student profile: Mr Shaojun Zhang


Thesis work

Thesis title: Artificial Intelligence at the Edge

Supervisors: Wei LI , Albert ZOMAYA

Thesis abstract:

The Deep Learning scheme has been successuflly applied in the data-center based environment, which always has numerous GPUs and CPUs. Due to the Gigabytes data transmission would occupy rather a lot of network and may related to several provicy problems, it encourages us to carry out DL on Edge devices. However, when it comes to the the Edge environment, which always locates near the edge of network and lack of luxury computing capbility, it always poses several challenges for DL. The first one is that we may not have sufficent labeled data to train a DL model. The second one relates to the training jobs of DL, which may not be carried out in the CPU-based Edge devices. The third one is that the target data domain in the Edge may vary a lot from the source data domain in the data-center, and lacks labels for supervised learning. It would be suitable to train a DL model on source data in the data-center with GPU servers and GPUs, and then apply trainsfer learning on the target data domain which may dosen't have sufficient labels in the Edge environment. So, we apply trainsfer learning to obtain models in the Edge, and compress the massive DL model to rather smaller one.

Note: This profile is for a student at the University of Sydney. Views presented here are not necessarily those of the University.