Real-time Modeling: Theory versus Real-World Applications

Bayesian inference is a powerful method for extracting statistical information from noisy, corrupted, or non-linear measurements. A growing number of applications rely on real-time Bayesian inference—assessing data and making predictions and models as data comes in—mainly in the fields of wireless communications, imaging, and radar. While state-of-the-art algorithms for Bayesian inference have been designed for time-insensitive tasks, real-time applications commonly rely on simplistic (mostly linear) methods. This disparity between recent theoretical advances, cutting-edge algorithms, and practical circuit realizations is mainly caused by the fast progress on the theory and algorithm sides, and by the limited theoretical expertise of hardware designers.

With this National Science Foundation CAREER award, Christoph Studer, Electrical and Computer Engineering, is working to resolve the dichotomy between recent advances on the theory side and real-world hardware constraints. He’s achieving this by pursuing a bottom-up approach in which hardware limitations drive efforts on the algorithm and theory levels. This unconventional research paradigm requires consideration of the major challenges on all levels, which is the core expertise of Studer’s research group. The team is evaluating modifications on the hardware level that are key for the design of efficient, digital, very-large scale integration circuits and is investigating new algorithms that enable more efficient hardware architectures. The group is also analyzing the impacts of circuit- and algorithm-level optimizations on the complexity and quality of inferences. This work will greatly advance the ability to accurately model and predict outcomes in real-time applications.

Cornell Researchers

Funding Received

$600 Thousand spanning 5 years

Sponsored by

Other Research Sponsored by National Science Foundation