2015机器学习十大问题

Machi ne Learning is a very vast field , a nd mu ch of it is still an active resea r ch are a . Th ere  are many interesting problems that can easily be nominated a s  a "t op" pr oble m,  h o w ever I d on't think any of them will be sol ved  (completely) b y the end of 20 15 .

A s  an example , h ere is a fun pap er b y  Dr Pedro Domingos   written in 2007 a b out  Ten Problems for the Next Ten Years . Its  almost  2015, a nd we are no t able solve any one of them completely ye t, s o I think these problems will r emai n "top  problem s" at  least for a year!

But my 2 cent are on these topics (in  no particular order) -

1. Structured prediction
   
2. Inductive transfer
   
3. Marginal MAP problem
   
4. Learning architecture of Deep Net
5. Combining Deep Learning with Statistical Relational Learning
6. Combining SVM with Probabilistic Graphical Models
7. Determining learning rate (No More Pesky Learning Rates)
8. Learning in presence of partial data (Expectation–maximization)
9. Learning Probabilistic programs
10. Banishing "black arts" from machine learning (A Few Useful Things to Know about Machine Learning)