Program
Date
Venue
- Shaoke
Guan Building, Yuquan Campus, Zhejiang University, 38 Zheda Road,
Hangzhou
- 杭州市浙大路38号浙江大学玉泉校区邵科馆
Conference Schedule
Notes:
- Each
keynote speech will be given one hour including answering questions.
- Each
invited speech will be given forty minutes including answering
questions.
- For
each keynote speaker and invited speaker, it is better to leave 5-10
minutes for answering potential questions.
- We will
offer free lunch at conference venue on August 4-5.
|
Thursday, August 4
|
|
|
8:40-9:00
|
Opening Ceremony
|
|
9:00-10:00
|
Keynote speech : Open universes
and nuclear weapons
Stuart Russell, University of California, Berkeley
|
|
10:00-10:20
|
Coffee Break
|
|
10:20-11:00
|
Invited Speech: Sparse
modeling: some unifying theory and "subject-imaging"
Bin Yu, University of California, Berkeley
|
|
11:00-11:40
|
Invited speech: Learning and
Mining using Visual Data on the Web
Jiebo Luo, Kodak Research Laboratories
|
|
11:40-12:20
|
Invited speech: GoDec:
Randomized Low-rank & Sparse Matrix Decomposition in Noisy Case
Daocheng Tao,University of Technology Sydney
|
|
12:20-14:00
|
Lunch
|
|
14:00-14:40
|
Invited speech: Active
Learning: When does it work?
Liwei Wang, Peking University
|
|
14:40-15:20
|
Invited speech: Sparse Anova
Models
Jinzhu Jia, Peking University
|
|
15:20-15:40
|
Coffee Break
|
|
15:40-16:20
|
Invited speech: A Hierarchical
Framework for Bayesian Sparse Learning
Zhihua Zhang, Zhejiang University
|
|
16:20-17:00
|
Invited speech: Regularized
Latent Semantic Indexing
Hang Li, Microsoft Research Asia
|
|
17:00-17:40
|
Invited speech: Learning
Semantics from Web Images for Cross-media Retrieval
Jianping Fan, University of North Carolina at Charlotte
|
|
Friday, August 5
|
|
|
9:00-10:00
|
Keynote speech : Algorithmic
and statistical perspectives on large-scale data analysis
Michael Mahoney, Stanford University
|
|
10:00-10:20
|
Coffee Break
|
|
10:20-11:00
|
Invited Speech: Some Results on
Dictionary Learning for Sparse Signal Representation
John Wright, Microsoft Research Asia
|
|
11:00-11:40
|
Invited speech: Large-scale
Visual Recognition via Unsupervised Feature Learning
Kai Yu, NEC Laboratories America. Inc
|
|
11:40-12:20
|
Invited speech: On the
usefulness of multi-views in exploiting unlabeled data
Zhi-Hua Zhou, Nanjing University
|
|
12:20-14:00
|
Lunch
|
|
14:00-14:40
|
Invited speech: Theory and
Applications of Greedy Algorithms in Machine Learning
Tong Zhang, Rutgers University
|
|
14:40-15:20
|
Invited speech: Learning with
Exclusive Regularizations: Formulations and Solution
Shuicheng Yan, National University of Singapore
|
|
15:20-15:40
|
Coffee Break
|
|
15:40-16:20
|
Invited speech: Parallel Vector
Field Embedding
Xiaofei He, Zhejiang University
|
|
16:20-17:00
|
Invited speech: Imaging fusion
for diffuse optical tomography and X-ray computed tomography
Yuanzheng Si and Ming Jiang, Peking University
|
|
17:00-17:40
|
Invited speech: Yi Ma,
University of Illinois at Urbana-Champaign and Microsoft Research Asia
(pending)
|
Registration
There is no registration fee for
each participant. However, since the space is limited, only participants
who sent out the registration form before July 3 to
us can attend the conference.
Stuart
Russell at University of California, Berkeley
Title: Open universes and nuclear weapons
Abstract: I will discuss a formal unification of probability
theory and full (open-universe) first-order logic that allows for
uncertain reasoning about unknown objects and events within a
general-purpose formal language. Applications range from citation
information extraction to monitoring compliance with the Comprehensive
Nuclear-Test-Ban Treaty. The second half of the talk will describe the
latter application in detail.
