timr: The difference between this and normal outsourcing is that I'm not outsourcing the design aspect of it, merely the algorithmic development. Think of it as a netflix challenge style problem. The problem would be specced roughly as:
"Here is an input vector, here is an output vector, write an algorithm that can translate input -> output with 98%+ accuracy". Of course, there would be validation data as well which they never get to see.
scott_s: Of course, lots of research is by necessity vague but a lot of it can be tightly scoped as well. Handwriting recognition, for example, simply involves trying to improve an accuracy measure.
sireat: There's what? 1000 universities in China and I don't read Chinese. Such an approach seems really hit or miss to me.