Anomaly Detection with Spiking Neural Networks

In recent years, there has been a revolution in Machine Learning (ML) techniques that can be attributed to the large increase in computational power, mainly harnessing the efficiency for matrix multiplication and convolution granted by Graphical Processing Units (GPUs). The most widespread of these techniques was the Artificial Neural Network (ANN), whose architectural principles were inspired by information processing and distributed communication in biological systems. Even though ANNs were created to mimic brain-like calculations, facilitating communication between artificial neurons and performing calculations inside these neurons, neuroscientists were quick to point out that biological systems perform completely differently from those proposed by any current type of ANN. As such, Artificial Intelligence (AI) has not reached the truly limitless potential that was once promised. Neuromorphic hardware and Spiking Neural Networks (SNNs) [1], which behave like the human brain, aim to fix this problem, bridging the gap between artificial and biological intelligence. The inherent strengths of these SNNs would be extremely useful at the Large Hadron Collider (LHC) with their need for fast inference and accurate data-processing of petabytes of time-series events.