[auton-users] Fwd: FRC Seminar: "Large-scale multi-task machine learning via Gaussian processes", Arman Melkumyan. This Monday 20th June @ 11am. GHC2109.

[Julian R.]

Maybe of interest to some of you

---------- Forwarded message ----------
From: Stephen Nuske <nuske at cmu.edu>
Date: Thu, Jun 16, 2011 at 6:05 PM
Subject: FRC Seminar: "Large-scale multi-task machine learning via Gaussian
processes", Arman Melkumyan. This Monday 20th June @ 11am. GHC2109.
To: frc-seminar at cs.cmu.edu, robotics-seminar at ri.cmu.edu


**
Title
------
Large-scale multi-task machine learning via Gaussian processes*
*

Speaker
------
Arman Melkumyan
Research Fellow
Australian Centre for Field Robotics (ACFR)
University of Sydney


Time and Location
------
Monday 20th June @ 11am
GHC2109


Abstract
-----
Large-scale and multi-task learning remain difficult yet important problems
in machine learning. In large-scale Gaussian processes the difficulty arises
by the need to invert a potentially large covariance matrix during
inference. For the multi-task learning problems the main challenge is the
definition of valid kernels (covariance functions) able to capture the
relationships between different tasks. In this talk I’ll address the
complexity problem by constructing a stationary covariance function (Mercer
kernel) that naturally provides a sparse covariance matrix. The sparseness
of the matrix is defined by hyperparameters optimised during learning. This
covariance function enables exact GP inference and performs comparatively to
the squared-exponential one, at a lower computational cost.

The problem of multi-task learning will be addressed by presenting a novel
methodology for constructing valid multi-task covariance functions for
Gaussian processes allowing for a combination of kernels with different
forms. The method is based on Fourier analysis and is general for arbitrary
stationary covariance functions. Analytical solutions for cross covariance
terms between popular forms are provided including Matern, squared
exponential and sparse covariance functions.

I’ll also speak about our approach to classification of high dimensional
hyperspectral datasets and application of machine learning to solution of
differential and integral equations.

Experimental results will be discussed for both artificial and real datasets
demonstrating the benefits of the presented approaches.


Bio
----

Arman Melkumyan is a research fellow at the Australian Centre for Field
Robotics (ACFR) at the University of Sydney. Arman got his PhD in 1995 in
the field of Theoretical Mechanics and then worked as a postdoc at the
Centre for Advanced Materials Technologies at the University of Sydney. In
2008 Arman joined the ACFR as a researcher in the field of large-scale
multi-sensor machine learning modelling. Arman’s current work is mainly
concentrated on development of mathematical machine learning techniques for
probabilistic representations and information fusion for large scale
environments.




-- 

Julian.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mailman.srv.cs.cmu.edu/mailman/private/autonlab-users/attachments/20110617/9bfab1c1/attachment.html>