Workshop on Knowledge Discovery, Modeling, and Simulation (KDMS-2011)

http://discoveryandmodeling.org

San Diego, CA, August 21, 2011
(co-located with SIG-KDD 2011)

This first Workshop on Knowledge Discovery, Modeling and Simulation will be held in conjunction with the 17th ACM SIGKDD Conference on Knowledge Discovery and Data Mining that will take place on August 21-24, 2011 in San Diego, CA.

Knowledge discovery (KD) uncovers patterns and relationships within data. These patterns and relationships can, in turn, be used to create models of data. Simulations based on these models then can be used to generate data and start the cycle anew. Modeling and simulation (M&S) can benefit from KD by providing a means to evaluate the realism, novelty, interestingness, and utility of the created model. KD can benefit M&S by providing a means to understand the large amount of data generated by simulations. Recent developments in the ability to mine complex, highly-structured data of the form typically produced by large scale simulations may provide opportunities for useful knowledge discovery of simulation outputs.

Much of today’s KD research emphasizes the “learning” of correlated factors among data objects. Combining different models using hypothetical cause-effect relationships, we can investigate the direct/indirect conditioning of factors until the model matches up with the data. From the M&S research community, there are open questions about how to best import large-scale datasets into simulations and calibrate the models from this data. For example, how does a model (simulation) adapt to revised (joint) probability distributions when new patterns are discovered (over time)? Conversely, the KD community faces the issues of determining the pedigree of input data, as well as validation of the patterns discovered in the data (possibly reduced) generated by complex simulations.

The organizers of this workshop believe that combining KD and M&S will be useful in a variety of application domains, including social sciences, economics, earth sciences, and life sciences. Moreover, there are a number of sub-disciplines of machine learning and data mining that lie at the confluence of KD with M&S, including graphical models, statistical relational learning, evolutionary computation and clustering.

For example, models and simulations of financial transactions depend on accurate representation of social data, and the discovery of new patterns among these data. With the extremely large data volumes in the health care and energy domains, models and simulations intended to support resource planning rely on effective and rapid clustering techniques to learn usage patterns, as well as the early detection of patterns not previously represented in the model or simulation. And for those with a more physics-based modeling perspective, KD developments in time point-based and interval-based methods, as well as univariate and multivariate methods, are potentially applicable to complex climate models when combined with methods of spatial outlier detection.

The goal of this workshop is to bring together researchers from a variety of these disciplines to define a research agenda for the convergence of two previously separate research areas: KD and M&S. Specific objectives are to identify:

How knowledge discovery can produce patterns that can be useful in the development of models;
How to effectively and usefully mine the results of large numbers of potentially complex simulations created by executing models; and
Other potential synergies that may result from “closing the loop” between KD and M&S.

As this is the first workshop of its nature in the KDD conference community, we hope participants will walk away with a better sense of the collaboration possible across these disciplines and a better appreciation of the tools available for combining KD and M&S techniques. To begin this conversation, we hope to explore the state-of-the-art algorithms and methods, leveraging existing knowledge from KD sub-disciplines to identify real world applications and future challenges. In particular, we are interested in the following KD topics:

Probabilistic models for structured data
(Multi-) relational data mining/Statistical Relational Learning
Network analysis & mining
Large-scale learning and applications
Mining of spatial and temporal data
Semi-supervised learning
Clustering, Segmentation
Data mining feedback during simulation (dashboards, avatar guidance)
Active learning
Transductive inference
Transfer learning

Since our challenge is to identify the application of KD techniques and tools to real world applications, we call for papers in a variety of M&S domains, including:

Social science, e.g., business interactions, cultural dynamics, crisis management
Earth science, e.g., meteorology, astronomy
Life science, e.g., medical diagnosis, pharma, biology, pandemic modeling
Infrastructure/logistics, e.g., emergency management, production scheduling, traffic flow
Massive online gaming, e.g., massive virtual reality training and exercise environments
Engineering, e.g., aircraft development, building design, energy efficiency
Cyber security, e.g., agent-based, large-scale simulation of networks
Business processes, e.g., strategic planning, resource allocation, portfolio management, marketing

We invite researchers working at the confluence of KD and M&S to submit regular or position papers describing the major points and/or results they would present during a talk or discuss in a panel session. Regular papers are a maximum of 8 pages long in two-column format, position papers comprise 2 pages. All authors should use the ACM KDD conference paper format with these lengths, to include all graphics and references.

Authors whose regular papers are accepted to the workshop will have the opportunity to give a short (12-15 minute) presentation at the workshop. Authors whose position papers are accepted to the workshop will have the opportunity to present a 2 minute summary of their position, followed by participation in a panel session to promote interaction and dialogue.

Papers should be submitted by June 8, 2011.

The workshop itself currently is planned as a half-day workshop. We’ll begin with a keynote speaker, followed by domain-specific session showcasing accepted papers, and panel sessions to promote dialogue.

We'll conclude the workshop with presentation of certificates and monetary awards for best student papers.