Prof. Yun Peng
Title: Graph Learning for Combinatorial Optimization
Bio:
Yun Peng currently is a Professor of the Institute of Artificial Intelligence and BlockChain at Guangzhou University. He received his Ph.D degree from Hong Kong Baptist University in 2013. His research interests include graph learning, big graph data, and graph data security. He has published tens of papers in top-tier conferences and journals, including VLDB, ICDE, IJCAI, TKDE, and TSC, etc. He serves as the program committee member of IJCAI and the editor of International Journal of Intelligent Systems (IJIS). He is the reviewer of TKDE, ICDE, and SIGMOD, etc.
Abstract:
Graphs have been widely used to represent complex data in many applications, such as e-commerce, social networks, and bioinformatics. Efficient and effective analysis of graph data is important for graph-based applications. However, most graph analysis tasks are combinatorial optimization (CO) problems, which are NP-hard. Recent studies have focused a lot on the potential of using machine learning (ML) to solve graph-based CO problems. Most recent methods follow the two-stage framework. The first stage is graph representation learning, which embeds the graphs into low-dimension vectors. The second stage uses machine learning to solve the CO problems using the embeddings of the graphs learned in the first stage. The works for the first stage can be classified into two categories, graph embedding methods and end-to-end learning methods. For graph embedding methods, the learning of the embeddings of the graphs has its own objective, which may not rely on the CO problems to be solved. The CO problems are solved by independent down-stream tasks. For end-to-end learning methods, the learning of the embeddings of the graphs does not have its own objective and is an intermediate step of the learning procedure of solving the CO problems. The works for the second stage can also be classified into two categories, non- autoregressive methods, and autoregressive methods. Non-autoregressive methods predict a solution for a CO problem in one shot. A non-autoregressive method predicts a matrix that denotes the probability of each node/edge being a part of a solution of the CO problem. The solution can be computed from the matrix using search heuristics such as beam search. Autoregressive methods iteratively extend a partial solution step by step. At each step, an autoregressive method predicts a node/edge conditioned to current partial solution, which is used to its ex- tension. We will provide a thorough overview of recent studies of the graph learning-based CO methods and propose several remarks on future research directions.
Prof. Khan Muhammad
Title: Intelligent Fire Scene Analysis using Efficient Convolutional Neural Networks
Bio:
Khan Muhammad (S’16–M’18, SM’22) received his Ph.D. in Digital Contents from Sejong University, Republic of Korea in February 2019. He was an Assistant Professor in the Department of Software, Sejong University, from March 2019 to February 2022. He is currently the director of the Visual Analytics for Knowledge Laboratory (VIS2KNOW Lab) and an Assistant Professor (Tenure-Track) with the Department of Applied AI, School of Convergence, College of Computing and Informatics, Sungkyunkwan University, Seoul, Republic of Korea. His research interests include intelligent video surveillance, information security, video summarization, and smart cities. He has registered 10 patents and contributed 220+ papers in peer-reviewed journals and conference proceedings in his research areas. His contributions have received 10,940+ citations to date, with an H-index of 58. He is an Associate Editor/Editorial Board Member for more than 14 journals. He is among the most highly cited researchers in 2021, according to the Web of Science.
Abstract:
In today’s era, surveillance cameras are playing a major role in the detection of abnormal events such as fire, accidents, and violence. Among these events, fire is the most critical one, needing instance detection to minimize the overall damage to human lives and properties. For early detection of fire, several traditional and vision-based methods exist with a set of advantages and drawbacks. This talk will briefly discuss about the currently available approaches for early fire detection and will highlight some of their major drawbacks. Next, a few representative vision-based fire detection, segmentation, analysis methods will be discussed along with the available fire datasets. The talk will be concluded with major challenges in this area and a few directions for further research.