Timezone: »
Mathematically characterizing the implicit regularization induced by gradient-based optimization is a longstanding pursuit in the theory of deep learning. A widespread hope is that a characterization based on minimization of norms may apply, and a standard test-bed for studying this prospect is matrix factorization (matrix completion via linear neural networks). It is an open question whether norms can explain the implicit regularization in matrix factorization. The current paper resolves this open question in the negative, by proving that there exist natural matrix factorization problems on which the implicit regularization drives all norms (and quasi-norms) towards infinity. Our results suggest that, rather than perceiving the implicit regularization via norms, a potentially more useful interpretation is minimization of rank. We demonstrate empirically that this interpretation extends to a certain class of non-linear neural networks, and hypothesize that it may be key to explaining generalization in deep learning.
Author Information
Noam Razin (Tel Aviv University)
Nadav Cohen (Tel Aviv University)
More from the Same Authors
-
2020 Workshop: First Workshop on Quantum Tensor Networks in Machine Learning »
Xiao-Yang Liu · Qibin Zhao · Jacob Biamonte · Cesar F Caiafa · Paul Pu Liang · Nadav Cohen · Stefan Leichenauer -
2019 Poster: Implicit Regularization in Deep Matrix Factorization »
Sanjeev Arora · Nadav Cohen · Wei Hu · Yuping Luo -
2019 Spotlight: Implicit Regularization in Deep Matrix Factorization »
Sanjeev Arora · Nadav Cohen · Wei Hu · Yuping Luo