Finding patterns in timed data with spike timing dependent plasticity

Abstract

My research focuses on finding patterns in events - in sequences of data that happen over time. It takes inspiration from a neuroscience phenomena believed to be deeply involved in learning. I propose a machine learning algorithm that finds patterns in timed data and is highly robust to noise and missing data. It can find both coincident relationships, where two events tend to happen together; as well as causal relationships, where one event appears to be caused by another. I analyze stock price information using this algorithm and strong relationships are found between companies within the same industry. In particular, I worked with 12 stocks taken from the banking, information technology, healthcare, and oil industries. The relationships are almost exclusively coincidental, rather than causal.