AI/ML SECURITY

One of Undernet founder's first real job in machine learning space was designing automated interference mitigation systems for fighter jets air-to-air combat at the famous Institute of Air and Space Defense in the Soviet Union. One pilot, one fighter jet, and one radar. One neural network. While other systems were using the radars to maintain multiple targets, these same radar systems needed to detect interference activities, their frequencies, types, and intensity, qualify it all, and respond immediately, possibly giving a pilot just a few more milliseconds extra to react. Year was 1985, if your intelligence did not fit in the cockpit, it did not work. There were no football field sized ground support systems, high performance solid state drives, and certainly no gigabytes and gigabytes of data flying through the air. Science was not afraid to look for patterns across infinite spaces.  Fast forward two decades of AI winter, people disappearing in the middle of the night, Butyrka prison, Serbsky Institute, and stupidly dangerous defection on the third anniversary of Vadim Knizhnik's untimely and mysterious death on the Moscow metro; we face the same question today. How to refine the cognitive neural network between human and the machine? Ho to embed security into the models, into data, into operator verifiers from the very start. How to protect the learning system or operator without collapsing. Ever. Thank You Uncle Vadim, May you have eternal peace and happiness in knowing that your math works even better today! One of our research engineers described it as if Quantum Fields Theory and Complexity Theory finally had a baby. They did, and she turned out to be a great kid. Reach out to learn more. This is all real.