Notes of Todays talk

Todays talk is about NP hard problems in machine learning, under certain kind of assumptions of the input and the output, could be solved in polynomial problem.

The high level message is that some NP hard problem, if you consider about the input in a general case, the worst-case complexity would be horrible. However, if you add more assumptions, such as the structure of the data (depends on the specific problem you are working on), could make a huge difference on the complexity. 

But the most challenging put is about making the assumption. This also give me similar feeling in the problem we are attacking right now. as the data are high dimensional and complicated, we don't have good weapon on firing at these problems. i feel that we are approaching by the same philosophy. 

Another thing worthy to mention is that Bin jie talked to me today. I just found she is cute when she "critisize" me, but unfortunely she found it. But it helps. I should be more serious in many situation now.