Workshop Abstracts
Advances in Structured Learning for Text and Speech Processing -- Koby
Crammer
This workshop is intended for researchers and students interested in developing and applying structured classification methods to text and speech processing problems. Recent advances in structured classification provide promising alternatives to the probabilistic generative models that have been the mainstay of speech recognition and statistical language processing. However, powerful features of probabilistic generative models, such as hidden variables and compositional combination of several kinds of evidence, do not transfer cleanly to all structured classification methods.
http://www.cis.upenn.edu/~crammer/workshop-index.html
Automatic
Discovery of Object Categories
-- Ian Fasel
This workshop is on learning to identify
both the presence and location of objects in arbitrary scenes using minimal
training information (i.e., without hand-segmenting objects in training
images).
Bayesian
Methods for Natural Language Processing -- Hal Daume III, Yee Whye Teh
Models for NLP problems are often
complex, and there is never enough data to properly estimate all the required
parameters. The application of
Bayesian methods could potentially result in more effective models. This workshop aims to bring together
researchers from the Bayesian community and the NLP community to enable
cross-fertilization.
http://www.isi.edu/~hdaume/BayesNLP/
Computational Methods for Gene Expression Analysis in Neuroscience -- Michael
Hawrylycz
The goal of this workshop is to
present and develop emerging problems and informatics tools needed for atlas based gene
expression. Speakers with
representative expertise in neuroscience and neuroinformatics, computational methods in gene
expression analysis and brain atlas development is being assembled who will
present current problems and
techniques. We encourage contributions describing progress on new informatics techniques applicable
to this area and work that is
substantially different from standard approaches.
http://www.brainatlas.org/main.asp?section=workshop
Decision-making in Complex Nervous Systems: A Multi-Level
Perspective -- Patricia Churchland
Moving appropriately requires
integration of signals, and a decision about what to do next, and perhaps next
after that. Are there distinct
decision-stages? How is relevant
background information tapped? How
are relevant options selected?
What is the role of attention? Research at many levels is beginning to
reveal the structure and organization of decisions, choices and plans.
Decoding
Brain States with Machine Learning: Applications to fMRI -- Luiz Pessoa
In the past few years, a growing
number of studies have taken a different approach to fMRI data analysis.
Instead of contrasting fMRI signals associated with two conditions, the
direction of analysis is reversed in order to probe whether brain signals can
be used to predict perceptual, motor, or cognitive states. In this framework,
fMRI data analysis can be viewed within a standard pattern classification
framework. Therefore, established statistical and machine learning techniques
can be employed, such as neural networks, Bayesian networks, and support vector
machines. Recently, this line of research has attracted considerable attention
in the neuroimaging field. By discovering distributed patterns of activation,
such multivariate techniques offer the potential to go beyond the simple
subtractive-univariate standard largely adopted in the field and have the
potential to provide a much richer characterization of neuroimaging data.
http://www.brown.edu/Research/LCE/NIPS.html
Foundations
of Active Learning -- Claire Monteleoni
This workshop brings together
researchers interested in providing solid foundations to active learning. The
focus includes theoretical characterizations of the problem, and algorithms
with various forms of performance guarantees, or intuitive justification and
empirical success, and appropriate applications. We also encourage work towards
a solid framework for benchmarking AL algorithms.
http://www.cs.huji.ac.il/~ranb/aln05/
Game
Theory, Machine Learning and Reasoning under Uncertainty -- Iead Rezek
Game theory is concerned with
understanding the decision processes and strategic actions of competing
individuals. Having initially found applications in economics and diplomacy,
game theory is increasingly being used to understand the interactions that
occur within large multi-agent systems, and has been applied in areas as
diverse as allocating resources within grid computing and coordinating the
behaviour of multiple autonomous vehicles.
