Marian Verhelst
Verhelst moved to Intel in Portland in 2008 and returned to Leuven after about three years. Her work focuses on chip architectures that use less energy. Verhelst and Bert Moons published a precision-scalable convolutional-network processor in 2016. The work combines parallel processing, sparse computation and adjustable precision to reduce power while maintaining throughput. The interview explains how her team designs neural-processing units for tasks such as image recognition. These specialized chips support AI calculations on small devices while limiting battery use.
Timeline
-
2008
Industrial chip research
Verhelst moved to Intel in Portland in 2008 and returned to Leuven after about three years. Her work focuses on chip architectures that use less energy.
-
2016
A processor for convolutional networks
Verhelst and Bert Moons published a precision-scalable convolutional-network processor in 2016. The work combines parallel processing, sparse computation and adjustable precision to reduce power while maintaining throughput.
-
As of October 2026
Dedicated neural processing
The interview explains how her team designs neural-processing units for tasks such as image recognition. These specialized chips support AI calculations on small devices while limiting battery use.
Sources and credits
Photo: set.kuleuven.be; creator not stated. License: Not stated by source. Source: set.kuleuven.be. Cropped to a square and resized. Ask us to remove or replace this photo.