Enabling Ubiquitous Artificial Intelligence with Algorithm-Hardware Co-design
“Can machines think?”, the question brought up by Alan Turing, has led to the development of the field of brain-inspired computing, wherein researchers have put substantial effort in building smarter devices and technology, that have the potential of human-like understanding. However, there still remains a large efficiency gap (for instance, several orders-of-magnitude energy gap) between the human brain and computers, that attempt to emulate even some facets of its functionality.
In this talk, 1) I will describe design techniques exploiting the inherent variability in the difficulty of input data to scale down the computational requirements of a deep learning network with minimal impact on performance. 2) I will also delve into the biologically plausible (and computationally efficient) spiking neural networks and describe the advantage of a temporal learning scheme to address catastrophic forgetting. I will present our proposed ‘Learning to Forget’ rule that offers a promising solution for real-time lifelong learning without expensive re-training. In conclusion, I will discuss how algorithm-hardware co-design techniques hold promise for understanding the energy-accuracy tradeoff, as well as, gauging the robustness of learning systems.
Ph.D. Candidate, Purdue University on February 28, 2019 at 10:15 AM in 454 Monteith Research Center.
Priyadarshini Panda is a final year PhD student at Purdue University in the Electrical & Computer Engineering Department working with Prof. Kaushik Roy. She received her Bachelor’s in Electrical & Electronics and Master’s in Physics from BITS Pilani, India in July 2013. Priya’s research interests span energy-efficient deep learning, neuromorphic computing and adversarial security of learning systems. She won the outstanding poster award for her work on ‘Opportunities and Challenges with Liquid State Machines’ in C-BRIC Annual Review 2018 and honorable mention for her work on ‘Efficient Gesture Recognition Learning’ in Intel Labs Intern Showcase 2017.
The Department of Electrical and Computer Engineering, sometimes in conjunction with other NC State departments makes an ongoing effort to invite the brightest experts from academia, business, and research to come speak and share their ideas, experiences, and research with faculty, students, and alumni at NC State.