Manifolds in Machine Learning and Computer Vision

Number of lecture hours

15 hours cross 5 days


Brief introduction to the course

This postgraduate course will feature a short theoretical overview about the role of manifold methods and methodologies in and for machine learning and computer vision.  It will introduce the background of manifolds and Riemannian geometry. Moreover, both linear and non-linear manifold learning techniques will be described. Besides, this course will feature extensive study materials about the geometry of three important matrices, namely the orthogonal matrices, positive cone, and low-rank matrices. Furthermore, it will embrace constrained optimization and Riemannian geometry. Finally, the connections between the Riemannian geometry and the groundbreaking deep learning methods will be highlighted.

 

Biography of the lecturer

Dr. Mehrtash Harandi,

1. Senior Researcher in the Data61-CSIRO Machine Learning Research Group in Canberra, Australia

2. Adjunct Lecturer position in the Australian National University at the College of Engineering and Computer

Email: Mehrtash.Harandi@data61.csiro.au

Dr. Mehrtash Harandi received his PhD from the University of Tehran in Tehran, Iran in 2009. In his PhD dissertation, Mehrtash studied various solutions for identifying humans from visual data using the concept of reinforcemnet learning. Since late 2009, Mehrtash has been developing and evaluating Statistical Machine Learning methods for visual analysis. He works as a Senior Researcher in the Data61-CSIRO Machine Learning Research Group in Canberra, Australia and holds an Adjunct Lecturer position in the Australian National University at the College of Engineering and Computer. Before joining Data61, Mehrtash worked for National ICT Australia (NICTA) as a Researcher in the computer vision research group. Mehrtash has over seventy publications with 30 co-authors from 9 countries, including EPFL, University of Oxford, Amazon and the Tsinghua University.  His research interests are developing theoretical Machine Learning, Mathematical Modeling, Computer Vision and Signal Processing. Dr. Harandi is a recognized scientist in both the machine learning and computer vision communities for his contributions in non-linear analysis and especially Riemmanian geometry in addressing learning problems. He is acting as a reviewer for various premier journals in his field including IEEE Trans. on Pattern Analysis and Machine Intelligence (TPAMI), IEEE Trans. on Neural Networks and Learning Systems (TNNLS) and Int. Journal of Computer Vision (IJCV). He is also a program committee of premier conferences in computer vision and machine learning including IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Int. Conference on Computer Vision (ICCV) and Neural Information Processing Systems (NIPS). He was the area chair of IEEE Winter Conference on Applications of Computer Vision (WACV), 2017, and was the organizer of various workshops and tutorials at premier conferences such as International Workshop on Differential Geometry in Computer Vision and Machine Learning at CVPR.

Last updated: 23.5.2018