Title: Tensors and their Applications in Machine Learning and Computer Vision.
Speaker: Ms. Asma Almutairi, CCIS, KSU
Tensors are multi-dimensional arrays, where a first order tensor is a vector and a second order tensor is a matrix, tensor of order three and higher are called higher order tensors. The use of tensor representation requires a different mathematical framework from vector and matrix algebra. It was only introduced in the 1899, and mostly known to have its applications in physics and engineering. Yet, only lately with the emergence of multi factor analysis as well as the need for better data reduction techniques, it started to trigger uses in machine learning and computer vision. In machine learning, linear subspace learning (LSL) algorithms aim to solve for an optimal linear mapping by representing data as vectors and mapping them to a lower dimensional space. Such treatment of data can lead to a loss of correlation in the original data. However, with Multiline subspace learning (MSL) high multidimensional data are represented in their natural form as tensors. Thus, it can preserve the neutral inherited structure of the data when extracting features. In computer vision, tensors are used in building features to work with multi-factor variations. One example would be identifying faces under different illuminations, expression, and views, where tensor based techniques had been proven to give reliable results compared to linear techniques, like principle component analysis (PCA).
- Overview of tensor math.
- Multiline subspace learning(MSL).
- Tensor decomposition and applications.
- Other applications and algorithms.
- a demo of MATLAB Tensor Toolbox, by Bader and Kolda.
Speaker's bio: Asma Abdullah Almutairi, TA and thesis master student at Information System Department at the college of computer and information sciences, King Saud University.
Date/Time: Tuesday, Nov 29, 2016 at 12-1pm
Location: 6F55 in Building 6 (Broadcast to Room 2077 in CCIS male side)