[Eecs_msee] EECS Research Seminar Talk-Revolutionizing Edge AI: Improved and Automated Model Compression

• Previous message: [Eecs_msee] EECS Research Seminar Talk-Towards Deterministic and Reliable Real-Time Cyber-Physical Systems
• Next message: [Eecs_msee] EECS Research Seminar Talk-Geometric and Geospatial Algorithms under Data Conundrums in Geographic Information Systems
• Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

Title
Revolutionizing Edge AI: Improved and Automated Model Compression
Bio
Kaiqi Zhao is a final-year Ph.D. candidate advised by Prof. Ming Zhao in the School of
Computing and Augmented Intelligence (SCAI) at Arizona State University (ASU). Her research
interests are machine learning (ML) and deep learning (DL) model compression as well as cloud
and edge computing. Central to her research are innovating solutions to optimize large-scale
ML/DL models for Internet-of-Things (IoT) data-driven applications. She has published
first-authored papers at top-tier AI conferences, e.g., AISTATS (2023 and 2024), ICASSP (Oral),
and InterSpeech (Oral), and top-tier edge computing conferences, e.g., SEC, and won the Best
Poster Award at SEC'24. She did three research internships at Amazon Web Services as an
Applied Scientist. One of her internship works about compressing speech recognition models
has been integrated into the Amazon Alex library for production usage. Additionally, she was
awarded the Graduate College Completion Fellowship, the most prestigious award for graduate
students at ASU, and SCAI Doctoral Fellowships in 2022 and 2023.
Abstract
Machine learning models are increasingly employed by smart devices on the edge to support
important applications such as real-time virtual assistants and privacy-preserving healthcare.
However, deploying state-of-the-art (SOTA) deep learning models on devices faces multiple
serious challenges. First, it is infeasible to deploy large models on resource-constrained edge
devices whereas small models cannot achieve the SOTA accuracy. Second, it is difficult to
customize the models according to diverse application requirements in accuracy and speed and
diverse capabilities of edge devices. This talk presents several novel solutions to
comprehensively address the above challenges through automated and improved model
compression. First, it introduces Automatic Attention Pruning (AAP), an adaptive,
attention-based pruning approach to automatically reduce model parameters while meeting
diverse user objectives in model size, speed, and accuracy. AAP achieves an impressive
92.72% parameter reduction in ResNet-101 on Tiny-ImageNet without causing any accuracy
loss. Second, it presents Self-Supervised Quantization-Aware Knowledge Distillation, a
framework for reducing model precision without supervision from labeled training data. For
example, it quantizes VGG-8 to 2 bits on CIFAR-10 without any accuracy loss. Finally, the talk
explores two more works, Contrastive Knowledge Distillation and Module Replacing, for further
improving the performance of small models. All the works presented in this talk are designed to
address real-world challenges, with applications extending to production environments such as
Amazon Alexa, and have been successfully deployed on diverse hardware platforms, including
cloud instances and edge devices, catalyzing AI for the edge.
________________________________________________________________________________
Microsoft Teams meeting
Join on your computer, mobile app or room device
Click here to join the meeting< https://teams.microsoft.com/l/meetup-join/19%3ameeting_MzIxZWVkM2MtMjg1MC00MjQ5LWIwY2MtZWM5ZmY2ZjdlNmY4%40thread.v2/0?context=%7b%22Tid%22%3a%22f3308007-477c-4a70-8889-34611817c55a%22%2c%22Oid%22%3a%22685c3f4f-29d5-4141-ada5-0fdeab8480e4%22%7d 
>
Meeting ID: 283 234 327 430
Passcode: Sf3wbd
Download Teams< https://www.microsoft.com/en-us/microsoft-teams/download-app 
> | Join on the web< https://www.microsoft.com/microsoft-teams/join-a-meeting 
>
Or call in (audio only)
+1 614-706-6572,,80448131#<tel:+16147066572,,80448131#>   United States, Columbus
Phone Conference ID: 804 481 31#
Find a local number< https://dialin.teams.microsoft.com/8f5f7319-0053-4423-a154-4f8b6e7fb7dd?id=80448131 
> | Reset PIN< https://dialin.teams.microsoft.com/usp/pstnconferencing 
>
[ https://www.ohio.edu/sites/default/files/2018-11/invite_logo_teams.jpg 
]
If you encounter issues with this meeting, please visit the Help link. If you are not able to resolve the problems, please contact the meeting organizer to let them know you are having difficulty.
Learn More< https://aka.ms/JoinTeamsMeeting 
> | Help< https://www.ohio.edu/oit/services/collaboration/teams/help 
> | Meeting options< https://teams.microsoft.com/meetingOptions/?organizerId=685c3f4f-29d5-4141-ada5-0fdeab8480e4&tenantId=f3308007-477c-4a70-8889-34611817c55a&threadId=19_meeting_MzIxZWVkM2MtMjg1MC00MjQ5LWIwY2MtZWM5ZmY2ZjdlNmY4@thread.v2&messageId=0&language=en-US 
>
________________________________________________________________________________

-------------- next part --------------
An HTML attachment was scrubbed...
URL: < http://listserv.ohio.edu/pipermail/eecs_msee/attachments/20240221/501048f5/attachment.html 
>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: text/calendar
Size: 11259 bytes
Desc: not available
URL: < http://listserv.ohio.edu/pipermail/eecs_msee/attachments/20240221/501048f5/attachment.ics 
>

• Previous message: [Eecs_msee] EECS Research Seminar Talk-Towards Deterministic and Reliable Real-Time Cyber-Physical Systems
• Next message: [Eecs_msee] EECS Research Seminar Talk-Geometric and Geospatial Algorithms under Data Conundrums in Geographic Information Systems
• Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]