Read the version of a lot of posts, found a lot of friends are asking "hot research direction", "the latest method" and so on. Some students suggest that a professor in China, or Cnki, or some sci periodicals. Whenever I see this kind of problem, I'm a little puzzled, why not read the papers at the top conference.
I have no intention of denying the value of the above documents, but in the field of machine learning, computer vision and artificial intelligence, the top conference is the king. Domestic textbooks and Cnki on the basic is n years ago the old thing. Some will question whether these meetings are just EI. Yes, it's really special: in many other areas, meetings are events such as society of neuroscience meetings, with tens of thousands of participants each, with an abstract and poster to go. But in several areas discussed, the importance of top-level meetings cannot be overemphasized.
Can be explained from the following points: (1) Because machine learning, computer vision and artificial intelligence in the field of rapid development, new work is endless, if the paper into the journal, a year or two after the publication of a little out. As a result, most of the latest work is first published in top-level meetings that reflect the "hot research direction" and "the latest approach". (2) A lot of classical work may lead to a paper in a top-level periodical, because the journal paper is more complete and full of experiments. But many are actually starting at the top of the conference. such as pLSA, latent Dirichlet allocation and so on. (3) If you pay attention to these areas Daniel's publications, it is not difficult to find that they very much value these top meetings, many people are 80% of the conference +20% of the periodical. That's why we're sending the latest work to a top-level meeting, and there's no reason not to go to the top.
(1) The following is an incomplete list, but the basic coverage.
Machine Learning Top conferences: NIPS, ICML, Uai, aistats; (Issue: JMLR, ML, Trends in ML, IEEE t-nn)
Computer vision and image recognition: ICCV, CVPR, ECCV; (Journal: IEEE T-pami, IJCV, IEEE T-ip)
Artificial Intelligence: IJCAI, AAAI; (journal AI)
Also related to Sigraph, KDD, ACL, Sigir, www and so on.
In particular, if you do machine learning, must, the nearly 4 years of NIPS, ICML turn several times, if you do computer vision, to the nearly 4 years of ICCV, CVPR, NIPS, ICML turn several times.
(2) Add: Most of the top conference papers can be downloaded from the Internet for free,
For example CV aspect: http://www.cvpapers.com/index.html;
nips:http://books.nips.cc/;
JMLR (journal): http://jmlr.csail.mit.edu/papers/;
Colt and ICML (annual official website): http://www.cs.mcgill.ca/~colt2009/proceedings.html. I hope this information will be of some help to everyone. (3) say some of your feelings. My research focuses on statistical learning and probabilistic mapping models, but it involves both computer vision and computational neuroscience, as well as some knowledge of data mining and IR. In these areas, statistical models (including probabilistic graphical model and statistical learning theory) are the mainstream and very influential methods, both in terms of methods and models. There is a very obvious trend: important methods and models appear first in Nips or ICML, and then applied to cv,ir and MM. Although specific problems and applications are also important, it is also meaningful to focus on and combine these approaches.
For the cattle in this field, the above is all plain nonsense, can be ignored completely. Welcome to discuss