[ back to Schedule ]
Bin Yu at
University of California, Berkeley
Title : Sparse modeling: some unifying theory and
"subject-imaging"
Abstract: Information technology has enabled collection
of massive amounts of data in science, engineering, social science,
finance and beyond. Extracting useful information from massive and
high-dimensional data is the focus of today's statistical research and
practice. After broad success of statistical machine learning on
prediction through regularization, interpretability is gaining attention
and sparsity is being used as its proxy.
With the virtues of both regularization and sparsity,
sparse modeling methods (e.g. Lasso) has attracted much attention for
theoretial research and for data modeling.
This talk discusses both theory and pratcice of
sparse modeling. First we present some recent theoretical results
on bounding L2-estimation error (when p>>n) for a class of
M-estimation methods with decomposable penalities. As special cases, our
results cover Lasso, L1-penalized GLMs, grouped Lasso, and low-rank
sparse matrix estimation. Second we present on-going research on "subject-imaging"
supported by an NSF-CDI grant.
This project employs sparse logistic regression to
derive a list of words ("subject-image") that associate with a
particular subject matter (e.g. "China") in, for example, New
York Times articles. The validity of such a list is supported by human
subject experiment results when compared with some other methods.
[ back to Schedule ]
Jiebo Luo
at Kodak Research Laboratories
Title: Learning and Mining using Visual Data on the Web
Abstract: Increasingly rich and large-scale image
related data are being posted to social network and media sharing
websites. Researchers from multidisciplinary areas, including machine
learning, computer vision, data mining, and human machine interaction,
are developing methods for employing such multi-modality data for various
applications. We present several recent advances in this arena of
opportunities and challenges. First, we address the multi-modality
feature issue by developing new machinery called Heterogeneous Feature
Machines (HFM), which builds a kernel logistic regression model based on
similarities that combine different features and distance metrics with
group LASSO constraints. Its power is demonstrated across a wide variety
of visual recognition tasks including scene, event, and action
recognition.
Second, we examine the recently popular data driven
approach that has seemingly diminished the need for machine learning in
favor of simply relying on large scale data. We believe it is important
to address crucial machine learning issues, in particular cross-domain
learning, in order to intelligently leverage large scale web data to
solve problems such as searching personal images by keywords and
recognizing events in personal videos. Finally, we explore the global trends
and sentiments that can be drawn by mining the sharing patterns of
uploaded and downloaded social multimedia. We consider that each time an
image or video is uploaded or shared, it constitutes an implicit yet
trustworthy vote for (or against) the subject of the image. By
aggregating such votes across millions of Internet users, we reveal the
wisdom that is embedded in social multimedia sites for prediction and
forecast in politics, economics, and marketing.
Bio: Jiebo Luo is a Senior Principal Scientist with
the Kodak Research Laboratories in Rochester, NY. His research interests
include image processing, computer vision, machine learning, social media
data mining, medical imaging, and computational photography. He has
authored over 160 technical papers and holds over 60 US patents. Dr. Luo
has been actively involved in numerous technical conferences, including
serving as the general chair of ACM CIVR 2008, program co-chair of IEEE
CVPR 2012 and ACM Multimedia 2010, area chair of IEEE ICASSP 2009-2011,
ICIP 2008-2011, CVPR 2008 and ICCV 2011, and an organizer of ICME
2006/2008/2010 and ICIP 2002. Currently, he serves on several IEEE
Technical Committees (IMDSP, MMSP, and MLSP) and conference steering
committees (ACM ICMR and IEEE ICME). He is the Editor-in-Chief of the
Journal of Multimedia, and has served on the editorial boards of the IEEE
Transactions on Pattern Analysis and Machine Intelligence (TPAMI), the
IEEE Transactions on Multimedia (TMM), the IEEE Transactions on Circuits
and Systems for Video Technology (TCSVT), Pattern Recognition (PR),
Machine Vision and Applications (MVA), and Journal of Electronic Imaging
(JEI). He is a Fellow of the SPIE, IEEE, and IAPR.
[ back to Schedule ]
Daocheng
Tao at University of Technology Sydney, Australia
Title: GoDec: Randomized Low-rank & Sparse Matrix
Decomposition in Noisy Case
Abstract: Low-rank and sparse structures have been
profoundly studied in matrix completion and compressed sensing. In this
paper, we develop “Go Decomposition” (GoDec) to efficiently and robustly
estimate the low-rank part L and the sparse part S of a matrix X = L + S
+ G with noise G. GoDec alternatively assigns the low-rank approximation
of X − S to L and the sparse approximation of X − L to S. The
algorithm can be significantly accelerated by bilateral random
projections (BRP). We also propose GoDec for matrix completion as an
important variant. We prove that the objective value ||X − L −
S|| converges to a local minimum, while L and S linearly converge to local
optimums. Theoretically, we analyze the influence of L, S and G to the
asymptotic/convergence speeds in order to discover the robustness of
GoDec. Empirical studies suggest the efficiency, robustness and
effectiveness of GoDec comparing with representative matrix decomposition
and completion tools, e.g., Robust PCA and OptSpace.
