Truth or backpropaganda? An empirical investigation of deep learning theoryDownload PDF

25 Sept 2019, 19:30 (modified: 27 Apr 2020, 18:07)ICLR 2020 Conference Blind SubmissionReaders: Everyone
Original Pdf: pdf
Abstract: We empirically evaluate common assumptions about neural networks that are widely held by practitioners and theorists alike. In this work, we: (1) prove the widespread existence of suboptimal local minima in the loss landscape of neural networks, and we use our theory to find examples; (2) show that small-norm parameters are not optimal for generalization; (3) demonstrate that ResNets do not conform to wide-network theories, such as the neural tangent kernel, and that the interaction between skip connections and batch normalization plays a role; (4) find that rank does not correlate with generalization or robustness in a practical setting.
Keywords: Deep learning, generalization, loss landscape, robustness
TL;DR: We call into question commonly held beliefs regarding the loss landscape, optimization, network width, and rank.
15 Replies