学术信息

首页

电信学院学术讲座:澳大利亚墨尔本大学Saman K. Halgamuge教授学术系列讲座

  报告时间:2019年12月15日(周日)14:30;12月16日(周一)14:30

  报告地点:北辰校区电信学院楼102会议室

  报告嘉宾:Saman K. Halgamuge 教授


  

  Lecture 1:Deep Neural Networks in an increasingly Networked World of Transducers

  An inclusive framework for learning algorithms for Deep Neural Networks will be presented discussing the “known unknowns” and speculating about “unknown unknowns” in learning algorithm development. A paradigm shift can be observed in wide-ranging application domains such as energy management, image processing, neural engineering, bioinformatics, mechatronics etc, which are empowered by rapidly-advancing technologies ,e.g.,Internet of Things (IoT), that can generate large quantities of “imperfect” data for analysis of processes, compounds and organisms.  These applications are increasingly demanding transparency thus the need for moving away from completely backbox approaches for learning. These technologies have been spurred by the improvements in processor technology (e.g. GPU), that have allowed practitioners and researchers to overcome the computational limitations of Deep Neural Networks that depend on fully human curated or labelled data (i.e. Supervised Learning). The following fundamental question then naturally arises: What happens when curated information or labels capture only a subset of critical classes, or the curation process itself is not fault- or error-free? Undoubtedly, the algorithm’s perceived reality will distort any subsequent analysis of these data, which may have detrimental downstream effects when new discoveries and critical decisions are made on a basis of these analyses. In such scenarios, learning algorithms that can find models –underlying structures or distinct patterns within data – without relying on labels (i.e. using Unsupervised Learning), have made great progress toward answering these sorts of questions; however, these algorithms only address part of the problem. Unsupervised Learning algorithms used in Unsupervised Deep Neural Networks do not consider any available and potentially reliable information or domain knowledge, which could prove useful in developing a robust model of the data. It can be advantageous to consider such information as well as any other available domain knowledge, not as ground truth but as a starting point to build a more complete picture of the problem under investigation. Application of Deep Neural Networks in Internet of Things (IoT) enabled world opens up the need for extensive new research.  Some of the landmark contributions by my research groups at University of Melbourne and Australian National University are also highlighted. The recent work on Generative Adversarial Networks (GANs) and Self Organizing Nebulous Growths (SONG) are such research contributions.

 

  Lecture 2:Optimization Algorithms for Energy

  Optimization algorithms in AI are increasingly been applied in many areas of Engineering, including energy, communication networks, transport planning and construction. Since we do not know much about the fitness landscape of a real world optimization problem we try to solve, it can be challenging to pick the right method for agiven complex real problem.  Comparisons of algorithms using numerous benchmarks may reveal the better algorithms suitable for the benchmarks. However, we do not know whether the set of benchmarks includes a problem similar to the one we try to solve. Selecting the correct optimization algorithm to a given problem can be achieved through the characterization of the fitness landscape. The solution to this problem becomes even more challenging when the fitness landscape changes dynamically. We report on some exciting new insights on the algorithm selection problem and its applications based on three PhD projects in my lab and some work of others. These PhD projects include Operational Optimization of Smart Grids with storage, road network planning and the integration of multiple renewable energy technologies including shallow geothermal energy generation in to the grids.

 

  Saman K. Halgamuge教授简介:

  Prof Saman K. Halgamuge, FIEEE is a Fellow of Institute of Electrical and Electronics Engineering (IEEE), USA, and aDistinguished Lecturer/Speaker appointed by IEEE in the area of Computational Intelligence.Heiscurrently a Professor in the Department of Mechanical Engineering, School of Electrical, Mechanical and Infrastructure Engineering at the University of Melbourne, an honorary Professor of School of Electrical, Energy and Materials Engineering at Australian National University (ANU).  He was previously the Director of the Research School of Engineering at the Australian National University (2016-18) and held Professor, Associate Dean International, Associate Professor and Reader and Senior Lecturer positions at University of Melbourne (1997-2016).  He graduated with Dipl.-Ing and PhD degrees in Data Engineering from Technical University of Darmstadt.

  His fundamental research contributions are in Unsupervised and Near Unsupervised type learning as well as in transparent Deep Learning and Bioinspired Optimization. His h-index is 43 (9200 citations) in Google Scholar and he graduated 45 PhD students as the primary supervisor. He has also been a keynote speaker for 40 research conferences.

 

  Saman K. Halgamuge教授简介(中文):

  Saman K. Halgamuge教授,IEEE Fellow,IEEE在计算智能领域任命的杰出讲师演讲者。现任墨尔本大学电、机和基础设施工程学院机械工程系教授,澳洲国立大学(ANU)电子电气、能源和材料工程学院名誉教授。曾任澳洲国立大学工程研究学院院长(2016-2018年),并在墨尔本大学担任教授、国际学院副院长、副教授和高级讲师等职位(1997-2016年)。他毕业于德国达姆施塔特工业大学,获数据工程专业硕士学位和博士学位。

  他的基础研究贡献在于无监督和近无监督类型学习以及透明深度学习和生物启发式优化。在Google Scholar中他的H指数为43(9200次引用),并指导45名博士研究生毕业。另外,他还是40个研究会议的主讲人之一。