Abstract
In this paper, we propose a new Expectation-Propagation (EP) algorithm to address the problem of joint robust linear regression and sparse anomaly detection from data corrupted by Poisson noise. Adopting an approximate Bayesian approach, an EP method is derived to approximate the posterior distribution of interest. The method accounts not only for additive anomalies, but also for destructive anomalies, i.e., anomalies that can lead to observations with amplitudes lower than the expected signals. Experiments conducted with both synthetic and real data illustrate the potential benefits of the proposed EP method in joint spectral unmixing and anomaly detection in the photon-starved regime of a Lidar system.
Original language | English |
---|---|
Title of host publication | 2020 28th European Signal Processing Conference (EUSIPCO) |
Publisher | IEEE |
Pages | 2463-2467 |
Number of pages | 5 |
ISBN (Electronic) | 9789082797053 |
DOIs | |
Publication status | Published - 18 Dec 2020 |
Event | 28th European Signal Processing Conference - Amsterdam, Netherlands Duration: 18 Jan 2021 → 22 Jan 2021 https://eusipco2020.org/ |
Publication series
Name | European Signal Processing Conference |
---|---|
ISSN (Electronic) | 2076-1465 |
Conference
Conference | 28th European Signal Processing Conference |
---|---|
Abbreviated title | EUSIPCO 2020 |
Country/Territory | Netherlands |
City | Amsterdam |
Period | 18/01/21 → 22/01/21 |
Internet address |
Keywords
- Anomaly detection
- Approximate Bayesian inference
- Expectation-Propagation
- Linear regression
- Poisson noise
ASJC Scopus subject areas
- Signal Processing
- Electrical and Electronic Engineering