|
Prof. Nick Freris
(Keynote Speaker)
Senior Member of ACM and IEEE
University of Science and Technology of China, China
Speech Title: Adaptive Compression of Deep Neural Networks
Abstract:Model compression is crucial for accelerating deep neural networks while maintaining high prediction accuracy. In this talk, I will present a lightweight compression method termed Adaptive SensiTivity-basEd pRuning (ASTER) which dynamically adjusts the filter pruning threshold concurrently with the training process. This is accomplished by computing the sensitivity of the loss to the threshold on the fly (without re-training), as carried with minimal overhead on the Batch Normalization (BN) layers. ASTER then proceeds to adapt the threshold so as to maintain a fine balance between pruning ratio and model accuracy. Extensive experiments on numerous neural networks and benchmark datasets illustrate a state-of-the art trade-off between FLOPs reduction and accuracy, along with formidable computational savings.
Biography: Nick Freris is Professor in the School of Computer Science at USTC, and Vice Dean of the International College. He received the Diploma in Electrical and Computer Engineering from the National Technical University of Athens (NTUA), Greece, in 2005, and the M.S. degree in Electrical and Computer Engineering, the M.S. degree in Mathematics, and the Ph.D. degree in Electrical and Computer Engineering all from the University of Illinois at Urbana-Champaign
(UIUC) in 2007, 2008, and 2010, respectively.
His research lies in AIoT/CPS/IoT: machine learning, distributed optimization, data mining, wireless networks, control, and signal processing, with applications in power systems, sensor networks, transportation, cyber security, and robotics. Dr. Freris has published several papers in high-profile conferences and journals held by IEEE, ACM, and SIAM and holds three patents. His research has been sponsored by the Ministry of Science and Technology of China, Anhui Dept. of Science and Technology, Tencent, and NSF, and was recognized with the National High-level Talent award, the USTC Alumni Foundation Innovation Scholar award, and the IBM High Value Patent award. Previously, he was with the faculty of NYU and, before that, he held senior researcher and postdoctoral researcher positions at EPFL and IBM Research, respectively.
Dr. Freris is a Senior Member of ACM and IEEE, and a member of CCF and SIAM.<Personal Webpage>
llllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllll |
|
Prof. Philippe Fournier-Viger
(Keynote Speaker)
Shenzhen University, China
Speech Title: Advances and challenges for the automatic discovery of interesting patterns in data
Abstract:Intelligent systems and tools can play an important role in various domains such as for factory automation, e-business, and manufacturing. To build intelligent systems and tools, high-quality data is generally required. Moreover, these systems need to process complex data and can yield large amounts of data such usage logs, images, videos, and data collected from industrial sensors. Managing the data to gain insights and improve these systems is thus a key challenge. It is also desirable to be able to extract information or models from data that are easily understandable by humans. Based on these objectives, this talk will discuss the use of data mining algorithms for discovering interesting and useful patterns in data generated from intelligent systems and other applications.
The talk will first briefly review early study on designing algorithms for identifying frequent patterns. Then, an overview of recent challenges and advances will be presented to identify other types of interesting patterns in more complex data. Topics that will be discussed include high utility patterns, locally interesting patterns, and periodic patterns. Lastly, the SPMF open-source software will be mentioned and opportunities related to the combination of pattern mining algorithms with traditional artificial intelligence techniques for intelligent systems will be discussed.
Biography:Philippe Fournier-Viger (Ph.D) is a Canadian researcher, distinguished professor at Shenzhen University (China). Five years after completing his Ph.D., he came to China in 2015 and became full professor after receiving a talent title from the National Science Foundation of China. He has published more than 375 research papers related to data mining algorithms for complex data (sequences, graphs), intelligent systems and applications, which have received more than 11,000 citations. He is the founder of the popular SPMF data mining library, offering more than 250 algorithms to find patterns in data, cited in more than 1,000 research papers. He is former associate edito-in-chief of the Applied Intelligence journal and has been keynote speaker for over 15 international conferences and co-edited four books for Springer. He is a co-founder of the UDML, PMDB and MLiSE series of workshops held at the ICDM, PKDD, DASFAA and KDD conferences. <Personal Webpage>
llllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllllll |