GSoC Introduction - generic multiclass learning framework
2012-04-24 07:39:23 GMT
My name is Chiyuan Zhang (张驰原), also known as pluskid in IRC (and almost everywhere else). I'm a CS student from Zhejiang University, China. I have finished my current degree and will start my PhD pursuit at CSAIL, MIT this Sept. I'm luckily selected for the GSoC task of developing a generic multiclass learning framework in shogun, mentored by Cheng Soon Ong.
My work is mainly consists of two parts:
- Designing and implementing a generic framework in shogun for multiclass learning
- Implement specific multiclass learning algorithms
The first part is (very) roughly done, since there were multiclass classification support in shogun. During the past weeks, I have helped to reorganize the existing code and framework so that it is more unified and better extend-able. There are still many issues to be solved in the current framework, but things will turn out to be clear as we go on.
For the 2nd part, I have a draft plan in my proposal, but I think that is somewhat preliminary, and I need to make the plan more accurate and specific with my mentor during the next weeks. However, any suggestions from all of you are welcomed! But to start with, I'll implement the simple ECOC way of multiclass classification.
Other details are to be decided during the following weeks. We will also have a kick off meeting as far as I know. Looking to see you there.
I will also work closely with Sergey Lisitsyn, since he has strong interest in multiclass learning, and is an expert of the shogun codebase. There might be potential collaboration with Fernando J. Iglesias García, too. Since multiclass learning and SO learning are somewhat related. For example, we might both want to extend the current CLabels class for our purpose.
Finally, thank you very much for this awesome project!