Novelty Search & Secure Computing for Robust AI in Open World

Artificial intelligence (AI) has become a pervasive component of data-driven interactive applications (i.e., interactive games, vision-based vehicular autonomy, autonomous military equipment, etc.) for enhancing human experiences in an era of digital revolution. Such applications require interaction in open-world environments, which may arise multiple novel scenarios alongside posing computing threats to the AI component of the system. Compromise in AI components or computation may cause severe life-threatening events in safety-critical applications such as AI-based autonomous navigation, space platforms, etc. Therefore, establishing the trustworthiness of AI components by specifying the robustness, quantifying the task complexity, and computing accuracy in the presence of persistent threats in the open world is inevitable for amplifying its widespread adoption in human-centric critical applications. To address the problems in the open world, we propose (i) a novelty specification framework for outlining the robustness of AI components (or agents) in the presence of novelties, (ii) a diversity and redundancy-based adaptive and fault tolerant computing framework for heterogeneous computing platform and (iii) a domain complexity quantification framework for distributed learning environments.