http://www.robots.ox.ac.uk/~argus/NipsOverview.html
Inductive
Transfer: 10 Years Later -- Daniel L. Silver
Inductive transfer or transfer
learning refers to the problem of applying the knowledge learned in one or more
tasks to more efficiently develop a more effective hypothesis for a new
task. While all learning involves
generalization across problem instances, transfer learning emphasizes the
transfer of knowledge across domains, tasks, and distributions that are similar
but not the same. For example, learning to recognize chairs might help to
recognize tables; or learning to play checkers might improve the learning of
chess. While people are adept at inductive transfer, even across widely
disparate domains, there exists little associated computation learning theory
and few systems that exhibit knowledge transfer.
http://iitrl.acadiau.ca/itws05/
Intelligence
Beyond the Desktop -- Carlos Guestrin
In non-traditional computer
architectures like wireless sensor networks or multi-robot systems, machine
learning problems cannot faithfully be viewed in terms of a data set and an
objective function to optimize; physical aspects of the system impose
challenging new constraints. This
workshop will bring together researchers to discuss new applications of machine
learning in these systems, the challenges that arise, and emerging solutions.
http://ai.stanford.edu/~paskin/ibd05
Interclass
Transfer Workshop: Why Learning to Recognize Many Objects Might be Easier Than
Learning to Recognize Just One
-- Michael Fink
A recognition system benefits from
interclass transfer if multiple target classification tasks share common
underlying structures that can be used to facilitate training or detection. The
workshop is aimed at bringing together researchers interested in how interclass
transfer might be utilized by human and artificial object recognition systems.
http://www.cs.huji.ac.il/~fink/nips2005
Kernel
Methods and Structured Domains
-- Arthur Gretton
Substantial recent work in machine
learning has focused on the problem of dealing with inputs and outputs on more
complex domains than are provided for in the classical
regression/classification setting. Structured representations can give a more
informative view of input domains, which is crucial for the development of
successful learning algorithms: application areas include determining protein
structure and protein-protein interaction; part-of-speech tagging; the
organisation of web documents into hierarchies; and image segmentation.
Likewise, a major research direction is in the use of structured output
representations, which have been applied in a broad range of areas including
several of the foregoing examples (for instance, the output required of the
learning algorithm may be a probabilistic model, a graph, or a ranking). In particular, kernel methods have been
especially fertile in giving rise to efficient and powerful algorithms for both
structured inputs and outputs, since (as with SVMs) use of the "kernel
trick" can make the required optimisations tractable: examples include
large margin Markov networks, graph kernels, and kernels on automata. More generally, kernels between
probability measures have been proposed (with no a-priori assumptions as to the
dependence structure), which have motivated particular kernels between images
and strings.
http://nips2005.kyb.tuebingen.mpg.de/
Large Scale
Kernel Machines -- Leon Bottou
This workshop investigates
computationally efficient ways to exploit large datasets using kernel machines.
It will show how adequately designed kernel machines can efficiently process
millions of examples. It will also debate whether kernel machines are the best
way to achieve such objectives.
http://nipsworkshop.bottou.org/
Learning
from Heterogeneous Data -- William Noble
The goal of this workshop is to
present emerging methods for computational learning from heterogeneous
data. The workshop will include
descriptions of formal representational frameworks, particular algorithmic
approaches, and applications of new and existing methods to particularly
challenging integration problems in computational biology and other application
domains.
Email: noble@gs.washington.edu
http://noble.gs.washington.edu/hdata
Learning
to Rank -- Shivani Agarwal
The problem of
ranking, in which the goal is to learn an ordering or ranking over objects, has
recently gained much attention in machine learning. Progress has been made in formulating different forms of the
ranking problem, proposing and analyzing algorithms for these forms, and
developing theory for them.
However, a multitude of basic questions remain unanswered. This workshop aims to provide a forum
for discussion and debate among researchers interested in the topic of ranking,
with a focus on these basic questions.
