I'm looking at research topics for ML, and this seems to be something that regularly comes up. One paper I found is this, where it uses a phenomenon called Rule Transfer to speed up domain transfer of knowledge.
Some questions I now have are: With popularity of deep nets and auto encoders, is this problem interesting anymore? Given that we can automatically learn interesting characteristics, does it make sense to look at this problem? Why not just learn everything?
It would be great to get some insight about this topic. I would appreciate some help. Thanks!
[link][comment]