Security

Don't know if been posted
Dark Trace, Spikenet, AIMergence: these AI and security start-ups

20 April 2017 , Written by Informatelecom

Deep learning / machine learning helps to identify and stop a cyber attack and then repair its damage.  It is also used for the recognition of faces in a crowd.  Many start-ups are emerging in this innovative sector.

Security as well as cyber security are high-use artificial intelligence domains.  This is evidenced by Emmanuel Meriot , director in France of the British Dark Trace , whose Antigena solution has equipped, over the last four years, 2,000 companies in the world . "Our algorithms offer an immune system against cyber attacks. Above all, since last year, they stopped the attacks in progress.  Especially those of the Ransomware that encrypt the data of the infected hard disks before the ransom request.  These algorithms are embedded in machines that analyze the normal data traffic in the company network in order to create a reference model.  As soon as an anomaly appears, even in the form of weak signals, it will be detected and generate an alert.  Especially in the case of a software program, the suspect encryption of the hard disks will be stopped immediately.  Moreover, after the management of the crisis, the administrator can conduct the (forensic) investigation to trace the criminal .
Auxiliaries also of the fight against terrorism, the solutions of " machine learning " want to make speak the video images. "Our software understands and analyzes what happens in an image within a real-time stream , " explains Stéphane Bidault , president of TEB (180 million euros in turnover in 2016).  Then, the system extracts useful information: counting people who pass by the camera, analyzing objects (abandoned parcels, for example), automatic license plate reading, detection of abnormal people or assemblages.  The images are then transformed into data that will be interpreted in relation to a normal situation in order to trigger actions.  As the prepositioning of intervention forces or the guidance of the flow of people to evacuate ...
For their part, the neural networks of the Toulouse-based Spikenet are used, among others, by the Directorate General of Internal Security (DGSI) and the  Prefecture of police of Paris.  The algorithms succeed a tour de force. "Their learning does not need to be constantly supervised by humans. Our neural networks are rewiring on their own , says Hung Do-Duy, president of Spikenet Technology SAS, who has been working for 11 years with the Ministry of the Interior. Our research on unsupervised neural networks began twenty years ago. "
AIMergence launches an autonomous robot
Recently, Spikenet passed a test in real conditions with video surveillance cameras in Paris.  The faces of 150,000 individuals at the rate of 5 to 200 faces per individual (1.5 million images in total) were loaded into the software to find a single person.  You might as well look for a needle in a haystack. "In just a few seconds, the face of the person sought came out 8 times out of the first 20 results," said Hung Do-Duy, who for the last three years has carried his algorithms on hardware platforms.
"We started on Raspberry Pi [nano-computers the size of a credit card at less than 50 euros, Editor's note]. We now realize neuromorphic electronic chips on reprogrammable integrated circuits (FPGA). " A strategy spotted by the American Brainchips who bought the Toulouse company last September.  What to build networks of neural networks by integrating these chips with servers, network equipment, cameras, robots ...

Learn More