Joint Signal Recovery and Graph Learning from Incomplete Time-Series

Amirhossein Javaheri,Arash Amini,Farokh Marvasti,Daniel P. Palomar

2023-12-28

Abstract:Learning a graph from data is the key to taking advantage of graph signal processing tools. Most of the conventional algorithms for graph learning require complete data statistics, which might not be available in some scenarios. In this work, we aim to learn a graph from incomplete time-series observations. From another viewpoint, we consider the problem of semi-blind recovery of time-varying graph signals where the underlying graph model is unknown. We propose an algorithm based on the method of block successive upperbound minimization (BSUM), for simultaneous inference of the signal and the graph from incomplete data. Simulation results on synthetic and real time-series demonstrate the performance of the proposed method for graph learning and signal recovery.

Machine Learning,Signal Processing

What problem does this paper attempt to address?

The core problem that this paper attempts to solve is **learning graph structures from incomplete time - series data and restoring signals**. Specifically, the author focuses on how to simultaneously infer the graph model and the graph signal without complete data statistics.

### Problem Background

1. **Challenges in Graph Learning**:
- Most traditional graph - learning algorithms rely on complete data statistics, but in practical applications, data are often incomplete.
- Graph - learning algorithms usually assume that data are complete, while graph - signal - restoration methods require prior knowledge of the graph model.

2. **Semi - Blind Restoration Problem**:
- When the graph model is unknown, how to restore signals from incomplete time - series data is a challenging problem.

### Research Objectives

- **Joint Estimation of Graph Model and Signal**: The author proposes an algorithm based on the Block Successive Upper - Bound Minimization (BSUM) method for simultaneously inferring the graph model and the signal from incomplete time - series data.
- **Handling Incomplete Data**: This method can perform effective graph learning and signal restoration in the case of missing data.

### Formula Representation

This problem can be represented by the following optimization problem:

\[
X^*, w^*=\argmin_{X, w \geq 0} f(X, w)
\]

where,

\[
f(X, w) \triangleq\|Y - M \odot X\|_F^2+\alpha \text{Tr}(L(w) \Delta(X) \Delta(X)^{\top})-\beta \log \det(L(w)+J)+\gamma\|w\|_1
\]

- \( Y \) is the observed incomplete data matrix.
- \( M \) is the binary sampling mask matrix.
- \( X \) is the original signal matrix.
- \( w \) is the edge weight of the graph.
- \( L(w) \) is the Laplacian matrix operator.
- \( \Delta(X)=X - XD \) represents the first - order time difference of the signal.
- \( D \) is an upper triangular matrix.
- \( J = \frac{1}{n} \mathbf{1} \mathbf{1}^{\top} \).
- \( \alpha, \beta, \gamma \) are regularization parameters.

### Method Overview

1. **X Update Step**: Fix \( w \) and minimize \( f_X(X) \).
2. **w Update Step**: Fix \( X \) and minimize \( f_w(w) \).
3. **Main Algorithm**: Alternately update \( X \) and \( w \) until the stopping condition is met.

Through this method, the author can effectively learn the graph structure from incomplete time - series data and restore the signal.

### Experimental Results

The author verifies the effectiveness of this method through synthetic data and real - world data (such as PM2.5 concentration data), and the results show that this method outperforms other benchmark algorithms in graph learning and signal restoration.

### Summary

This paper proposes a new method that can simultaneously learn the graph structure and restore the signal in incomplete time - series data, solves the problem that traditional methods rely on complete data, and demonstrates its superior performance in multiple experiments.