Brain-Inspired Hyperdimensional Computing: An Efficient Cognitive Machine
Halls department, Hall 4
Wednesday, 26 December 2018
15:15 - 16:15
We live in a world where technological advances are continually creating more data. By the year 2020, about 1.7 megabytes of new information will be created every second for each human on the planet. With the emergence of the Internet of Things, devices will generate massive data streams demanding services that pose huge technical challenges due to limited device resources. Sending all the data to the cloud for processing is not scalable, cannot guarantee the real-time response, and is often not desirable due to privacy and security concerns. To achieve real-time performance with high energy efficiency, we need to rethink not only how we can accelerate machine learning algorithms in hardware, but also we need to redesign the algorithms themselves using strategies that more closely model the ultimate efficient learning machine: the human brain. In this talk, we present a brain-inspired Hyperdimensional (HD) computing as an alternative computing paradigm. HD computing performs cognitive tasks by representing the values as patterns of neuron's activity in a high dimensional space. This approach enables a robust and highly efficient implementation of the most commonly used learning algorithms in both software and hardware including classification, clustering, regression, and recommendation systems.
Mohsen Imani is a PhD candidate at Computer Science and Engineering Department at UC San Diego. He is a project leader at System Energy-Efficient Laboratory (SeeLab), mentoring over 20 undergraduate and graduate students. He has published over 80 papers at top tiers conferences and journals including HPCA, DAC, DATE, and FPGA. He has also gotten several awards including the best doctoral award for excellence in research, nomination for the best leadership award, and powell fellowship award at UC San Diego. Mr. Imani research interests are brain-inspired computing, computer architecture, and machine learning.