perhaps with a harsh criticism from an arrogant reviewer who thinks he is clever and funny. But hey, that is his job and a paper does not define you as a person anyway. So be thankful, use the feedback for improvement, resubmit, learn from it and move on.

    There is also a risk that your idea turns out not being novel or your experiments will show no advantage over the baseline. This happens a lot, but instead of losing it – learn from this, try to think of more contributions to convince the reader (that is the reviewer) that you work is worth knowing.

    Finally, you may fail completely – get bored, do a startup, run out of time, or a hundred of other things. This happens a lot and so it may to happen to you. Accept the possibility but know that focus and dedication will get you to the finish line.

  3. Start publishing and going to conferences early

    It is funny, but for most PhD’s the number of papers per year seems to grow exponentially. This means that you can easily double the number of papers by publishing your first paper one year earlier. Well, the catch is that your first paper will most likely be rejected. Nevertheless, in the worst case you will learn something from this, and in the best case you will go somewhere exotic, warm and nice, get some drinks and have some fun.

  4. Find the conferences you want to go to and read the best papers from the past two or three years …, and copy that

    Firstly, I am not talking about plagiarism – that will not get you anywhere good. Nevertheless, most of quality papers from the same conferences turn out being shockingly similar in their writing style, structure, terminology, methodology, data sets, etc. Copying that will get you quite far – if it looks like a good paper and sounds like a good paper, it must be a good paper!

    Secondly, this will give you some understanding of what is the interest area of the conference, what is hot and what you can try to improve, and of course what makes a paper really good.

    Finally, it is just so much more convenient to read a well-written related work section instead of digging up all scrap from the last ten years in the research field.

  5. Success is 30% novelty, 30% presentation and 30% luck

    You do not need a brilliant idea nor solve all the problems at once. However, you must show that your contributions are new and significant in one way or another, submit before the deadline, cross your fingers and start working on the next thing.

    Moreover, novelty is so important that some of the researchers I know have spent a lot of time looking for problems with a minimum or no previous work, and then writing a paper focusing on the novelty of the problem instead of the complexity of the solution. The biggest win of this approach is that in the next few years other researchers will use this work as a seemly naive baseline and therefore generate tons of citations.

  6. Minimize all programming, backup daily, and use version control, unit tests, checkstyle/findbugs/etc.

    Writing the most advanced code ever will not get you anywhere and only take the precious time from achieving your actual goal. However, it also sucks having to redo all your experiments due to a bug or not being able to remember when this or that part was added/changed and why.

    The best approach to research is actually this. First, find the conference you want to submit to – this will give you a deadline, the number of pages, etc. Then, find what would be a suitable idea and plan the evaluation. The best researchers I know will also write the whole paper apart from the experimental results at this point. Only then, implement the whole framework and do the experiments.

    Of course, as mentioned in the point two, your method may not beat the baseline. In that case, find out why it does not work, tackle that issue and add this to your contribution list. Lather, rinse and repeat…

    A version control system such as Git is a great tool for organizing your work (including your writing) and tracking your progress. With good commit messages you will able to find when any given part has been changed or what you did on any given day, in addition to keeping all possible versions of your work backed-up and neatly organized. Unit tests and code analysis tools will allow you to modify any part of your code without worrying about a newly introduced bug, thus keeping you more focused towards the deadline.

    Documenting your code and making it reusable as much as you can will speed-up your future work. However, it is also a highly time consuming task, thus you need to find the right balance. Consider also uploading your code to GitHub, thus letting others to learn and reuse.

    Finally, log everything you can from every experiment, not just the very minimum you wanted to present, because this will give you a future opportunity to extend your work without having to redo all the experiments.

  7. Collaborate and go abroad early

    I know the feeling: your supervisor is too busy, none of your family members and friends understands your work, all reviewers are against you, it takes a hell lot of time to publish a paper and none of your ideas seem to work. So what to do?

    – Collaborate! Talk to your fellows, even if they are working in completely different fields. Go to their presentations, get them into some kind of discussion, even if they look very annoyed and ask you to leave – oddly enough, this may result in a great paper! Find another institution that is doing great in your field and go there even without a pay. Find mentors, work with them and learn, and ask for more work after you get back!

  8. Become a reviewer

    As mentioned before, publishing a paper is primarily about convincing the reviewers that your work is great. For that you need to understand what makes a paper great, and the best approach for that is obviously reviewing other people’s papers.

    Of course, you can review any paper you find, but being a reviewer at a conference has a nice bonus – that is access to the not yet published work, which means you will know what are the topics the researchers out there are working on right now. And even better, you can help them to improve their work, which will bring you a great amount of professional satisfaction and self-realization.

  9. Keep calm and carry on

    You time is limited, most of the PhD programs are over three years – that is more than a thousand days even subtracted vacations and days off. So in the beginning it sounds like an infinite amount of time, but suddenly you find out that a half of that time has gone and you haven’t gotten anywhere. You slip into a state where you cannot think about work when you are there and you cannot think about anything else but work when doing anything else…

    – Ok, think this: Eventually it will work out. If not, there will be something else. One way or another, in five years from now you will be in a completely different state and none of this will matter at all. In fact, you will look back and remember only the best parts – that is the exotic places, banquets and drinks.

