Bayesian Continual Learning on Edge Devices
Project Description
While recent advancements of Continual Learning (CL) have shown an impressive performance on learning a new task and/or data distribution in a variety of applications such as computer vision and mobile sensing tasks, being able to provide more information in addition to a probability of a certain class becomes favorable in such areas as healthcare and mobile sensing, e.g., gait monitoring of Parkison’s disease patients.
Thus, to make a CL model more intuitive to human interpretation, Bayesian concepts have been utilized to provide uncertainty estimation for outputs of the model. However, the Bayesian CL suffers from high computational demands in training and inference. Thus, in this work, we plan to examine these challenges in algorithmic and system aspects and attempt to address them.
Contact: If you are interested in this project, please contact Young D. Kwon (ydk21@cam.ac.uk) and Pan Hui (panhui@cse.ust.hk) for more details.
Reference papers (can be a good starting point)
[1] 2019. ICLR. Uncertainty-guided Continual Learning with Bayesian Neural Networks. Sayna Ebrahimi et al.
[2] 2021. SEC. The Benefit of the Doubt: Uncertainty Aware Sensing for Edge Computing Platforms. Lorena Qendro et al.
[3] 2019. ArXiv. Latent Replay for Real-Time Continual Learning. Lorenzo Pellegrini et al.
Thus, to make a CL model more intuitive to human interpretation, Bayesian concepts have been utilized to provide uncertainty estimation for outputs of the model. However, the Bayesian CL suffers from high computational demands in training and inference. Thus, in this work, we plan to examine these challenges in algorithmic and system aspects and attempt to address them.
Contact: If you are interested in this project, please contact Young D. Kwon (ydk21@cam.ac.uk) and Pan Hui (panhui@cse.ust.hk) for more details.
Reference papers (can be a good starting point)
[1] 2019. ICLR. Uncertainty-guided Continual Learning with Bayesian Neural Networks. Sayna Ebrahimi et al.
[2] 2021. SEC. The Benefit of the Doubt: Uncertainty Aware Sensing for Edge Computing Platforms. Lorena Qendro et al.
[3] 2019. ArXiv. Latent Replay for Real-Time Continual Learning. Lorenzo Pellegrini et al.
Supervisor
HUI Pan
Quota
3
Course type
UROP1000
UROP1100
UROP2100
UROP3100
UROP4100
Applicant's Roles
Required skills/aptitudes to successfully complete the project are:
1. Self-motivated and proactive, which means you should be able to set your own milestones and finish them on time.
2. Basic knowledge of neural networks, bayesian methods, and embedded software/hardware are plus.
1. Self-motivated and proactive, which means you should be able to set your own milestones and finish them on time.
2. Basic knowledge of neural networks, bayesian methods, and embedded software/hardware are plus.
Applicant's Learning Objectives
Benefits of participating in this project are:
1. Opportunity to gain hands-on experience in on-device ML/DL research.
2. Opportunity to tackle a challenging and novel problem that no one has solved before.
3. Opportunity for having a top-tier academic publication, depending on the quality and the amount of novel contribution of the project.
1. Opportunity to gain hands-on experience in on-device ML/DL research.
2. Opportunity to tackle a challenging and novel problem that no one has solved before.
3. Opportunity for having a top-tier academic publication, depending on the quality and the amount of novel contribution of the project.
Complexity of the project
Challenging