Event detection from data streams (for example, detection of computer breaches or new distant stars) typically involves a labor-intensive preprocessing step by experts, which is slow, error-prone, and expensive. Additionally, many of these events have low intensity signals that are obscured by random noise. This technology uses “Information Annihilation” (IA), a computational method to detect events by comparing two data streams without prior data interpretation. This technology provides a fast, low-cost method of event detection from any form of data stream, including low intensity data. Such high-sensitivity signal processing has wide applications in many fields, including bioinformatics, astrophysics, communications, and computer security.
This technology uses comparison of data streams to accomplish the difficult task of event detection without prior data interpretation. While standard stochastic signal analysis relies on expert pre-processing, this algorithm instead uses a universal metric to assess similarities between data streams. Specifically, every data stream, regardless of source, has an anti-stream of data. The IA step combines the stream and anti-stream to produce randomly generated noise. If the combined data significantly deviate from flat white noise, an event is present. Even events with low signal intensity will deviate from noise after IA, allowing for high sensitivity detection.
A prototype of the technology has been tested and was highly sensitive in a variety of detection tasks, including identification of abnormal heart activity from heart murmur recordings. This technology was also able to detect astronomical events that were missed by traditional stochastic analysis methods.
Patent Pending (US 20150242469)
Tech Ventures Reference: IR CU16056