Featurespace invents next generation of Machine Learning
Featurespace’s Cambridge R & D team has mined previously untapped capability in machine learning and behavioural technology to provide unparalelled defence against online financial fraudsters.
The team – boasting some of the best brains in the world – have delved deeper than DeepTech has ever gone in the cards and payment industry to fortify the defences of clients and transactions involving millions of their customers worldwide.
The result is the launch this week of Featurespace’s transformative Automated Deep Behavioral Networks (ADBN) architecture for payments in real-time, which scuttles scams and fends off fraud faster and more forcefully than ever before.
One of the many gamechanging elements of the invention is a novel Recurrent Neural Network architecture available through the latest version of the company’s ARIC™ Risk Hub.
Patterns and timings of fraud attacks in various territories and against different social or economic backdrops can be identified with much more certainty.
That, in turn, enables users of ADBN to spot and thwart a detected raid before there is any threat of money leaving someone’s account – or indeed any attempt to take over an account.
In times of downturn and pandemic, for example, attacks can escalate as financial criminals endeavour to exploit opportunities such as social, mental and medical vulnerabilities.
They may see opportunities to be exploited in terms of availability and auctions for COVID vaccines, for example.
Financial fraud is already costing companies and their customers many billions of dollars every year the length and breadth of the planet. Featurespace is already stemming the tide.
The company’s R & D heroes, directed by Dr Dave Sutton, have worked with many of the company’s international customers in sculpting the new technology to their exact requirements and road testing the results in partnership.
Continued innovation will enable the solution to grow with the requirements of the businesses that deploy it – and that, in turn, could lead to potential uses for the technology in other industry sectors.
Featurespace co-founder Dave Excell, who heads up the firm’s Atlanta operation, has no doubts about the potential power and adaptability of the technology.
He says: “The significance of this development goes beyond the scope of addressing enterprise financial crime. It’s truly the next generation of Machine Learning.”
It’s the modus operandi of attackers and the behaviours of the consumers – coupled with anomaly detection – that enables Featurespace to spot previously undetectable fraud attacks.
That depth of experience has informed the forensics studied and utilised by the Cambridge research think tank in creating the Automated Deep Behavioral Networks technology.
Inventing the new architecture would have been impressive enough. The achievement becomes even more remarkable when one reconnects with the core challenge.
In terms of cards and payments, until now, deep learning technology has been incapable of understanding time gaps in transaction flow.
To ensure detection success in this particular field, memory cells need to recall a range of events, critical to accurately predicting behaviour. Automated Deep Behavioral Networks solves this challenge for a range of users and uses.
Deep learning, which is a subset of Machine Learning, has become widely used in other industries – for example in language processing where deep learning algorithms predict the next word in a sentence.
Deep learning adoption by Featurespace’s sector has been limited due the technology’s lack of memory relating to the complexity of time as a factor in accurately classifying and predicting good from bad transactions.
Featurespace has solved this complex challenge by inventing ADBN specifically for its industry.
Performance tests have been extensive on all payment types including cards and ACH/BACS, wires and peer to peer transactions. Essential system performance tests have also been conducted to ensure that enterprise customers who depend on their technology for business critical, real-time decisioning can have confidence that ADBNs provide stable scoring with high throughput and low latency response times, even under surge conditions.
In terms of analytics, the Automatic Deep Behavioral Networks model feeds data in what Excell terms a “more discriminatory way” to be able to determine whether to allow or disallow a transaction, as it occurs.
The conventional wisdom may be that faster payments equal faster fraud; one hallmark of real-time payments is that those transactions are irrevocable. Featurespace’s new architecture closes the zone of uncertainty for customers in this regard.
Excell says the need for Featurespace technology has never been greater and it has experienced a continuing spike in demand on every Continent.
Just as its territorial expansion has blossomed, so the headcount has soared and Business Weekly can reveal that the payroll of 330 people currently will top 400 by the end of the year.
Excell recalls that soon after the dawn of the business, after he and the inspirational Bill Fitzgerald founded it with high hopes, he did the payroll for the five people working at Featurespace and wondered how it would be covered.
Nor did he imagine how much legs there was in the technology, although he soon realised the business would never have the luxury of standing still. He certainly did not envisage that he would be addressing the world’s media in February 2021 unveiling the mind-boggling ML technology Featurespace has now conceived and is set to commercialise.
Excell shared a staggering statistic with me: In the US, 40 per cent of Americans have less than $300 in their savings accounts. Online transactions predominate. As a microcosm of the global marketplace, Featurespace would appear to have a huge opportunity for massive growth under its own tenure for many years to come.
While it has thriving US and Singapore operations, Cambridge remains the global headquarters for Featurespace as it utilises brainpower from the University among other sources.
It’s a model built to last as the Cambridge R & D team continues to push the boundaries of what is possible in terms of continuous evolution for its proprietary and potent iP. That’s real and tangible comfort in an era where plastic reigns.