Lilian Weng • 3/14/2019

Are Deep Neural Networks Dramatically Overfitted?

This technical article investigates the apparent contradiction of deep neural networks (DNNs) generalizing effectively despite their high parameter counts and capacity for perfect training accuracy. It examines classic concepts like Occam's Razor and the Minimum Description Length principle, and references modern research like the Lottery Ticket Hypothesis to understand the generalization capabilities of DNNs.

0 comments

#Machine Learning #Neural Networks #Deep Learning