Graduate Seminar with Rohit Sahu: Towards Enabling Pervasive Extreme Edge Intelligence


March 20, 2024    
1:10 pm - 2:00 pm


3043 ECpE Bldg Addition
Coover Hall, Ames, IA

Event Type

Title: Towards Enabling Pervasive Extreme Edge Intelligence

Abstract: Pervasive Intelligence, which involves ubiquitous deployment of smart sensor nodes has a has potential to revolutionize everything from agriculture to industries to manufacturing and beyond. Neural networks, which are shown to be highly effective in common tasks such as image, speech, and including time series classification, are natural candidates for enabling such intelligence on pervasive edge devices. However, since the application often requires deployment of large numbers of nodes in hard to access locations such as farms, bridges and even space they need to be extremely low maintenance and low cost. Thus, these devices are limited to using ultra-low power processors which makes them severely resource constrained in compute and memory capability. Additionally, they are often powered solely by energy harvested from ambient environmental sources such as radio, solar, thermal, and including vibration energy — that is, these nodes are batteryless. Due to the unpredictable and time-varying nature of energy harvesting batteryless nodes often operate intermittently with very short operating and varying off times. These characteristics are at significant odds to the memory, compute, and application latency requirements of neural network inference. In his seminar today, Rohit will provide an overview of these challenges and his proposed solutions to overcome them.

Bio: Rohit is a PhD candidate at Iowa State University. Rohit got his bachelor’s degree in Electronics and Communication Engineering from India. He then worked for two years as a Systems Engineer. During his Ph.D., he has also worked as an AI compiler research intern at LatentAI. Rohit was awarded a teaching excellence award (TEA) for his contributions to the development of “CPRE 487/587: Hardware Design for Machine learning”. His research broadly focuses on developing efficient neural network architecture search, partitioning, and deployment strategies for severely resource constrained edge devices such as ultra-low power MCU’s and FPGA’s. Rohit envisions deployment of cloud independent smart sensor nodes equipped with artificial intelligence (AI) that could take intelligent decisions both locally, collaboratively and in real time.


Please click this URL to start or join.

Or, go to and enter meeting ID: 968 1097 2944 and password: 334840

Join from dial-in phone line:

Dial: +1 309 205 3325 or +1 312 626 6799

Meeting ID: 968 1097 2944

Participant ID: Shown after joining the meeting

International numbers available: