Data Science Day @ Columbia University

"Due to the anticipated slowdown of speed and efficiency improvements in integrated circuits, researchers are looking for new scalable ways to get useful computation with existing silicon technology. Our work explores using analog electronic circuits to assist conventional digital computers to obtain high performance and low energy computation.
Our prototype analog co-processor solves nonlinear and linear systems of equations, delivering approximate solutions which are useful in physical simulations and machine learning tasks. Commonly perceived downsides of analog computing, such as low precision and accuracy, limited problem sizes, and difficulty in programming are all compensated for using methods we discuss. Based on our findings, small-scale uses of analog computing can bring efficient physics simulation to energy-efficient, Internet-of-Things devices. On the other hand, large-scale uses of analog computing can speed up the training of machine learning applications.
"