The goal is not to find immediate answers, but rather to discuss
possible methods and applications, develop intuition, brainstorm on possible
directions and, in the process, encourage dialogue and collaboration among
researchers with complementary ideas.
http://web.mit.edu/shivani/www/Ranking-NIPS-05/
Machine Learning Based Robotics in
Unstructured Environments
-- Greg
Grudic
The dream of robots that work alongside or in lieu of people in natural
environments has long evaded researchers. Clearly, our approach to programming
robots needs to be examined and perhaps fundamentally changed. This workshop will investigate and
propose Machine Learning based approaches to autonomous robotic problems in
unstructured environments.
http://www.cs.colorado.edu/~janem/NipsMLR.html
Machine
Learning for Implicit Feedback and User Modeling -- Samuel Kaski
The workshop introduces a new challenging application area. Half of
the presentations come from a PASCAL Challenge of inferring intent of users
based on eye movement signals. The rest will broaden the scope towards other
implicit feedback signals such as click streams, and towards more general
problems of user modeling.
http://www.cis.hut.fi/inips2005/
Machine Learning in Finance -- John Moody
Machine learning (ML) and related methods have
produced some of the financial industry's most consistently profitable
proprietary trading strategies during the past 20 years. With markets, trade execution and financial
decision making becoming more automated and competitive, practitioners
increasingly recognize the need for ML.
This workshop brings together researchers from machine learning,
academic finance and the financial industry to discuss problems in finance where
ML may provide an edge.
ML themes include reinforcement learning, optimization
methods, recurrent and state space models, on-line algorithms, evolutionary
computing, kernel methods, bayesian estimation, wavelets, neural nets, SVMs,
boosting and multi-agent simulation.
Financial topics include high frequency data, trading strategies,
execution models, forecasting, volatility, extreme events, credit risk,
portfolio management, yield curve estimation, option pricing, and selection of
indicators, models and equilibria.
http://www.icsi.berkeley.edu/~moody/MLFinance2005.htm
Models of Behavioural Learning -- Chris Watkins
We believe that new models of behavioural learning are needed. The
proposed workshop would bring together talks on the widest possible range of
approaches to machine learning of behaviour, and would encourage interaction
between researchers in machine learning, robotics, and animal behaviour. The
purpose would be to suggest new models of behavioural learning, suitable as
frameworks for theoretical analysis and the development of practical
algorithms. Talks would concentrate on identifying and motivating various
formal models of behavioural learning.
http://www.cs.rhul.ac.uk/home/chrisw/NIPSworkshop.htm
New Methods and Problems in
Computational Biology -- Gunnar Raetsch
The field of computational biology has seen a dramatic growth over
the past few years, both in terms of new available data, new scientific
questions and new challenges and for learning and inference. In particular,
biological data is often relationally structured and highly diverse, thus
requires to combine multiple weak evidence from heterogeneous sources. These
could include sequenced genomes of a variety of organisms, gene expression data
from multiple technologies, protein sequence and 3D structural data, protein
interactions, gene ontology and pathway databases, genetic variation data (such
as SNPs), and an enormous amount of textual data in the biological and medical
literature. The new types of scientific and clinical problems, require to
develop new supervised and unsupervised learning approaches that can use these
growing resources.
http://www.fml.tuebingen.mpg.de/nipscompbio
Open Problems and Challenges for
Nonparametric Bayesian Methods in Machine Learning -- Matthew Beal and Yee Whye Teh
A forum for presentation and discussion of recent advances in
research using nonparametric Bayesian techniques from the statistical
disciplines to bear on problems in Machine Learning. We will discuss new techniques, new problems including
temporal and spatial processes, computational/inferential issues, as well as
the appropriateness, or not, of NPB.
http://aluminum.cse.buffalo.edu:8079/npbayes/nipsws05
Open
Problems in Gaussian Processes for Machine Learning -- Joaquin Quinonero Candela
The purpose of this workshop is first to bring together and compare
the state-of-the-art in Gaussian Process (GP) methods that are used in Machine
Learning, and then especially to focus on the corresponding open problems. To
guide the discussion, we have identified a list of 6 relevant open problems in
GPs, given at the end of this proposal. While it might be optimistic to expect
the workshop to provide answers to some of the open questions, we hope it will
help build a consensus about the promising research directions to follow. As
far as we can remember, there has never actually been a GP workshop at NIPS.
http://gp.kyb.tuebingen.mpg.de/
Reinforcement Learning Benchmarks
and Bake-offs II -- Martin Riedmiller
It is widely agreed that the field of reinforcement learning would
benefit from the establishment of standard benchmark problems and perhaps
regular competitive events (bake-offs).
