DeepMind's PathNet -- artificial general intelligence (AGI)

(Michael Guia) #1

Not sure if this paper will be included in the curriculum but it seemed relevant to the course.

Essentially, the authors are stacking neural networks and sharing gradients in clever ways.

More details in the blog post below

(Jeremy Howard) #2

There is, happily, some progress being made on transfer learning recently - I think it’s a little early to see which approaches are going to stick around, and I haven’t had the time yet to do my own research based on these recent papers; so I won’t be covering them in part 2, but I certainly plan to look into them in detail in the coming months.