Program Type
Graduate
Faculty Advisor
Dr. Jerry Wood, Dr. Tolga Ensari
Document Type
Poster
Loading...
Location
Online
Start Date
18-4-2024 8:30 AM
Abstract
Abstract:
Anomaly detection, the identification of rare or unusual patterns that deviate from normal behavior, is a fundamental task with wide-ranging applications across various domains. Traditional machine learning techniques often struggle to effectively capture the complex temporal dynamics present in real-world data streams. Spiking Neural Networks (SNNs), inspired by the spiking nature of biological neurons, offer a promising approach by inherently modeling temporal information through precise spike timing. In this study, we investigate the use of Spiking Neural Networks (SNNs) for detecting anomalies or unusual patterns in data. We propose an SNN model that can learn what constitutes normal behavior in an unsupervised manner, without requiring labeled examples of anomalies during the training process. Our SNN approach models the temporal patterns inherent in the data through precise spike timings, allowing it to capture the complex dynamics often present in real-world data streams. Importantly, the model can continually adapt and detect anomalies. Evaluations of diverse datasets demonstrate the effectiveness of our SNN-based approach for anomaly detection compared to traditional methods. Our design achieves high detection accuracy. Furthermore, we leverage the interpretable nature of SNNs to provide insights into the learned representations and decision boundaries, enabling better understanding and trust in the anomaly detection process. Our findings highlight the potential of SNNs as a powerful tool for accurate, robust, and explainable anomaly detection, applicable to a wide range of domains and applications.
Recommended Citation
Bhandari, Shruti and Gogineni, Vyshnavi, "Anomaly Detection with Spiking Neural Networks (SNN)" (2024). ATU Research Symposium. 8.
https://orc.library.atu.edu/atu_rs/2024/2024/8
Anomaly Detection with Spiking Neural Networks (SNN)
Online
Abstract:
Anomaly detection, the identification of rare or unusual patterns that deviate from normal behavior, is a fundamental task with wide-ranging applications across various domains. Traditional machine learning techniques often struggle to effectively capture the complex temporal dynamics present in real-world data streams. Spiking Neural Networks (SNNs), inspired by the spiking nature of biological neurons, offer a promising approach by inherently modeling temporal information through precise spike timing. In this study, we investigate the use of Spiking Neural Networks (SNNs) for detecting anomalies or unusual patterns in data. We propose an SNN model that can learn what constitutes normal behavior in an unsupervised manner, without requiring labeled examples of anomalies during the training process. Our SNN approach models the temporal patterns inherent in the data through precise spike timings, allowing it to capture the complex dynamics often present in real-world data streams. Importantly, the model can continually adapt and detect anomalies. Evaluations of diverse datasets demonstrate the effectiveness of our SNN-based approach for anomaly detection compared to traditional methods. Our design achieves high detection accuracy. Furthermore, we leverage the interpretable nature of SNNs to provide insights into the learned representations and decision boundaries, enabling better understanding and trust in the anomaly detection process. Our findings highlight the potential of SNNs as a powerful tool for accurate, robust, and explainable anomaly detection, applicable to a wide range of domains and applications.