How to best thesis into

　In 2007 the international multimedia Conference, Microsoft Research Asia of a paper won the best paper award (see previous blog article Augsburg dual award myth). Recently, the Institute for intern Liu on one of the main authors of the paper's first victory over fellow did a story on the award of the interview behind. Author: Liu, University of science and doctoral

Q1: first of all congratulate you get 2007ACM international multimedia conference best paper Awards, can give us a brief description of award-winning thesis?

A1: thank you.

This essay is about video annotations, belongs to the video index. Video index is given to do analysis of video, extract useful semantic information. The most advanced semantic information is text, video annotations is to create video and text between semantically. This is a very difficult problem, but in video search and other fields have an urgent application needs. Early video callout is artificial, and time consuming. Subsequently began to study with machine learning methods for automatic video annotations. We consider a problem is that corresponds to a video of multiple keywords, the previous solution is a single keyword, respectively, and then the results be fusion, this will lead to error propagation. We introduced a new approach, taking into account the more keywords the intrinsic link between the creation of a complete framework to address multiple keywords video annotations.

Q2: for this article, "best papers" that you have and what kind of experience to share?

A2: bluntly, it's no secret (smiling).

First of all, a good thesis behind is solid, has a theory and practical application of the value of the work, this is a good paper. Only a good idea to eventually produce a good thesis. Secondly, in writing, I think it should be simple, both in-depth analysis of the problem — and you want to go out to brief — clearly express their ideas. Our papers are changed many times before. Of course, doing presentations in the General Assembly is also important. In order to prepare this paper reports on the scene, we group internal rehearse three or more, I prepared a more times.

Q3: are you referring to the solid work is very important, so in this paper, behind which the accumulation and experience?

A3: there are many.

We start from the 2005 video retrieval, video tagging work in this area, in the top-level academic conferences and journals are published. This ACMMultimedia2007, we made a 5 story long papers, (57 articles) 9%, this is not difficult to prove that our work was a peer recognition. Video annotations alone on this issue, we have been a number of exploration process. For example, the previously used some half-supervision of machine learning approach, then also used active learning, also have to consider some of the features of the video itself, for example, time consistency, to improve annotation results. Over the past year, we introduced a concept of Multi-Multi, multiple keywords, multiple modes, multiple levels and multiple instance and so forth, in a unified framework for consideration. This article, "best papers" is to belong to this a big topic.

(China first victory fellows and interns Qi Jun on their way in to the General Assembly)

Q4: you mentioned these methods is more mainstream.

A4: of course the mainstream for research is need to know, at least to "know".

From another perspective, this video annotations can be viewed as machine learning theories, methods and tools of an applied research. Since it is a tool, you must understand, but not necessarily all intensive. However, as an applied research, only put these tools take over as black box is available, that are mostly not be successful. This need to grasp the scale. I advocate, or should be on the tools have a basic understanding of and response to a question, try to use suitable or improved even innovative tools. Another way is directly with the problem to find tools, this may not be I agree.

Q5: for your students, you have special instructions?

A5: I have some guidance method and philosophy, may and others not so.

First of all my students are treated differently, for example, new students, I will ask them to submit a report every day, write about today's progress and plans for tomorrow, practice time longer students don't have to write.

Q6: daily report, is it necessary? students may also be a burden.

A6: just arrived here to practice for new students, often doing research also is not very experienced, although we practice groups also have weekly meetings, students and the instructor may still be some misunderstanding, then if he writes daily report, you can identify issues and communicate with him.

On the other hand, writing, reporting on students ownIs an exercise, the novice acquire good habits are important daily summary is a good habit.

Q7: for new students, you still have a way to help he accumulated research experience?

A7: start from the easy problems.

This is important. Research needs to have the confidence to do an easy problem is a good way to help novices build their self-confidence. Confident, bound, not overcautious, ideas are easy to open. In addition from successful examples you can learn more experience. Some students do not start is very good, he found an easy problem, write a good dissertation, send in a better Conference or journal, starting from this moment as if suddenly see that future research to get yueshun. Students stayed here for some time, more experienced, I will give him more freedom.

Q8: what are the so-called freedom?

A8: mainly to encourage students to think Idea, I will be very happy and students discussed his Idea to any one.

Sometimes I will experience there the Idea cannot be used, but this time I generally is not a simple negation, but rather to point out to make him think. Maybe a look bad Idea, as amended may get a very good idea. Research is a creative work to do before we all have limitations. If a student in me for a couple of years, his Idea was not able to go beyond my imagination, I might think that this is a failure.

Q9: what do you think Research Institute in the Dr to these interns research environment which features?

A9: compared with the University environment, doing research in Microsoft Research Asia has its own characteristics.

First of all, here we do mostly forward direction and useful research, is the industry and the academic community are interested in. Secondly, the study of the atmosphere here is very good, everyone has a spontaneous passion, hard work, forming a good group. There is also a study of Microsoft Research Asia has some very good tradition, this is from top to bottom, with disciples such band master, are able to exercise. But Microsoft and other companies, research institutions or not, here you can also do some usability not so strong, basic research, for the academic community is also a good thing.