We present Duplex- a methodology for search and optimization. Duplex can optimize nonconvex, nonlinear functions as well as functionals. It consistently outperforms known numerical methods in optimization, and can provide good solutions to functional optimization problems that are known to be highly complex for optimization techniques. Duplex is based on the random tree data structure. Duplex's efficiency is due to partitioning and separating the problem space into multiple smaller spaces such as input, state and the function space. Duplex simultaneously controls, biases and monitors the growth of random trees in the partitioned spaces.
We have applied Duplex to solve practical problems in analog and mixed signal validation like directed input stimuli generation, compressing analog stress tests, worst-case eye diagram analysis, and performance optimization. In each case, we consistently show orders of magnitude improvement over the state-of-the-art.
Shobha Vasudevan is an associate professor in the departments of Electrical and Computer Engineering and Computer Science at the University of Illinois at Urbana-Champaign. Her research interests span hardware systems and algorithms. In hardware- verification, analog/mixed signal verification and post Silicon validation; in algorithms- non-convex and functional optimization, graph search, feature engineering and their applications to data analysis. She has won several best paper awards including one at DAC 2014. She has won other honors like ACM SIGDA Outstanding New Faculty Award, IEEE early career award, IBM faculty award, Dean's award for research excellence in UIUC, NSF CAREER award, and a UIUC award for service to women in engineering. GoldMine, a verification software from her group has been developed into a commercial product since 2014 and licensed by multiple semiconductor and electronic design automation companies from UIUC.