    For now, just focus on the small things – writing a good paragraph, finding a clever way to present a result, typing another block of code, running another experiment, enjoying the paper you read, etc. Whatever you do right now – focus only on this, and afterwards leave your work at the office.

  10. Enjoy your spare time, meet new people and friends

    Most PhD students struggle with loneliness. In fact, it is the greatest challenge you will encounter and the exact thing that turns your brain into a stressed monkey. So join a Judo club – some of those guys may become your best friends forever, and getting smashed against a mat for an hour and a half will surely keep you away from overestimating the importance of your research troubles. In fact, any other sport will do.

    Go out, meet a friend, have a drink and enjoy being single and young (that is the loud music and the smell of perfume and sweat). And surely talk to that beautiful girl that looks like Scarlet Johansson – little do you know, a few years later she will be your wife, and you will be sitting there 4 AM in the morning glued to an attention sick little baby that for some odd reason does not want to sleep, so tired that all the all–nighters and deadlines you pulled through your PhD sound like eight hours of the good old sleep. Or maybe not at all – she will just roll her eyes and leave, but take a chance anyway!

Read more

I was lucky to attend CIKM this year. This was the first research conference I went to since I defended my thesis, as well the first conference to attend under a non-academic affiliation. Besides giving an invited talk on things we do at Cxense right now presenting a paper on things I did at Yahoo! Labs a few years ago at the LSDS-IR workshop, meeting old colleagues and friends, I was highly interested in attending as many presentations I could. Especially, I was interested in the keynotes and industrial presentations, which there were many of. Here some of my favorites.

For the keynotes, Jaime Teevan held an exciting speech about so-called slow search, that is using more time for a better search experience; Xiaofang Zhou gave an interesting talk about mining in massive spatial trajectory datasets, some of its challenges and opportunities; and Andrew Tomkins held a mindblowing presentation about people making choice among several alternatives. Among these, I specially enjoyed Jaime’s talk for all the insightful examples, case studies and jokes.

Among the industry talks, I deeply enjoyed the Telecom data-mining talk held by Amy Shi-Nash about all the exciting work they do together with SingTel, Sofus Macskassy’s presentation about machine learning at Facebook and Wei-Ying Ma’s demonstration of the techniques used to build Xiaoice, the advanced chat-bot developed by Microsoft and becoming extremely popular in China these days.

After the conference, I had a short visit to Sydney and spent a week working at the Melbourne office of Cxense and bonding with my Australian colleagues, before heading back to my family in Norway. Overall, this was a nice trip, which I enjoyed pretty much despite a week of ten-hour jetlag and all the hustle.

Large-Scale Real-Time Data Management for Engagement and Monetization

Invited talk at the 12th International Workshop on Large-Scale and Distributed Systems for Information Retrieval, co-located with ACM CIKM 2015, October 23, 2015, Melbourne, Australia.


Cxense helps companies understand their audience and build great online experiences. Cxense Insight and DMP let customers annotate, filter, segment and target their users based on the consumed content and performed actions in real-time. With more than 5000 active websites, Insight alone tracks more than a billion unique users with more than 15 billions page views per month. To leverage the huge amounts of data in real-time, we have built a large distributed system relying on techniques familiar from databases, information retrieval and data mining. In this talk, we outline our solutions and give some insight into the technology we use and the challenges we face. This introduction should be interesting to undergraduate and PhD students as well as experienced researchers and engineers. [ Extended abstract/description: Preprint PDF, ACM DL ]

Presentation slides

Additional material

Cxense DMP Video

Demo: Associating Users with Audience Segments

Demo: Using Audience Segments to Drive Targeted Advertising with Cxense Display

How to pronounce Cxense?

Read more

Yet another year, and yet another Javazone. As always, lots of food, new people, old friends and many interesting talks. Long story short, here is my top-5 of the most entertaining talks:

  1. Java 8 JVM Memory and Thread Management by Ken Sipe – a very good explanation of the Java memory model and some practical tips.
  2. Coding Culture by Sven Peters – a set of very good suggestions on how to create a great work place for developers.
  3. The Rule of Three by Kevlin Henney – as always, a hilarious talk about everything and nothing.
  4. Building a Carputer: Java and teh Automotive Internet of Things by Simon Ritter – Java, Raspberry Pi and an Audi S3 - how awesome can it be?
  5. Production time profiling On-Demand with Java Flight Recorder by Klara Ward – nuff said!

Two more talks for those specially interested in the JVM internals:

  1. The Illusion of Execution by Nitsan Wakart.
  2. Java Concurrency Under The Hood by Gleb Smirnov.

And two more interesting talks for my Norwegian-speaking friends:

  1. Skrivetips for utviklere by June Henriksen.
  2. Anbefalinger fra Apache Mahout til Spark MLlib på 4 steg i FINN.no by Helge Jenssen.

These and other presentations and lightning talks can be found here.

I know lots of people who were passionate about computers since they were just little kids. I wasn’t. I wanted to be a chief officer (seaman), a lawyer, a scientist or a graphical designer at different parts of my childhood. The first piece of code I wrote in 2002 was something like this (this is a CASIO fx graphical calculator with a BASIC dialect).

13 years later I work as a programmer - but it is not the passion for computers that drives me forward. It is helping people to solve their problems, and of course getting a chance to put some simple mathematics into it :)