|
|
|
|
Rethinking Efficiency Bottleneck of Generative AI
|
**RSVP by Tuesday 4 August for catering purposes.
Accept this calendar invitation if you plan to attend the seminar in person and join us for a light lunch at approximately 12.30pm.
|
|
|
|
|
Generative AI has ushered in a new era of breakthroughs across science, industry, and society. Yet, the immense computational,
environmental, and financial costs required to train and deploy such models pose serious challenges to long-term sustainability. This talk presents a unified perspective on reducing these costs by addressing two fundamental bottlenecks: data redundancy and
model complexity. It introduces two cutting-edge strategies: neural data compression and generative model compression, and demonstrates how they can drastically reduce storage demands, training overhead, and inference latency, while maintaining or even enhancing
model performance. Through real-world case studies spanning efficient video, medical, and climate data compression, as well as acceleration techniques for large language models, video generation, and embodied AI systems, the talk demonstrates how principled
redundancy and complexity reduction can democratize access to generative AI. Ultimately, it argues that sustainable generative AI is achievable through efficiency-oriented design, enabling broader participation beyond tech giants and national megaprojects.
|
|
Dr Zhenghao Chen is a Lecturer (equivalent to Assistant Professor) at the University of Newcastle, Australia. He received
his B.Eng. H1. and Ph.D. from the University of Sydney in 2017 and 2022, respectively. He worked as as a Research Engineer at TikTok, a Research Fellow at the University of Sydney, and a Visiting Research Scientist at Microsoft Research and Disney Research.
Dr Chen’s research interests span generative AI, embodied AI, AI for science, and multimedia. He has published 33 papers in CORE A* conferences and JCR Q1 journals, including flagship conferences such as CVPR, ICCV, ICML, ICLR, ECCV, MM, and AAAI, as well
as leading journals including IEEE Transactions on Image Processing, IEEE Transactions on Pattern Analysis and Machine Intelligence, and IEEE Transactions on Medical Imaging. His research has received more than 2,800 citations. Dr Chen actively translates
his research into global industrial impact and cross-disciplinary applications. His industrial innovations have resulted in multiple patents and have been deployed in enterprise-scale generative AI systems worldwide. He also collaborates with scientists to
apply AI to major scientific challenges, resulting in interdisciplinary publications in Nature Portfolio journals. In recognition of his achievements, Dr Chen has received the Google Australia Prize, the Australian Government Research Training Program International
Scholarship, the ACM SIGMM Outstanding PhD Thesis Award, and the Microsoft Research Asia StarTrack Fellowship.
|
|
|
|
|
|
|
|