[ back to Schedule ]
Liwei Wang
at Peking University
Title: Active Learning: When does it work?
Abstract: Traditional supervised learning assumes the
availability of a training set of labeled examples. In many tasks
labeling the examples is costly while unlabeled data are easy to collect.
Active learning is a model which aims to reduce the nunber of label
requests. The idea is to adaptively query the labels of the most
informative examples based on previous observations. It is important to
understand whehter active learning really requires fewer labels than
supervised learning. Although there are examples in which active learning
yields lower label complexity than supervised learning, there are also
simple cases for which active learning does not save labels at all,
leading to the suspection that active learning is not useful in practice.
In this talk, we give intuitively reasonable conditions under which
active learning provably requires strictly fewer lables than active
learning. We show that the smoothness of the classification boundary is
cruicial to active learning: Under some noise condition, if the boundary
is smooth to a finite order, active learning achieves polynomial lower
label complexity than supervised learning; if it is infinitely smooth,
the label saving is exponential.
[ back to Schedule ]
Jinzhu Jia
at Peking University
Title: Sparse Anova Models
Abstract: The LASSO has been a popular method to
simultaneously select predictors and build models. However in many
applications, the relationship between predictors and the response are
nonlinear and there are interaction effects between predictors.
We present sparse Anova models for modeling the
relationship between predictors and the response. A sparse Anova model
decomposes the predictive function into main effects and interactions.
Our goal is to select a few predictors and select a few main effects
and/or interactions. The sparse Anova models extend the Lasso to
non-linear and non-parametric models.
[ back to Schedule ]
Zhihua
Zhang at Zhejiang University
Title: A Hierarchical Framework for Bayesian Sparse
Learning
Abstract: In this talk, we present Bayesian sparse
learning problem. We propose a unified framework for the construction
of sparsity-inducing priors. In particular, we define such priors
as a mixture of exponential power distributions with a generalized inverse
Gaussian density (EP-GIG). The special cases include extant normal-gamma
distributions and laplace-inverse gamma distributions. The densities of
EP-GIG can be explicitly expressed. Moreover, the corresponding poster
distribution also follows a generalized inverse Gaussian distribution.
These properties lead us to EM algorithms for Bayesian sparse learning.
We will show that the EM algorithm bears a strong resemblance with the
iteratively reweighted $\ell_2$ or $\ell_1$ method.
[ back to Schedule ]
Hang Li at
Microsoft Research Asia
Title: Regularized Latent Semantic Indexing
Abstract: In this talk, I will introduce a topic
modeling technique, called Regularized Latent Semantic Indexing, which we
have developed at MSRA. I will start my talk by explaining the motivation
of the work, namely, solving the term mismatch problem in web search. I
will also give a high level introduction of the machine learning
techniques which we have developed for addressing term mismatch. Next, I
will describe in details about the Regularized Latent Semantic Indexing
(RLSI) techniques. RLSI is a new method for topic modeling, formalized as
the minimization of a quadratic loss function regularized by L1 norm
and/or L2 norm One advantage of RLSI is that it has significantly better
scalability than existing methods. I will also make comparison between
RLSI and other topic modeling techniques such as LDA, PLSI, LSI, and NMF.
Hang Li is senior researcher and research manager at
Microsoft Research Asia. He is also adjunct professors at Peking
University, Nanjing University, Xi’an Jiaotong University, and Nankai
University. His research areas include information retrieval, natural
language processing, statistical machine learning, and data mining. He
graduated from Kyoto University in 1988 and earned his PhD from the
University of Tokyo in 1998. He worked at the NEC lab in Japan
during 1991 and 2001. He joined Microsoft Research Asia in 2001 and
has been working there until present. http://research.microsoft.com/en-us/people/hangli/
[ back to Schedule ]
Jianping
Fan at University of North Carolina at Charlotte
Title: Learning Semantics from Web Images for
Cross-Media Retrieval
Abstract: In this presentation, I will introduce our
recent work on learning the semantics of large-scale web images by
achieving more accurate alignments between the web images and the
auxiliary text terms. First, large-scale cross-media web pages (i.e., web
images and their auxiliary text documents) are crawled and automatic web
page segmentation is performed to extract the informative images and
align the informative images with their most relevant auxiliary text
blocks. Second, distributed image clustering is performed to partition
large-scale web images into a set of image clusters according to their
visual similarity contexts. Such distributed image clustering process can
significantly reduce the uncertainty on the relatedness between the
semantics of the web images and their auxiliary text terms. The semantics
for the web images in the same cluster can be described effectively by a
same set of auxiliary text terms that co-occur frequently in the relevant
text blocks. Finally, random walk is performed over a term correlation
network to refine the relevance scores for achieving more precise
alignments between the semantics of the web images and their auxiliary
text terms.
