An artificial intelligence-focused research team from Michigan State University in East Lansing was recognized as fourth in the world and first in the United States and Canada at the recent Google MicroNet Challenge in Vancouver, Canada.
The team is from MSU’s Department of Electrical and Computer Engineering and was recognized for its contributions to the challenge’s CIFAR-100 Track at the conference on neural information processing, or NeurIPS’19. The challenge took place Dec. 8-14.
NeurIPS’19 attracted about 13,000 participants from across the globe for artificial intelligence and machine learning. The annual meeting fosters an exchange of research on neural information processing systems from biological, technological, mathematical, and theoretical perspectives.
According to Mi Zhang, the assistant professor who heads the team, the challenge looks for solutions in developing the most efficient deep neural network architecture for resource-constrained devices such as mobile phones and the internet of things.
“The solution provided by the MSU team is built on an efficient backbone design with a combination of sparsification and quantization techniques,” Zhang says. “The MSU team’s solution is now open source. We hope it can push the research area of AI for mobile and IOT forward.”
This is the team’s third international competition win in the past four years. In 2016, the team won the Champion of the NIH Pill Image Recognition Challenge. In 2017, it took third place at the NSF Hearables Challenge.
Other team members include Ph.D. students Yu Zheng, Shen Yan, Xiao Zeng, and Biyi Fang.