【学术报告会】
时间：2023.5.10，周三 13:00-16:00
地点：SPEIT 323会议室
【议程】
▲13:00-13:45, Ali Djafari, Bayesian Deep Learning or Deep learning with Bayesian principles
▲13:45-14:30, 初宁，Modal composition integrated fast-rotating sound source localization under Bayesian framework for axial fans
▲14:30-15:00 自由交流
▲15:00-16:00, 竺烨，Diffusion Models - From Data Generation, Editing to Astrophysics
【嘉宾介绍】
Ali Djafari
ABSTRACT
In this seminar, I discuss about Neural Nets(NN),Deep Learning(DL),Bayesian inference, Bayesian DL and Deep Learning with Bayesian principles.
The main content of this talk, follows:
1.Introduction to NNs. What's are good for?
2.Learning by training, testing, evaluating and using NNs
3.Main established tasks of NN and Deep Learning
4.Bayesian inference and computational methods
5.Deep Learning and/or Bayesian Learning?
6.Deep Learning with Bayesian principles
7.Available tools, challenges
PROFILE
Dr. Ali Djafari was once the research director of the French National Science Research Center, a tenured professor of the University of Paris Saclay, France, and the famous expert in the field of intelligent detection, playing a crucial role in the application of Bayesian inference methods in industrial intelligent diagnosis.
He is currently a strategic scientist and chief engineer of Zhejiang Shangfeng Special Blower Industry Co., Ltd.
初宁
ABSTRACT
In this talk, the RSP model is experimentally validated for the first time. In addition, a pioneering parameter selection scheme is proposed by investigating the applicable boundary of the MCB method, and its effectiveness is verified by simulation. Also, for the sound source localization results, three evaluation metrics are proposed, and the processing times of the related simulations are given to assist in verifying the above scheme. The proposed method of highresolution localization of rotating sound sources is fast in calculation, effective and high in resolution, and can realize the leaf tip noise localization of multi-blade high speed axial fans. And the proposed MCB parameter selection scheme can effectively guide the acoustic measurements and the practical application of the method.
PROFILE
Dr. Chu Ning has the B.S in NUDT, M.S and PhD. in Univ. Paris Saclay and postdoc in EPFL. He is the IEEE Senior member since 2020, currently the chief researcher of Zhejiang Shangfeng Special Blower Industry Co., Ltd. invited professor the "Belt and Road" Brunei class at Zhejiang University, and vice chairman of the Acoustics Society of Zhejiang Province.
He majors in Bayesian deep learning and multi-sensor information fusion, and successfully digitized the traditional ventilation equipment, which won the excellent case of national intelligent manufacturing in 2021 and the Industrial Internet platform of Zhejiang Province, and broadcasted by CCTV 2 in 2022 respectively.
竺烨
ABSTRACT
Diffusion Models (DMs) have become the mainstream approach in the generative field in computer vision and machine learning. However, despite its initial design as a generative model, diffusion models can be applied in a much wider context in addition to date generation. In this talk, I will introduce various applications of diffusion models. Firstly, we present the contrastive diffusion model in conditional cross-modality data generation tasks for music and images [Zhu et al., ICLR 2023]. Next, we introduce our boundary diffusion method that allows us to do image semantic manipulation and editing in an efficient learning-free manner [Zhu et al., preprint 2023]. Finally, we bring the diffusion models out of the computer vision field, but to the real astrophysical world to solve real-world scientific problems. Specifically, we successfully predict the molecular density in the universe via adopting DDPMs, which is a critical problem to infer the star formation in astronomy [Xu et al., ICLR Workshop Physics4ML 2023; Xu et al., The Astrophysical Journal 2023].
PROFILE
Ye Zhu is currently a senior Ph.D. student in Computer Science, affiliated with the Computer Vision and Multimedia Laboratory at Illinois Institute of Technology, USA, and Visual AI Laboratory at Princeton University, USA.
Her research mainly focuses on computer vision, with a special focus on multimodal learning and generation. She holds a B.S. and M.S. degree from Shanghai Jiao Tong University. She has published more than 10 papers in top-tier conferences and journals during her Ph.D., including ICLR, ECCV, TPAMI, etc. She is a recipient of the ACM-W scholarship, the Award for Excellence in Dissertation Research at Illinois Tech, the Microsoft Azure Cloud Computing Award, etc. She also serves as a reviewer for top computer vision and machine learning venues such as CVPR, ICCV, ECCV, NeurIPS, ICML, and AAAI.