.png){kind=icon}
Return
TinyML Based Physical Computing for Education
.png){kind=icon}
PRE
AI
DOI：10.1109/INDICON56171.2022.10039922.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Neural networks, a type of machine learning, are generally big in size, from tens of MBs to a few GBs because they have millions of parameters such as weights and biases. They also need huge number of computations leading to the requirement of high memory and computational capacity in systems running them. However, edge devices have limited resources such as memory and computing power, have limited internet connectivity and are powered by limited battery capacity. These big models need high end processors, graphics processing units and hardware accelerators to run their training and inference phases. So, these models cannot fit in the resource-constrained edge devices. However, edge devices are excellent for Physical Computing which offers an interactive paradigm for learning. At the same time, using design paradigms like TinyML, the size of the models can be reduced enabling their deployment for inference on edge devices. This paper first discusses TinyML, with Raspberry Pi (RPi) as an example edge device, and then focuses on using such devices running tiny models to study opportunities for application to education, primarily tertiary education. We demonstrate such an application that uses You Look Only Once (YOLOv3) model to provide on-demand informative content for images captured through RPi camera thus demonstrating learning through physical computing.
Keywords:
Machine learning
edge device
edgeAI
tinyML
raspberry pi,physical computing
Journal
I
IF:
0
Papers:
25
Citations:
0
.png){kind=icon}
