skip to main content


Title: Implicit Regularization in Matrix Factorization
We study implicit regularization when optimizing an underdetermined quadratic objective over a matrix X with gradient descent on a factorization of X. We conjecture and provide empirical and theoretical evidence that with small enough step sizes and initialization close enough to the origin, gradient descent on a full dimensional factorization converges to the minimum nuclear norm solution.  more » « less
Award ID(s):
1302662
NSF-PAR ID:
10025955
Author(s) / Creator(s):
; ; ; ;
Date Published:
Journal Name:
arXiv.org
ISSN:
2331-8422
Page Range / eLocation ID:
arXiv:1705.09280v1 [stat.ML]
Format(s):
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. In the past few decades, there has been rapid growth in quantity and variety of healthcare data. These large sets of data are usually high dimensional (e.g. patients, their diagnoses, and medications to treat their diagnoses) and cannot be adequately represented as matrices. Thus, many existing algorithms can not analyze them. To accommodate these high dimensional data, tensor factorization, which can be viewed as a higher-order extension of methods like PCA, has attracted much attention and emerged as a promising solution. However, tensor factorization is a computationally expensive task, and existing methods developed to factor large tensors are not flexible enough for real-world situations. To address this scaling problem more efficiently, we introduce SGranite, a distributed, scalable, and sparse tensor factorization method fit through stochastic gradient descent. SGranite offers three contributions: (1) Scalability: it employs a block partitioning and parallel processing design and thus scales to large tensors, (2) Accuracy: we show that our method can achieve results faster without sacrificing the quality of the tensor decomposition, and (3) FlexibleConstraints: we show our approach can encompass various kinds of constraints including l2 norm, l1 norm, and logistic regularization. We demonstrate SGranite's capabilities in two real-world use cases. In the first, we use Google searches for flu-like symptoms to characterize and predict influenza patterns. In the second, we use SGranite to extract clinically interesting sets (i.e., phenotypes) of patients from electronic health records. Through these case studies, we show SGranite has the potential to be used to rapidly characterize, predict, and manage a large multimodal datasets, thereby promising a novel, data-driven solution that can benefit very large segments of the population. 
    more » « less
  2. In practical instances of nonconvex matrix factorization, the rank of the true solution r^{\star} is often unknown, so the rank rof the model can be over-specified as r>r^{\star}. This over-parameterized regime of matrix factorization significantly slows down the convergence of local search algorithms, from a linear rate with r=r^{\star} to a sublinear rate when r>r^{\star}. We propose an inexpensive preconditioner for the matrix sensing variant of nonconvex matrix factorization that restores the convergence rate of gradient descent back to linear, even in the over-parameterized case, while also making it agnostic to possible ill-conditioning in the ground truth. Classical gradient descent in a neighborhood of the solution slows down due to the need for the model matrix factor to become singular. Our key result is that this singularity can be corrected by \ell_{2} regularization with a specific range of values for the damping parameter. In fact, a good damping parameter can be inexpensively estimated from the current iterate. The resulting algorithm, which we call preconditioned gradient descent or PrecGD, is stable under noise, and converges linearly to an information theoretically optimal error bound. Our numerical experiments find that PrecGD works equally well in restoring the linear convergence of other variants of nonconvex matrix factorization in the over-parameterized regime. 
    more » « less
  3. null (Ed.)
    We experimentally examine how gradient descent navigates the landscape of matrix factorization to obtain a global minimum. First, we review the critical points of matrix factorization and introduce a balanced factorization. By focusing on the balanced critical point at the origin and a subspace of unbalanced critical points, we study the effect of balance on gradient descent, including an initially unbalanced factorization and adding a balance-regularizer to the objective in the MF problem. Simulations demonstrate that maintaining a balanced factorization enables faster escape from saddle points and overall faster convergence to a global minimum. 
    more » « less
  4. The success of gradient descent in ML and especially for learning neural networks is remarkable and robust. In the context of how the brain learns, one aspect of gradient descent that appears biologically difficult to realize (if not implausible) is that its updates rely on feedback from later layers to earlier layers through the same connections. Such bidirected links are relatively few in brain networks, and even when reciprocal connections exist, they may not be equi-weighted. Random Feedback Alignment (Lillicrap et al., 2016), where the backward weights are random and fixed, has been proposed as a bio-plausible alternative and found to be effective empirically. We investigate how and when feedback alignment (FA) works, focusing on one of the most basic problems with layered structure n×m, the goal is to find a low rank factorization Zn×rWr×m that minimizes the error ∥ZW−Y∥F. Gradient descent solves this problem optimally. We show that FA finds the optimal solution when r≥rank(Y). We also shed light on how FA works. It is observed empirically that the forward weight matrices and (random) feedback matrices come closer during FA updates. Our analysis rigorously derives this phenomenon and shows how it facilitates convergence of FA*, a closely related variant of FA. We also show that FA can be far from optimal when r more » « less
  5. The success of gradient descent in ML and especially for learning neural networks is remarkable and robust. In the context of how the brain learns, one aspect of gradient descent that appears biologically difficult to realize (if not implausible) is that its updates rely on feedback from later layers to earlier layers through the same connections. Such bidirected links are relatively few in brain networks, and even when reciprocal connections exist, they may not be equi-weighted. Random Feedback Alignment (Lillicrap et al., 2016), where the backward weights are random and fixed, has been proposed as a bio-plausible alternative and found to be effective empirically. We investigate how and when feedback alignment (FA) works, focusing on one of the most basic problems with layered structure n×m, the goal is to find a low rank factorization Zn×rWr×m that minimizes the error ∥ZW−Y∥F. Gradient descent solves this problem optimally. We show that FA finds the optimal solution when r≥rank(Y). We also shed light on how FA works. It is observed empirically that the forward weight matrices and (random) feedback matrices come closer during FA updates. Our analysis rigorously derives this phenomenon and shows how it facilitates convergence of FA*, a closely related variant of FA. We also show that FA can be far from optimal when r more » « less