Speakers

Prof. Tianrui Li

Southwest Jiaotong University, China

Research Area: Artificial intelligence, data mining and knowledge discovery, cloud computing and big data, granular computing and rough set

Speech Title: Application Cases of Intelligent Techniques for Sustainable Cities

Abstract: With the rapid development of urbanization, massive data has been accumulat-ed around urban management related fields. Sustainable urban management has entered the era of big data intelligence. How to effectively obtain useful knowledge from these big data by deep mining and intelligent learning tech-niques has become the key problem to be solved in the current sustainable ur-ban development. This report focuses on application cases of intelligent tech-niques for sustainable cities, e.g., urban taxi route recommendations, rental sug-gestions, ambulance deployment strategies, food delivery optimization, and subway stop scheduling. It concludes by discussing the challenges of sustaina-ble urban big data analysis.

Prof. Yan Yang

Southwest Jiaotong University, China

Research Area: Artificial intelligence, big data analysis and mining, integrated learning and multi-view learning, cloud computing and cloud services

Speech Title: Prediction of Spatio-temporal Data based on Deep Learning

Abstract: Spatio-temporal data contains rich information and intrinsic value, which is of great significance. Through the analysis and exploration of spatial-temporal data, it can address typical challenges such as traffic congestion, thus providing invaluable support for the construction of smart cities. In this talk, I will introduce deep learning, multi-task learning, multi-view learning and transfer learning to fully extract the nonlinear and dynamic spatial-temporal dependencies within data. Several novel deep learning models are devised and the effectiveness of the proposed models is shown through the demands of traffic prediction.

Prof. Xiaolin Qin, Deputy Chief Engineer

Chengdu Institute of Computer Applications, Chinese Academy of Sciences, China

Sichuan Institute of Artificial Intelligence, China

Research Area: Automated reasoning, Algebraic vision, Big data intelligent analysis, Natural language processing technology

Speech Title: Efficient Edge Detection Methods for Special Materials

Abstract: The reflection property of the target surface is an important factor affecting the photoelectric ranging effect. Low reflectivity or unstable reflection characteristics will affect the return quality of the beam, which will lead to the decrease of measurement accuracy or the inability to obtain effective measurement data. Pure visual ranging does not depend on the reflection property, and has better robustness to special materials with low reflectivity and high transmittance. Aiming at the problem of special material boundary blurring in pure visual ranging, this report proposes a Fourier boundary feature network method that integrates a wide-domain capturer. It uses a sufficiently wide horizontal shallow branch to guide the fine-grained segmentation boundary, and reduces the excessive capture of deep semantic information from the geometric structure to avoid mutual interference between location information and semantic information. A cross-shift attention mechanism is proposed, which avoids the incomplete boundary region caused by weak spot reflection noise by embedding fine-grained features. In order to balance the shallow and deep fitting performance of the network, a learnable Fourier convolution controller is constructed to reasonably fuse the shallow and deep learnable information in a robust way. The proposed method is verified on three different data sets. The experimental results show that compared with the existing advanced methods, the proposed method has better edge detection performance.

Assoc.Prof. Chuan Luo

Sichuan University

Research Area: Research on data mining and knowledge discovery, granular computing and rough set, incremental learning and parallel computing

Speech Title: Distributed Feature Selection for Scalable Dimensionality Reduction

Abstract: The emerging Big Dimensionality presents an immediate challenge pertaining to the scalability issue in the data analytics and computational intelligence communities. Feature selection, as a type of dimension reduction technique, has been proven to be effective and efficient in handling high dimensional data. However, the appearance of large data explosion leads to the existing serial computing feature selection algorithms are extremely time-consuming due to the limited computational and storage resources. As a practical pathway to pursue the challenge of explosive growth and aggregation of data, parallelization of algorithms by exploiting high performance computing resources in a distributed computing environment have increasingly gained strengths in facilitating large-scale data analysis. This talk will introduce our recent research works targeting scalable feature selection from multiple perspectives: Spark rough hypercuboid approach for scalable feature selection, Large-scale meta-heuristic feature selection based on BPSO assisted rough hypercuboid approach, and RHDOFS: a distributed online algorithm towards scalable streaming feature selection.