Grant: Engage Grants Program
Ryerson University
$25,000.00
Implementing restricted Boltzmann machines using fusion technology for classifying and searching images and videos in large scale digital libraries
FY 2010-11
Summary
Search and management of large-scale image, video and multimedia data is one of the critical steps in bringing
the huge volume of digital media on the Internet into our everyday life. Currently, the progress in this field is
severely limited by the proper annotation of data. In particular, among 5 billion images on the Internet only
5-10% of the images are human-labelled, and to effectively search through the remaining images, one needs to
find an automated method of properly annotating them. At Ryerson University, we are actively researching into
the use of Restricted Boltzmann Machines (RBMs) for automated annotation of multimedia data on the
Internet. As part of this ongoing research, we are investigating efficient methods of implementing RBM-based
learning algorithms on off-the-shelf computers. In most of these computers, Graphic Processing Units (GPUs)
are particularly well suited for implementing RBMs since a significant amount of the RBM learning process
involves matrix operations. In the past, RBM implemented on discrete GPUs have suffered from extensive
performance-limiting overhead due to memory copying and thread synchronization operations. The close
coupling of the Central Processing Unit and the GPU in the new AMD Fusion architecture potentially can
significantly reduce this overhead and result in substantial increases in performance. In this project, we propose
to investigate the effectiveness of the AMD Fusion architecture in accelerating RBMs by designing a suite of
Application Programming Interface (API) functions for Fusion. These functions will be based on the OPENCL
standard and will exploit the unique features of the AMD Fusion architecture in order to maximize their
performance. The performance between the Fusion and the traditional discrete GPUs will then be compared in
order to quantify the performance advantage of the Fusion architecture. The API developed as part of this
proposal will be used in the development of RBM-based video event detection and image annotation
applications with the ultimate goal of fully exploiting the AMD Fusion architecture in the search and
management of multimedia data on Internet.
Program
Engage Grants Program
Recipient
Location
Ontario, Canada
Recipient Type
Academia
Type
Grant
Source
NSERC's Awards Data
As published
Application id
411523-2010
Name nom
Ye, Andy
Department département
Electrical and Computer Engineering
Organization id
43
Institution établissement
Ryerson University
Province en
Ontario
Province fr
Ontario
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2010
Competition year année de concours
2010
Award amount
25000
Program id
EGP
Program name en
Engage Grants Program
Program name fr
Programme de subventions d'engagement partenariat
Group en
Research partnerships programs
Group fr
Programmes de partenariat
Committee code
1553
Committee name en
Ontario Internal Decision Committee
Committee name fr
Comité de décision interne pour l'Ontario
Area of application code
513
Area of application group en
Manufacturing processes and products
Area of application group fr
Procédés et produits industriels
Area of application en
Electrical and electronic machinery and equipment (including computer hardware)
Area of application fr
Machinerie et équipement électriques et électroniques (y compris matériel informatique)
Research subject code
2701
Research subject group en
Information technology
Research subject group fr
Technologie de l'information
Research subject en
Computer hardware
Research subject fr
Matériel informatique
Partie
1
Nb partie
1
Application title
Implementing restricted Boltzmann machines using fusion technology for classifying and searching images and videos in large scale digital libraries
Keyword
NSERC
Application summary
Search and management of large-scale image, video and multimedia data is one of the critical steps in bringing
the huge volume of digital media on the Internet into our everyday life. Currently, the progress in this field is
severely limited by the proper annotation of data. In particular, among 5 billion images on the Internet only
5-10% of the images are human-labelled, and to effectively search through the remaining images, one needs to
find an automated method of properly annotating them. At Ryerson University, we are actively researching into
the use of Restricted Boltzmann Machines (RBMs) for automated annotation of multimedia data on the
Internet. As part of this ongoing research, we are investigating efficient methods of implementing RBM-based
learning algorithms on off-the-shelf computers. In most of these computers, Graphic Processing Units (GPUs)
are particularly well suited for implementing RBMs since a significant amount of the RBM learning process
involves matrix operations. In the past, RBM implemented on discrete GPUs have suffered from extensive
performance-limiting overhead due to memory copying and thread synchronization operations. The close
coupling of the Central Processing Unit and the GPU in the new AMD Fusion architecture potentially can
significantly reduce this overhead and result in substantial increases in performance. In this project, we propose
to investigate the effectiveness of the AMD Fusion architecture in accelerating RBMs by designing a suite of
Application Programming Interface (API) functions for Fusion. These functions will be based on the OPENCL
standard and will exploit the unique features of the AMD Fusion architecture in order to maximize their
performance. The performance between the Fusion and the traditional discrete GPUs will then be compared in
order to quantify the performance advantage of the Fusion architecture. The API developed as part of this
proposal will be used in the development of RBM-based video event detection and image annotation
applications with the ultimate goal of fully exploiting the AMD Fusion architecture in the search and
management of multimedia data on Internet.