FaDO (‘Fault DetectOr’) is an online machine-learning algorithm for detecting anomalies, frauds or faults from monitoring data flows; the FaDO algorithm will observe your data streams and throw alarms every time a deviation from the learned norm is detected. Simple as that. No more. No less.
The FaDo algorithm is designed as a first-line-of-defence. It can handle the raw data streams of the (cyber) battlefield, but it does not take strategical decisions for you. FaDO is devised as a support tool to help your cyber-security experts manage incoming traffic. By inspection of the smarts of FaDO, an expert knows exactly which transactions are passing through, and which are withheld. Hence FaDO is a digital filter – as in the good old engineering tradition – but then for your type of data. It’s designed to support the job of your security experts, and not to overtake any of it.
The FaDO algorithm itself is quite translucent (it can be implemented in 5 lines of Python code: it doesn’t take a degree to get it), but the real smarts emerge from the data it will get exposed to: this is machine learning at work. All this is kept within your own premises without need for extended (storage) hardware. No GDPR troubles, no haunting Patriot Act, no cloudy services needed, nada. You stay in full control of your own digital dust: as far as we’re concerned, you can build an extra impenetrable defence wall around your IT systems. Including military-grade bunkers. Don’t forget to add the Moorish ornaments.
We believe that the perfect company of the future is free from any (storage of) data. Fully transparent, no backlog, no unnecessary history. The FaDO initiative is the first to advocate and realise this – by immediately processing data and then actively forgetting this. The fact that we still do machine learning (and hence artificial intelligence) in such setting is quite novel, but ties in nicely with current concerns.
The FaDO algorithm builds on the counter-intuitive thought that ‘anomalies shape what is normal’. That is, normal behaviour is not defined by the common average, but instead by the large deviators from what was thought to be the average. We think this design principle to be rather intriguing. Quite Swedish, when you come to think of it.
Did I mention that FaDO is multi-variate? It uses the language of mathematics where transactions, items or measurements in the flow are expressed as high-dimensional numerical feature vectors. Give us hundreds of such features, or even thousands, FaDO can handle it. Learning exactly which features are relevant to the problem at hand should be part of the solution, it shouldn’t have to be pre-set. The benefit of multi-variate methods as FaDO is that such selection mechanism is inherently included in the box. It’s really linear algebra in all it’s glory driving this solution.
FaDO is artificial intelligence as it should be: explainable, accountable and provable. That is, FaDO is not a black-box solution, but decisions which are advised can perfectly be traced, reconstructed and be interpreted. Our academical inclination is also very happy about the connection to academic theory. And guess what, theory is not ivory-tower here: it translates into practical quality-of-service guarantees.
Our core competence then is ‘to consult this solution into your own IT system’, and look together with you for appropriate extensions. Then we put back our academical hat and get you the suitable quality-of-service theorems if possible. Mathematical truth is our touchstone to keep expectations within reason. Going out and hunting for the theoretical truth is not really a uneventful walk in the park, but almost 20 years of experience in this academic sport taught us how to manoeuvre this jungle. And in the end, the University is conceived to help society. So we see this initiative as pay-back time for the academicus in us.
Efficient, real-time and high-dimensional … where’s the catch? Well, you shouldn’t let the hacker community adopt FaDO before you do.