Our work have several potential applications: (a)
achieving more accurate web image indexing and enabling cross-media
retrieval; (b) creating large-scale labeled training images by harvesting
from Internet.
Bio for Jianping Fan: Jianping Fan received his MS
degree in theory physics from Northwestern University, Xian, China
in 1994 and his PhD degree in optical storage and computer science from
Shanghai Institute of Optics and Fine Mechanics, Chinese Academy of
Sciences, Shanghai, China, in 1997. He was a Postdoc Researcher at
Fudan University, Shanghai, China, during 1998. From 1998 to 1999, he was
a Researcher with Japan Society of Promotion of Science (JSPS),
Department of Information System Engineering, Osaka University,
Osaka, Japan. From 1999 to 2001, he was a Postdoc Researcher in the
Department of Computer Science, Purdue University, West Lafayette, IN. At
2001, he joined the Department of Computer Science, University of North
Carolina at Charlotte as an Assistant Pofessor and then become Associate
Professor. His research interests include image/video analysis,
semantic image/video classification, personalized image/video
recommendation, surveillance videos, and statistical machine learning.
[ back to Schedule ]
Michael
Mahoney at Stanford University
Title: Algorithmic and
statistical perspectives on large-scale data analysis
Abstract: Computer scientists and statisticians have
historically adopted quite different views on data and thus on data
analysis. In recent years, however, ideas from statistics and scientific
computing have begun to interact in increasingly sophisticated and
fruitful ways with ideas from computer science and the theory of
algorithms to aid in the development of improved worst-case algorithms
that are also useful in practice for solving large-scale scientific and
Internet data analysis problems. After reviewing these two complementary
perspectives on data, I will describe two recent examples of improved
algorithms that used ideas from both areas in novel ways. The first example
has to do with improved methods for structure identification from
large-scale DNA SNP data, a problem which can be viewed as trying to find
good columns or features from a large data matrix. The second example has
to do with selecting good clusters or communities from a data graph, or
demonstrating that there are none, a problem that has wide application in
the analysis of social and information networks. Understanding how
statistical ideas are useful for obtaining improved algorithms in these
two applications may serve as a model for exploiting complementary
algorithmic and statistical perspectives in order to solve applied
large-scale scientific and Internet data analysis problems more
generally.
[ back to Schedule ]
John Wright
at Microsoft Research Asia
Title: Some Results on Dictionary Learning for Sparse
Signal Representation
Abstract: The idea that many important classes of
signals can be well-represented by linear combinations of a small set of
atoms selected from a given dictionary has had dramatic impact on the
theory and practice of signal processing. For practical problems in which
an appropriate sparsifying dictionary is not known ahead of time, a very
popular and successful heuristic is to search for a dictionary that
minimizes an appropriate sparsity surrogate over a given set of sample
data. While this idea is appealing, the behavior of these algorithms is
largely a mystery; although there is a body of empirical evidence
suggesting they do learn very effective representations, there is little
theory to guarantee when they will behave correctly, or when the learned
dictionary can be expected to generalize. In this talk, we describe
several steps toward such a theory. We show that under mild hypotheses,
the dictionary learning problem is locally well-posed: the desired
solution is indeed a local minimum of the L1 norm. Namely, if A is an
incoherent (and possibly overcomplete) dictionary, and the coefficients X
follow a random sparse model, then with high probability (A;X) is a local
minimum of the L1 norm over the manifold of factorizations (A,X)
satisfying AX = Y , provided the number of samplesis sufficiently large.
For overcomplete A, this is the first result showing that the dictionary
learning problem is locally solvable. Our analysis draws on tools
developed for the problem of completing a low-rank matrix from a small
subset of its entries, which allow us to overcome a number of technical
obstacles; in particular, the absence of the restricted isometry
property. We also discuss several more restricted situations in which the
problem can be globally solved by efficient algorithms.
