Prof ShaoLin Hu
Guangdong University of Petrochemical Technology
Big Data, Safety Control, Statistical Learning, Fault Detection and Diagnosis
Abstract: Time series data clustering is an important branch and difficult subject in the field of data clustering. Based on the definition of morphological similarity of time series data sequence, a morphological similarity measurement method with affine invariance is established in this paper, and a morphological clustering algorithm based on morphological similarity measurement is established. By means of morphological similarity measurement and morphological clustering algorithm, two sets of time series data anomaly detection algorithms are established, which can be used for states monitoring of petrochemical industrial process and other practical fields, including morphological consistency detection at different periods of the same process and anomaly change detection algorithms of different objects with similar characteristics at multiple monitoring points. The results described above are very useful for multi-source time series data mining, clustering, modeling, statistical learning, as well as detection and diagnosis of process abnormal changes.
Assoc. Prof. Ag. Asri Ag. Ibrahim
Universiti Malaysia Sabah (UMS), Malaysia
Abstract: The increasing volume of data or information nowadays has led to a certain ineffectiveness and inefficiency of existing data processing applications. Storage is no longer a problem due to the availability of high storage capacity devices and lower market prices. As a result, the question has now mutated into ‘how can we effectively handle, present and understand the data?’ Much research has been carried out to investigate the best way to understand and deliver a massive amount of information to users. This has encouraged research activities in other fields such as big data, data analytics, data mining, data exploration, data visualization and so forth.
Graphical representation currently dominates the field of external representation, but sound is now seen as an alternative and its complement. Previous research has shown the success of using sound in several areas, especially for blind or visually impaired users; or in situations where the user’s eyes are occupied with other tasks such as looking at a patient in medical diagnosis; or something which is difficult to represent using graphics, such as multidimensional data. Thus, besides looking at the data, we could, in fact, also listen to the data. This field of data representation is known as Sonification. This talk will give an overview understanding of sonification research field
Assoc. Prof. Junjie Chen
Inner Mongolia Agricultural University，China
Abstract: Nowadays, deep learning methods have become popular in many fields and have been successfully applied to NLP tasks. Language has sequential characteristics, so recurrent neural network is the dominant framework in research. However, the document is organized in structure, and the relationship between words is determined by the syntax, so the sequence model is not enough to represent the feature of words in text.
Comparing with other data structures, such as sequences and trees, graphs are generally more powerful in representing complex correlations among entities. How to properly model text is a long-existing and important problem in NLP area, where several popular types of graphs are knowledge graphs, semantic graphs and dependency graphs. The graph neural network is an important deep leaning model on graph structure. In this talk, I will present recent work in graph neural network and text graph structure construction, and highlight opportunities and challenges in this area.