A Majorization-Minimization Gauss-Newton Method for 1-Bit Matrix Completion

Xiaoqian Liu,Xu Han,Eric C. Chi,Boaz Nadler

2024-09-24

Abstract:In 1-bit matrix completion, the aim is to estimate an underlying low-rank matrix from a partial set of binary observations. We propose a novel method for 1-bit matrix completion called Majorization-Minimization Gauss-Newton (MMGN). Our method is based on the majorization-minimization principle, which converts the original optimization problem into a sequence of standard low-rank matrix completion problems. We solve each of these sub-problems by a factorization approach that explicitly enforces the assumed low-rank structure and then apply a Gauss-Newton method. Using simulations and a real data example, we illustrate that in comparison to existing 1-bit matrix completion methods, MMGN outputs comparable if not more accurate estimates. In addition, it is often significantly faster, and less sensitive to the spikiness of the underlying matrix. In comparison with three standard generic optimization approaches that directly minimize the original objective, MMGN also exhibits a clear computational advantage, especially when the fraction of observed entries is small.

Machine Learning

What problem does this paper attempt to address?

### Problems Addressed by the Paper

The paper aims to address the **1-bit matrix completion problem**. Specifically, given a partially observed binary matrix \( Y \), the goal is to estimate an underlying low-rank matrix \( \Theta^* \). In this problem, the observed data is a set of binary observations, which are independently generated from each element of the underlying matrix \( \Theta^* \).

#### Main Contributions
The paper proposes a novel method called the **Majorization-Minimization Gauss-Newton (MMGN)** method to solve the 1-bit matrix completion problem. This method is based on the majorization-minimization principle, transforming the original optimization problem into a series of standard low-rank matrix completion subproblems, and solving each subproblem using the Gauss-Newton algorithm. Compared to existing methods, MMGN has the following advantages:

1. **Faster Speed**: MMGN is typically at least an order of magnitude faster than existing 1-bit matrix completion methods while maintaining comparable or even higher estimation accuracy.
2. **Insensitivity to Matrix Spikiness**: MMGN's performance degrades more slowly as the spikiness of the underlying matrix increases.
3. **Ease of Implementation**: MMGN can be implemented using standard linear algebra tools, whereas existing methods are computationally complex and difficult to implement on large matrices.

#### Experimental Validation
Through simulation experiments and real data examples, the paper demonstrates the performance of MMGN under different settings, including variations in noise levels, observation proportions, and other parameters. The experimental results show that MMGN significantly outperforms other general optimization methods (such as gradient descent, quasi-Newton methods, etc.) in terms of runtime and also exhibits excellent estimation accuracy.