NtechLab the Algorithm on recognition of dipfeyk

Main articles:

• DeepFake
• Technologies of biometric identification
• Systems of face recognition

2020: Recognition steady against deception using so-called competitive to the attacks

NtechLab, the company in the field of solutions for video analysts on a basis neural networks and the technology partner of State corporation Rostec, reported on December 11, 2020 that algorithm the companies on recognition dipfeyk it was called the steadiest against deception using so-called competitive attacks. Such conclusion was drawn in the scientific article published on the website of Cornell University behind authorship of researchers from laboratory Facebook AI and the University of California in San Diego.

Dipfeyk (deepfake) — the picture synthesis technique based on technology of neuronets. Are used by Dipfeyki for connection and imposing of the existing images and video on source images or videos. Can be used by Dipfeyki for replacement of certain image elements by desirable images, in particular for creation of rollers which can be used further for political manipulation, personal benefit or revenge. The danger of uncontrollable production of dipfeyk is recognized at the international level. In particular, in the USA and China a number of the laws regulating production of similar videos was adopted.

In June, 2020 the results of the international competition in recognition of dipfeyk on video Deepfake Detection Challenge held by Amazon, Facebook and Microsoft were summed up. According to the results of a competition the algorithm NtechLab took the third prize. After summing up tender the team of researchers from AI Facebook laboratory and the University of California in San Diego subjected three algorithms which took the highest places in Deepfake Detection Challenge, to additional tests - they decided to check as far as winner algorithms are steady against the so-called "competitive attacks". Within "the competitive attack" the image is exposed hardly noticeable to a human eye to distortions with the purpose to deceive a neuronet. In this research "competitive attacks" were applied not to allow an algorithm to define whether is on video dipfeyk or not.

The algorithm NtechLab is steadier against the competitive attacks in comparison with other models.

The engineer of NtechLab the Azat Davletshin, the author of the algorithm provided on Deepfake Detection Challenge on recognition of dipfeyk, initially set the task to create not just model for successful participation in tender, and steadily working solution steady against different types of the attacks. The source code of an algorithm NtechLab, in accordance with the terms tender published in open access, showed worthy result in attempts to deceive him even without additional training at broad bases of photos. We hope that its use will enter a contribution to world fight against uncontrollable production of dipfeyk - the co-founder of NtechLab and the director R&D told the company Artyom Kukharenko.

In commercial operation NtechLab is going to use the improved version of an algorithm on recognition of dipfeyk. Practices will help the company to reveal complex cases of fraud when using face recognition in facepay systems for payment of goods and services.