Isaac Sánchez Leal presented his work at his halftime seminar

Mon 20 Mar 2023 09:31

On the 14th of March PhD-student Isaac Sánchez presented his work on a half-time seminar with the title " Embedded IoT Implementation of Convolutional Neural Networks using Intelligence Partitioning”. Now Isaac continues his work toward a doctoral thesis defense.

A man in a white shirt stands in front of a large screen and presents.

He discussed the challenges faced by computer vision systems in IoT sensor nodes due to main node constrains, energy, latency, and memory. Previous works have attempted to address this issue by reducing model size or distributing processing load to other nodes. However, the impact of these approaches on energy consumption and system latency remains unclear.

His research seeks to contribute to this understanding by exploring the impact of partitioning a Convolutional Neural Network (CNN) between an IoT node and an edge device, with a focus on the partition point, quantization method, and communication technology. Specifically, he identifies possible partitioning points and apply quantization and compression techniques to reduce the data sent between the two partitions. Results show that a 99.8% reduction in transmitted data while keeping a high accuracy of the CNN model. Therefore, the quantization of data facilitates CNN network partitioning with significant reductions in overall latency and node energy consumption.

As a future work, he’d like to deepen in data reduction and pruning methods to find a new CNN partitioning method that facilitates the implementation of CNNs in IoT devices.


Recommended

The page was updated 3/20/2023