[ back to Schedule ]
Kai Yu at
NEC Laboratories America. Inc
Title: Large-scale Visual Recognition via Unsupervised
Feature Learning
Abstract: It has been widely known that the quality of
features is the key bottleneck for achieving good performances of visual
recognition. Rather than hand-crafting features, we advocate
automatically learning better image features from unlabeled data. Our
work extended sparse coding to a broader family of new unsupervised
nonlinear coding methods, including local coordinate coding and
super-vector coding, which explore the geometrical structure of sensory
image data and has theoretical justification for improving supervised
learning. The coding of image local features gives rise to significantly
better image representations, which enable simple linear classifiers to
achieve stronger performance than traditional nonlinear classifiers. The
methods achieved state-of-the-art results on a range of challenging scene
classification & object recognition tasks, including Caltech 101,
Caltech 256, PASCAL VOC, and large-scale problems like ImageNet,
involving millions of images and one thousand object categories.
Short Bio:Kai Yu is Head of Media Analytics Dept. at
NEC Laboratories America, where he manages an R&D division
responsible for the company's technology innovation in image recognition,
multimedia search, video surveillance, sensor mining, and human-computer
interaction. He has led his team to develop cutting-edge technologies
that won numerical awards, as well as commercial products reported by
CNN, Wall Street Journal, and CCTV. He is Area Chair for ICML 2010, ICML
2011, NIPS 2011, and ACML 2011. He is also an Adjunct Faculty at Computer
Science Dept., Stanford University, teaching "Introduction to
AI". Before joining NEC, he was a Senior Research Scientist at
Siemens. He received Ph.D in Computer Science from University of Munich
and B.Sc from Nanjing University.
[ back to Schedule ]
Zhi-Hua
Zhou at Nanjing University
Title: On the usefulness of multi-views in exploiting
unlabeled data
Abstract: In many real applications there are abundant
unlabeled data but the amount of labeled training examples are limited,
since labeling the data requires extensive human effort and expertise. To
effectively utilize unlabeled data to help improve performance,
semi-supervised learning and active learning have attracted much
attention during the past decade. It is noteworthy that when the data
have multiple views, such as in cross-media applications, helpful
information contained in the multi-views will be able to enable strong
process of semi-supervised learning and active learning. In this talk we
will introduce some advances in this area.
Bio: Zhi-Hua Zhou is a Professor at the Department of
Computer Science and Technology, Nanjing University. His research
interests are mainly in machine learning, data mining, pattern
recognition, and artificial intelligence. In these areas he has published
over 80 papers in leading international journals or conferences, and
holds 11 patents. He is an Associate Editor-in-Chief of Chinese
Science Bulletin, Associate Editor of IEEE Transactions on
Knowledge and Data Engineering and ACM Transactions on
Intelligent Systems and Technology, and on the editorial boards of
various other journals. He also served as guest editor for journal such
as Machine Learning, Pattern Recognition, IEEE
Intelligent Systems, etc. He is the Founding Steering Committee
Co-Chair of ACML, and Steering Committee member of PAKDD and PRICAI. He
served as Program Committee Chair/Co-Chair of PAKDD'07, PRICAI'08 and
ACML'09, Vice Chair or Area Chair or Senior PC for various conferences.
He is the Chair of the Machine Learning Technical Committee of the China
Association of Artificial Intelligence (CAAI), the Vice Chair of the
Artificial Intelligence and Pattern Recognition Technical Committee of
the China Computer Federation (CCF), and the Chair of the IEEE Computer
Society Nanjing Chapter.
[ back to Schedule ]
Tong Zhang
at Rutgers University
Title: Theory and Applications of Greedy Algorithms in
Machine Learning
Abstract: I will discuss various forms of greedy
algorithms that have been studied in the literature of machine learning,
statistics, signal processing, and optimization. In machine learning,
greedy algorithms are best known as boosting procedures for classification;
in statistics they have been studied for functional estimation; in signal
processing, they have been proposed for sparse recovery/compressed
sensing; and in optimization, they are investigated as numerical
procedures for some large scale optimization applications. We will
discuss connections among variations of greedy algorithms in these
different fields, their theoretical analysis and applications.
