Cross-modal Knowledge Fusion and Collaborative Reasoning for Large Model Applications

Minghe Yu

Northeastern University, China, minghe.yu@neuc.edu.cn

10:00-10:30 am. *All timings are as per Japan Standard Time (JST) (UTC + 09:00)

Abstract:

With the rapid development of the new generation of artificial intelligence technology, large models have become one of the popular technologies for analyzing massive data. However, due to their weak ability for understanding and complex reasoning in domain knowledge, there exist many challenges to put large models into practical applications, mainly, hallucination, misunderstanding, and weak logic reasoning. This talk presents a cross-modal knowledge fusion and collaborative reasoning approach to improving the reliability of large models in solving complex problems. The key techniques include: 1) The construction and quality optimization of cross-modal domain knowledge graphs for extracting trustworthy knowledge; 2) Multi-source knowledge fusion for alignment and disambiguation of various types of knowledge; 3) Collaborative reasoning based on domain knowledge graphs and large models for acquiring new knowledge and making insightful predictions. It aims to build a data-driven, knowledge-based large model and promote the development of general artificial intelligence technology. This talk will introduce the relevant background, the solutions, the key techniques, and the prospects for future work.

Speaker Bio:

Minghe Yu is currently an Associate Professor in the Software College at Northeastern University, China. She received her B.S. degree in Computer Science and Technology from Northeastern University, China, in 2012, and her Ph.D. degree in Computer Science and Technology from Tsinghua University, China, in 2018. She visited the Electrical Engineering and Computer Science Department at the University of Michigan, Ann Arbor, USA, between 2016 and 2017. Her research interests include database systems, information retrieval, and intelligent education. Her works have been published in multiple top-tier venues, including TKDE, VLDBJ, ICDE, WWW, CIKM, DASFFA, and FCS,etc.

Protecting Users' Location Privacy in Traffic Flow and Crowdsensing Data

Libin Zheng

Sun Yat-sen University, libin.zheng@sysu.edu.cn

11:00-11:30 am. *All timings are as per Japan Standard Time (JST) (UTC + 09:00)

Abstract:

Urban residents' spatial activities are facing increasing privacy risks, especially the disclosure of their spatial locations. In this task, we target the users' location privacy in traffic and spatial crowdsensing data. Traffic flow data is the composition of personal driving trajectories, which can reveal sensitive information such as home and work locations, leading to privacy issues. To address this challenge, noises are added to the flow data with respect to the differential privacy paradigm, which ensures data safety but deteriorates its utility. We find that the inherent relations of the flow data inherited from the road network structure can be used to correct data without hurting the privacy property. Hence, we propose post-processing techniques, which exploit the data’s inherent relations for corrections. For the users' crowdsensing activities, we exploit a k-switch mechanism to improve the task assignment quality based on Geo-Indistinguishability. The mechanism engages users to undertake geographical sensing tasks while easing the concern of exposing their real-time locations.

Speaker Bio:

Libin Zheng is currently an Associate Professor at the School of Artificial Intelligence at Sun Yat-sen University. He received his Ph.D. degree from the Department of Computer Science and Engineering at the Hong Kong University of Science and Technology (HKUST), where he was recognized as an outstanding postgraduate. His research mainly includes spatiotemporal data mining, spatial mobility management, and crowdsourcing. He has won several accolades, including the Best Paper Runner-up Award at DASFAA 2020 and the Best Paper Nomination at CIKM 2021. Additionally, he won the Natural Science Second Prize for his contributions to IoT data management, awarded by the Guangdong Computer Foundation, China. Libin Zheng is also serving as a local chair for VLDB 2024.

2 hours 50 minutes (from 9:50-12:30) *All timings are as per Japan Standard Time (JST) (UTC + 09:00)

9:50-10:00 Opening

10:00-10:30 Invited Talk One

Title: Cross-modal Knowledge Fusion and Collaborative Reasoning for Large Model Applications

Speaker: Minghe Yu, Northeastern University, China.

10:30-10:45 Short Presentation One

Title: Co-Estimation of Data Types and Their Positional Distribution

Speaker: Shin-Ya Sato

(Nippon Institute of Technology, Japan)

10:45-11:00 Break

11:00-11:30 Invited Talk Two

Title: Protecting Users' Location Privacy in Traffic Flow and Crowdsensing Data

Speaker: Libin Zheng, Sun Yat-sen University, China.

11:30-11:45 Short Presentation Two

Title: Establishing a Decentralized Diamond Quality Management System: Advancing Towards Global Standardization

Authors: Duc Bui Tien (Nguyen Tat Thanh University), Bang Le (FPT University), Trung Phan Huyn Tan (FPT University)

11:45-12:00 Short Presentation Three

Title: Enhancing Load Forecasting with VAE-GAN-based Data Cleaning for Electric Vehicle Charging Loads

Authors: Wensi Zhang, Shuya Lei, Yuqing Jiang, Tiechui Yao, Yishen Wang(State Grid Smart Grid Research Institute Co., Ltd.) Zhiqing Sun (State Grid Hangzhou Power Supply Company)

Chengliang Chai

Beijing Institute of Technology, China, ccl@bit.edu.cn

Associated Professor

Yuyu Luo

The Hong Kong University of Science and Technology, China, yuyuluo@hkust-gz.edu.cn

Assistant Professor

Xiaoou Ding

Harbin Institute of Technology, China, dingxiaoou@hit.edu.cn

Assistant Professor