TY  - JOUR
T1  - Noise Contrastive Estimation of Coupled Energy-Based Model for Anomaly Detection
AU  - Kim, Dong Kook 
JO  - The Journal of Korean Institute of Communications and Information Sciences
PY  - 2025
DA  - 2025/1/1
DO  - 10.7840/kics.2025.50.10.1497
KW  - energy-based model
KW  - noise contrastive estimation
KW  - anomaly detection
KW  - deep neural networks
AB  - In this paper, a new coupled energy-based model (CEBM) for anomaly detection and noise contrast estimation (NCE) technique for learning it are presented. The structure of CEBM consists of the product of two unnormalized probability distributions, each of which is defined by an energy function with a deep neural network. We derive the objective function of NCE by using one distribution of CEBM as a noise distribution and present a parameter update method using the gradient descent method. An experiment using ECG, UNSW and MNIST/Fashion-MNIST is conducted to evaluate the anomaly detection performance of the proposed technique. As a result of the experiment, the CEBM learned with the proposed NCE shows higher F1-score than the existing EBMs in all data sets.