Guaranteeing AI Robustness against Deception (GARD)

2020: Intel headed the project

At the beginning of April, 2020 the chipmaker of Intel headed a new initiative of a military and research wing of the U.S. Government (DARPA) to provide cyber defense of the models of machine learning used in work of the Pentagon.

Cyber attacks can interfere with algorithms of machine learning and change results of object recognition or work of autonomous vehicles. Effects of such interventions, even the most imperceptible, can be catastrophic.

Intel headed the project of the Pentagon on protection of military against cyber attacks to the systems of machine learning

The Pentagon hopes to secure the algorithms, having developed the Guaranteeing AI Robustness against Deception (GARD) program. The existing means of protecting from the attacks, as a rule, are based on the predeterminated rules, but the Pentagon hopes that GARD will be able to become the self-training rule for counteraction to different types of the attacks. Intel stated that it will act as the general contractor within this program together with Georgia Tech.

The chief engineer of Intel Labs Jason Martin who headed a command of Intel of GARD told that both companies will work together, trying "improve capability of AI and algorithms of machine learning to react to the competitive attacks". At the first stage of the program optimization of technologies of detection of objects using space, time and semantic coordination both for still images, and for video will become the main objective of Intel.

The example of a protective system on the basis of scenarios which we intend to create is our own immune system which reveals the attacks, stops them and remembers to provide more effective answer during future battles, - the program manager at the Information and innovation office of the Pentagon doctor Hava Siegelmann explained. - We should guarantee that nobody will be able to deceive our algorithms of machine learning.[1]

Robotics

Notes