January 24, 2019: Seminar by Prof. Asim Roy

Machine Learning at the Edge of IoT using the Parallel Computing Framework of FPGAs and GPUs
by
 
Asim Roy
Arizona State University 
USA
 
Date: Thursday, January 24, 2019
Time: 15:00 HRS
Venue: ECSU Seminar Room, 9th Floor, S. N. Bose Bhavan (Library Building)
 
Abstract
 
The global market size of Internet of Things (IoT) is predicted to be in the trillions of dollars within a few years andmachine learning will become a critical component in these IoT systems. In the current IoT architecture, data from sensorsis uploaded to the cloud for processing, including machine learning. However, thevolume of data generated by IoTsensors is enormous and many organizations would prefer to process the data closer to the source where it’s generated in order to reduce the cost of uploading to the cloud. Hence, there is tremendous focus on computing at the edge of IoT (edge computing) and that includes machine learning at the edge.
In this talk, I will provide an overview of a machine learning system that is meant for deployment at the edge of IoT. The system uses neural network algorithms for feature ranking and selection, classification, function approximation (nonlinear regression, time-series), clustering and anomaly detection. The basic building block for all these different algorithms is the Kohonen SOM. We have implemented these algorithms on FPGA and GPU devices in order to exploit the opportunity for massively parallel computing on these devices.Thus, wecan build hundreds of models in parallel in different feature spaces and using different hyperparameter settings.These machine learningsystems at the IoT edgeare expected to learn (train) models in real-time from high-velocity streaming data. We also use boosting and stacking to build a composite ensemble model where the component models can be in different feature spaces. With the advent of deep learning and its wide deployment, there is some push back against complex modelsbecause no one can explain how it works. This has resulted in the call for Explainable AI. Our machine learning system has Explainable AI in the sense that it can extract simple IF-THEN rules from the models very easily.
 

Brief Biography: Asim Roy, Ph.D. is a Professor of Information Systems at Arizona State University. He earned his B.E. in Mechanical Engineering from Calcutta University, India, his M.S. in Operations Research from Case Western Reserve University, Cleveland, Ohio, and his Ph.D. in Operations Research from the University of Texas at Austin. He has been a Visiting Scholar at Stanford University, visiting Professor David Rumelhart in the Psychology Department. Asim is currently working on hardware-based (GPU, FPGA-based) machine learning for real-time learning from streaming data at the edge of the Internet of Things (IoT). He is also working on Explainable AI.
 
Asim is a member of the Board of Governors of the International Neural Network Society (INNS) and founded two sections of INNS. One section is on Autonomous Machine Learning (AML Section) and the other on Big Data Analytics (Big Data Section). He started the Big Data conference series of INNS and was the General Co-Chair of the first INNS Conference on Big Data in San Francisco in 2015. Heis the IJCNN General Chair for WCCI 2020 in Glasgow, UK (https://www.wcci2020.org/).He is the Senior Editor of Big Data Analyticsand serves on the editorial boards of Neural Networks,and Cognitive Computation. He was the Program Chair for the ORSA/TIMS (Operations Research Society of America / The Institute of Management Sciences) National meeting in Las Vegas and General Chair of the ORSA/TIMS National meeting in Phoenix.
 
Asim’s brain theories: Asim has published four theories of the brain. The theories postulate that the brain encodes abstractions in single cells or neurons. Neurophysiological evidence is provided in support of these theories.
 
All are cordially invited.
 
Dipti Prasad Mukherjee
Head, Electronics and Communication Sciences Unit