Mosaic Looks to Outer Space for AI-Powered Surveillance

Specifically, the ESA is releasing two sets of its machine-learning algorithms, which were initially designed to monitor the thousands of instruments aboard deep-space satellites, to Mosaic, which will apply them to analyze trading behavior.

“Space exploration has always been a great engine of innovation, developing technologies that have been used across industry,” says Matt Hodgson, founder and CEO of Mosaic Smart Data. “This collaboration is the ESA’s first with a fintech company. They were looking to identify a company with the expertise in both markets and data science to find new applications for their technology and, importantly, share the results of the partnership back with the ESA so they can further develop their own models.”

In the same way that the ESA uses the algos to stop problems in their exploration craft before they happen, Mosaic hopes that this same methodology can be used to prevent compliance breaches before they occur.

“These machine-learning models spot potential technical issues on satellites before things go seriously wrong by learning what ‘normal’ behavior is and then spotting anomalies in the data from the tens of thousands of telementary parameters,” Hodgson explains. “In a similar way, when it comes to market surveillance, you are trying to spot anomalies, the behavior that lies outside the normal trading patterns which might indicate an error or trading in bad faith of some kind. Just like on the satellite, you are trying to establish what data points lie within the normal distribution and then pick out those unexpected novelties for closer examination by the compliance team.”

Machine learning is one of the subsets of artificial intelligence (AI) that has emerged as an area of particular interest for market surveillance. Particularly when combined with elements of robotics process automation, machine-learning algorithms have the ability to sift through thousands of data points in time scales that are infinitesimally shorter than what it would take human analysts, often with a higher degree of accuracy.

To date, most applications that are actually in deployment are focused on alerting. AI assists surveillance officers and investigators by using various risk factors to assign cumulative risk scores to alerts, ranking those that are most likely to be taken further into suspicious activity reports (SARs) and presenting them first to the analyst, in an attempt to reduce the time spent on weeding out false positives.

Mosaic’s partnership with the ESA is still in its initial phases, and the vendor will conduct a feasibility study on whether the technology can truly be applied to financial markets, or if more specific solutions are required. Part of the problem, Hodgson says, is that the data from financial markets is not as homogenous as that used by the ESA. However, he says, confidence is high.

“These models could be applied both internally within a bank or large buy-side institution, or by an exchange. The data science problem, surfacing novel activity from the noise of normal trade flow, is the same. The clients we have spoken to about this, which include tier-one global banks, have been very enthusiastic about the project. The work we are doing is tackling a more complex data problem than the ESA’s application of these algorithms. The data from satellites is quite regular; financial market data, on the other hand, is far less predictable and flows continuously. Adapting the algorithms for this more complex set of data is the key challenge we face in this project.

“That said, we are confident that this project will yield useful technology for the financial markets. We think the use-case for these novelty detection algorithms is very clear and we are comfortable with the risks in the R&D program,” he says.