PhD Forum Abstract: Deep Learning Model Composition for Edge Intelligence

The high demands the Deep Learning based applications pose on the computation resources stagger their deployment, particularly in the constrained settings of the end devices. As a result, device-edge collaborative computation setups have emerged as a popular solution these days for the utilization o...

Full description

Saved in:
Bibliographic Details
Main Author: Nimi, Sumaiya Tabassum
Format: Conference Proceeding
Language:English
Subjects:
Analytical models
Collaboration
Computational modeling
Conferences
Deep learning
Multitasking
Training data
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:The high demands the Deep Learning based applications pose on the computation resources stagger their deployment, particularly in the constrained settings of the end devices. As a result, device-edge collaborative computation setups have emerged as a popular solution these days for the utilization of AI capabilities at the end devices. However we noted several limitations in the existing solutions: @ It was not considered whether it was not necessary to analyze all the incoming samples, particularly considering that in real-life experience many of the incoming samples could be Out-of-Distribution (OOD) to the training data distribution of the considered deep learning models and @ Device-edge collaborative inference for multitasking deep learning models was not studied. In this extended abstract, we explore two device-edge collaborative setups we have developed towards addressing these limitations.
ISSN:2693-8340
DOI:10.1109/SMARTCOMP55677.2022.00051