UMass Amherst Joins DARPA’s Surrogate Modeling AI Project

The University of Massachusetts Amherst has joined a Department of Defense project to develop an artificial intelligence machine learning framework that can rapidly simulate complex systems by generating surrogate models.

UMass Amherst said its Biologically Inspired Neural and Dynamical Systems Laboratory will support the Defense Advanced Research Projects Agency’s DITTO Intelligent Auto-Generation and Composition of Surrogate Models project.

“Our proposed MOKA system will incorporate knowledge about self, its inputs, and other components it may interface with, already starting at the neural architecture. This will lead to computing that is informed of itself and its environment,” said Hava Hiegelmann, director of Umass Amherst BINDS laboratory.

MOKA, or the Modular Knowledge AI system, was designed to combine AI “meta-cognition” with the design of a neural compiler that aggregates models into a modular system.

Hiegelmann said the technology will greatly reduce neural networks’ reliance on training and the time spent on it while significantly improving capabilities and accuracy.

By incorporating knowledge into neural architecture, MOKA reduces the amount of time and resources that can be lost to design flaws in hardware simulations, the university said.

UMass Amherst said it will collaborate with Lockheed Martin Advanced Technology Laboratories on the DARPA project. ATL is a Lockheed applied research and development center that specializes in applying technologies in human systems, robotics, automation, spectrum systems and data analytics.

Janet Wedgwood, Lockheed’s lead engineer on the DITTO project, said the company will combine its experience in integrated circuits design and testing with the BINDS laboratory’s machine learning neural networks. 

The BINDS laboratory is part of UMass Amherst’s computer science department. BINDS advances research in biological computing, AI, neuroscience and other related fields.