Representation Power of Neural Networks
257 views · Published 22 June 2016 · 59:01 · Indexed 14 November 2026
Channel: Microsoft Research · 2016 · Science & Technology
This talk will survey a variety of classical results on the representation power of neural networks, and then close with a new result separating shallow and deep networks: namely, there exist classification problems where any shallow network needs exponentially as many nodes to match the accuracy of certain deep or recurrent networks
