LOGiCS: Learning-Driven Oracle-Guided Compositional Symbiotic Design of Cyber-Physical Systems

About

LOGiCS is a project funded by the DARPA program on Symbiotic Design of Cyber-Physical Systems. The aim of the project is to develop design tools and techniques that augment human designers with an AI co-designer towards achieving significant improvements in design productivity as well as design 'surprises'. We believe that design surprises come from two main sources: (i) lateral thinking, guided through expert human insight but powered by pattern recognition and computational search, and (ii) design corners, defined by specifications and intent expressed by humans but usually identified by computational engines for analysis and verification. In this project, we aim to blend computationally-enabled lateral thinking with human-guided design corner epxloration. To this end, we take a learning-driven, oracle-guided, approach to Symbiotic CPS Design. Within LOGiCS, we aim to address three challenges:

The first challenge is to design for the interaction between design algorithms that define and explore the design space and oracles, both human and computational, that guide the design algorithms by answering queries and providing key inputs. A symbiotic design loop requires languages and representations to succinctly capture a wide design space along with multi-domain design contexts, in formalisms efficiently learnable from data and easily interpreted and guided by expert human designers.
The second challenge is to improve design productivity through compositional methods, cross-domain reasoning, and synthesis algorithms. Here, the theory of contract-based design, suitably extended to multi-domain heterogeneous CPS, and augmented to reason about specifications learned from data, is playing a pivotal role.
Thirdly, we are developing novel algorithmic methods for learning, optimization, and design space exploration for the multi-objective design problems arising in CPS. In addition, to reduce designer and algorithmic bias, we are employing our work on algorithmic improvisation to generate and explore a diverse space of designs.

Overview of the LOGiCS Approach

Figure 1 shows the three main areas that we will work on: representation of the design problem, design procedures, and computational engines. The concept of oracles is central in the way that these different aspects are combined, and in the way that we interact with the human designers. The figure also clarifies the interfaces with our project with project partners in the other technical areas (TA 2 and TA 3).