[ back to Schedule ]
Shuicheng
Yan at National University of Singapore
Title: Learning with Exclusive Regularizations:
Formulations and Solution
Abstract: In this talk, exclusive regularization is
explored for two tasks. In the first task, we address the problem of
multi-class classification problem in semi-supervised setting. A
regularized multi-task learning approach is presented to train multiple
binary-class Semi-Supervised Support Vector Machines (S3VMs) using the
one-vs-rest strategy within a joint framework. A novel type of
regularization, namely Positiveness Exclusive Regularization (PER), is
introduced to induce the following prior: if an unlabeled sample receives
significant positive response from one of the classifiers, it is less
likely for this sample to receive positive responses from the other
classifiers. In the second task, we introduce a novel
approach to multilabel image classification which incorporates a new type
of context — label exclusive context — with linear representation and
classification. Given a set of exclusive label groups that describe the
negative relationship among class labels, our method, namely LELR for
Label Exclusive Linear Representation, enforces exclusive assignment of
the labels from each group to a query image. The problem can be
formulated as an exclusive Lasso (eLasso) model with group overlaps and
affine transformation. For these two formulations for two tasks, we
propose a Nesterov-type smoothing approximation algorithm for efficient
optimization. Efficiency and effectiveness are validated by experiments
on several datasets.
Short Bio:Dr. Yan Shuicheng is currently an Assistant
Professor in the Department of Electrical and Computer Engineering at
National University of Singapore, and the founding lead of the Learning
and Vision Research Group (http://www.lv-nus.org).
Dr. Yan's research areas include computer vision, multimedia and machine
learning, and he has authored or co-authored over 200 technical papers
over a wide range of research topics. He is an associate editor of IEEE
Transactions on Circuits and Systems for Video Technology, and has been
serving as the guest editor of the special issues for TMM and CVIU. He
received the Best Paper Awards from ACM MM’10, ICME’10 and ICIMCS'09, the
winner prize of the classification task in PASCAL VOC'10, the honorable
mention prize of the detection task in PASCAL VOC'10, 2010 TCSVT Best
Associate Editor (BAE) Award, and the co-author of the best student paper
awards of PREMIA'09 and PREMIA'11.
[ back to Schedule ]
Xiaofei He
at Zhejiang University
Title: Parallel Vector Field Embedding
Abstract: In this talk, I will introduce our recent work
on manifold embedding from the perspective of vector field. Unlike graph
based techniques which try to preserve the distance, our approach tries
to find a parallel vector field on the manifold and then reconstruct the
embedding function via the obtained vector field. When we restrict the
vector field to the gradient field, our approach is equivalent to finding
killing vector field on manifold. Our analysis of killing field on
Euclidean space shows, when the manifold is locally isometric to a
connected subset of Euclidean space, we can always recover the manifold
isometrically. I will also present some experimental results on both
synthetic and real data sets. Particularly, our approach is very
insensitive to the noise.
[ back to Schedule ]
Yuanzheng
Si and Ming Jiang at Peking University
Title: Imaging fusion for diffuse optical tomography and
X-ray computed tomography
Abstract: Many imaging modalities have been developed
and applied such as x-ray computed tomography (XCT), diffuse optical
tomography (DOT) for structural and functional imaging of small animals,
respectively. Each imaging modality has its advantages and limitations.
XCT provides high-resolution anatomical images in bone but performs
poorly for soft tissues. DOT can provide the functional information of
optical absorption and diffusion properties of tissues and has been
recognized as a non-invasive molecular imaging modality for its
sensitivity, specificity, and non-radiation characteristics. However, DOT
suffers from low spatial resolution and contrast. Currently,
multimodality image fusion techniques are developed to make the best use
of multimodality image data. The key issue is how to fuse information
from other imaging modalities. Past research utilized the high-resolution
image data from CT/MRI as a priori information to guide the
reconstruction process of DOT by sequential imaging of XCT/MRI and then
DOT.
We have developed a hybrid imaging system that can
conduct XCT and DOT without repositioning small animals in the system. In
our work, the reconstruction of XCT and DOT are performed simultaneously
rather than sequentially as in the past work. We use the Kullback-Leibler
divergence to characterize the similarity between images from XCT and DOT
and developed an image fusion reconstruction algorithm based on the
recent superiorization method. Preliminary studies demonstrated the
feasibility of the proposed approach. This is a joint work with Heng Mao,
Yanbin Lu, Jiansheng Yang.
[ back to Schedule ]
Yi Ma at University of Illinois at Urbana-Champaign and
Microsoft Research Asia(pending)
[ back to Schedule ]
|
