Scalable Exploration of Al-Enabled, Reliable, and Cost-Effective Architectures (SEARCH) Proposal

Abstract

State-of-the-art artificial intelligence (AI) techniques become increasingly more important for situational awareness and decision-making in cyber-physical systems (CPSs). However, their adoption can substantially increase system complexity and uncertainty, and make it difficult to design cost-effective systems with strong guarantees of correctness, dependability, and compliance with regulations. Reasoning about AI-enabled CPSs performing complex functions in highly dynamic and unpredictable environments is challenging. Research efforts aiming to provide assurance about their operation or verify robustness of AI artifacts have significantly increased over the past few years. However, holistic system design approaches that can effectively capture the uncertainty induced by AI-enabled components and their environments, and scale from component-level point solutions to more complex, heterogeneous CPSs are still elusive. This project develops SEARCH (Scalable Exploration of AI-Enabled, Reliable, and Cost-Effective Architectures), a framework for modeling, specification, and scalable design space exploration of AI-enabled CPSs that can support different architecture trade-offs and design objectives. If successful, SEARCH will provide the foundations for rapid, compositional, certified design of adaptive and resilient CPSs with applications ranging from autonomous vehicles and robotics to industrial automation and medical devices. Rather than focusing on AI elements in isolation, SEARCH will develop compact stochastic abstractions and specification formalisms that can capture the behaviors of AI-enabled components conditioned on the environment of operation and the overall system objectives and can propagate computationally tractable representations of uncertainty at different abstraction levels. Component models will be expressed by assume-guarantee contracts to favor modular certification of designs made of independently developed parts. Incremental and hierarchical exploration algorithms will map system-level requirements to netlists of components that satisfy the requirements and optimize a set of quality factors. The proposed framework and tools will be validated in the context of system development efforts pursued in collaboration with system engineering companies, targeting autonomous vehicles and avionics applications.

Document Details

Document Type
DoD Grant Award
Publication Date
Jul 23, 2020
Source ID
HR00112010003

Entities

People

  • Pierluigi Nuzzo

Organizations

  • Defense Advanced Research Projects Agency
  • University of Southern California

Tags

Fields of Study

  • Computer science
  • Engineering

Readers

  • Distributed Systems and Data Platform Development
  • Software Engineering.

Technology Areas

  • AI & ML
  • AI & ML - DoD AI Strategy
  • AI & ML - Machine Learning Algorithms
  • Autonomy
  • Autonomy - Autonomous System Control
  • Cyber
  • Space