Various incremental learning (IL) approaches have been proposed to help deep learning models learn new tasks/classes continuously without forgetting what was learned previously (i.e., avoid catastrophic forgetting). With the growing number of deployed applications that need to dynamically incorporate new tasks and changing input distribution from users, the ability of IL on-device becomes essential for both efficiency and user privacy. However, high computational costs of IL hinder the deployment of IL on-device.

In this work, we want to explore the potential of the recently developed TensorFlow.js which utilizes the WebGL backend to enable the GPU training on an iPhone. Can we overcome the computation bottleneck using phone GPUs?

Contact: Interested students can actively discuss and suggest their ideas and also contact Young D. Kwon (ydk21@cam.ac.uk), Dr. Jagmohan Chauhan (jc2161@cam.ac.uk)/J.Chauhan@soton.ac.uk, and Prof. Pan Hui (panhui@cse.ust.hk) for more details.

Reference papers (can be a good starting point)
[1] 2021. HotMobile. SplitEasy: A Practical Approach for Training ML models on Mobile Devices. Kamalesh Palanisamy et al.