Competitions can greatly increase the interest and focus in an area by
clarifying its objectives and challenges, publicly acknowledging the best
algorithms, and generally making the area more exciting and enjoyable. Standard benchmarks can make it much easier
to apply new algorithms to existing problems and thus provide clear first steps
toward their evaluation.
http://www.ni.uos.de/rl_workshop05
The Accuracy-Regularization
Frontier -- Nathan Srebro
The workshop is concerned with computing, and using, the entire
"regularization path": the path of predictors corresponding to all
values of a trade-off parameter, such as the SVM slack penalty. These make up the frontier of
predictors for which it is not possible to reduce both the empirical error and
the regularizer.
http://www.cs.toronto.edu/~nati/Front/
Towards Human-Level AI? -- Andrew Y. Ng
Modern machine learning has developed a large set of techniques
over the last 10 years, to
the point that they are practically applicable in circumscribed domains. They include:
Bayesian networks, Bayesian
learning, kernel methods, supervised and unsupervised learning, and reinforcement learning. Can these types of techniques be
integrated and applied to
make progress towards the human-level AI problem? If not,
where and why do they break?
Additionally, are there new research problems that now urgently need our attention, if our
long-term goal is human -- or even dog-level AI?
http://www.cs.stanford.edu/groups/nips05-AI-Workshop/
Theoretical
Foundations of Clustering
-- Shai Ben-David
Despite this large number of algorithms and applications of
clustering, the theoretical foundations of clustering seem to be distressingly
meager. We wish to initiate a
discussion on principles of clustering that are independent of any particular
algorithm, objective function, or generative data model. We hope that a theory of clustering
will enhance the transfer of tools developed in related areas (like machine
learning and information theory) to the clustering domain.
http://www.ipsi.fraunhofer.de/~ule/clustering_workshop_nips05/clustering_workshop_nips05.htm
Value of Information in Inference,
Learning and Decision-Making
-- Alina
Beygelzimer
Value of Information (VOI) analysis
provides a principled methodology
for acquiring information in a way
that optimally trades off the cost
of information gathering with the
expected benefit in some overall
objective. Maximizing VOI is a common goal in
diverse research topics
ranging from active learning to
Bayesian problem diagnosis to
exploration- exploitation
tradeoffs. The goal of the
workshop is to
bring together researchers from
several fields concerned with VOI
analysis in the hope of igniting
cross-fertilization between the
areas.
http://www.research.ibm.com/nips05workshop
Workshop on Activity Recognition
and Discovery -- Dieter Fox
Recognition of high-level events from data streams is a much needed
resource towards building intelligent machines that provide automatic and
autonomous support in our every day lives. Recently, there has been significant progress towards such
activity recognition and discovery from data. Research progress is evident in (a) building real systems
that extract information from a variety to sensors, to (b) developing and
extending statistical machine learning approaches to model data for recognition
and matching, to (c) study of specific domains to extract relevant higher-level
context which in turn can be leveraged to support recognition.
http://www.cs.washington.edu/nips05-ARD
Workshop on Machine Learning for
Implicit Feedback and User Modeling – Samuel Kaski
Te workshop introduces a new challenging application area. Half of
the presentations come from a PASCAL Challenge of inferring intent of users
based on eye movement signals. The
rest will broaden the scope towards other implicit feedback signals such as
click streams, and towards more general problems of user modeling.