Workshops

There will be 11 workshops held at the end of the 23rd International Conference on Machine Learning (ICML-2006), on Thursday, June 29, 2006 at Carnegie Mellon University, Pittsburgh, Pennsylvania, U.S.A.

Workshops provide organizers and participants with an opportunity to focus intensively on a specific topic in machine learning or a closely related area. Workshops can choose to concentrate on emerging research topics, but can also be devoted to application issues, or to questions concerning the economic and social aspects of machine learning and data mining.

Workshop participants will not be given workshop proceedings in a hardcopy format. Instead, organizers of the workshops will make the proceedings available on their website prior to the conference by June 18, 2006.

Click here for locations of workshops and other information.

All workshops have a minimum of 20 registrants to take place.

Workshops

Descriptions

Learning with Nonparametric Bayesian Methods.
Dirichlet Processes and other nonparametric Bayesian (NPB) methods have originally been developed in statistics but are finding growing interest in the machine learning community. It has already been demonstrated that in some important machine learning applications, NPB has clear advantages over parametric solutions. We hope that this workshop will serve as a platform to discuss basic issues and recent developments in NPB.
ROC Analysis and Machine Learning.
The ICML'06 workshop on ROC analysis and Machine Learning is the third in a series of workshops on ROC (Receiver Operating Characteristic) Analysis in Artificial Intelligence and Machine Learning. ROC Analysis is related in a direct and natural way to cost/benefit analysis of diagnostic decision making, and it has been widely used in medicine for many decades. This third workshop is intended to investigate on the hot topics identified during the two previous workshops (e.g. multiclass extension, statistical analysis, alternative approaches), on the one hand, and to encourage cross-fertilisation with ROC practitioners in medicine, on the other hand, thanks to an invited medical expert and a focus on appropriate datasets (to be announced).
Learning in Structured Output Spaces.
The prediction of structured outputs is relevant in several application areas including natural language processing, computational biology, computer vision, robotics, and information retrieval. Significant contributions in these fields led to applications that have matured to a point beyond proofs of concepts. But the applications arising from this vast field also drive further needs such as semi-supervised and large scale methods which we want to address with this workshop. Moreover, this interdisciplinary workshop will serve as a forum for exchanging ideas, leveraging existing knowledge to other domains, and identifying future challenges and applications.
Open Problems in Statistical Relational Learning.
Statistical relational learning (SRL) addresses the challenge of applying statistical inference to problems which involve rich collections of objects linked together in complex relational networks. The last decade of SRL research has explored many different ways to combine statistical and relational models. We now have a much improved understanding of the strengths and weaknesses of various SRL representation languages, and have software systems for learning both model structure and parameters. The goal of this workshop is to look forward to the next five years of SRL research. What are the open problems and challenges? Some new problems have become clear through practical experience; other problems were hinted at in early SRL work and still seem daunting today. With the shared vocabulary and experience that the SRL community has developed, we expect be able to formulate the important research problems much more precisely than we could five or ten years ago. Because our goal is to look forward, the bulk of the workshop will be organized into open problem sessions with focused discussion and contributed presentations.
Kernel Machines and Reinforcement Learning.
Reinforcement Learning (RL) is an adaptive method for learning to make good decisions in complex, stochastic, unknown environments. Real-world applications of RL require efficient function approximators. Kernel methods are at the heart of many modern machine learning techniques and make it possible to derive efficient algorithms that work in function spaces of high representation power and come with PAC-style theoretical results. This workshop will be dedicated to bridging the gap between kernel methods and reinforcement learning. Submissions (8-15 pages) should be sent to the workshop organisers by April 30, 2006.
Knowledge Discovery from Data Streams.
Many sources produce data continuously. Examples include customer click streams, telephone records, large sets of web pages, multimedia data, and sets of retail chain transactions. These sources are called data streams. If the process is not strictly stationary (as most of real world applications), the target concept could gradually change over time. This is an incremental task that requires incremental learning algorithms that take drift into account. Data streams are increasingly important in the research community, as new algorithms are needed to process this streaming data in reasonable time. This is an important issue for different research areas like data mining, machine learning, OLAP, databases, etc. Many researches in these areas are designing new approaches or adapting some of the traditional algorithms to data streams. The goal of this workshop is to convene researchers who deal with decision rules, decision trees, association rules, clustering, filtering, preprocessing, post processing, feature selection, visualization techniques, etc. from data streams and related themes.
Applications of Multiple-Instance Learning.
This workshop has been cancelled due to lack of participation.
Machine Learning Algorithms for Surveillance and Event Detection.
A common task in many domains involves monitoring routinely collected data for anomalous events. This task is prevalent in surveillance and also in analysis of scientific data. We will refer to this monitoring process as event detection. Event detection has the potential to impact a wide range of important real-world applications, ranging from security, finance, public health, medicine, biology, environmental science, manufacturing, astrophysics, business and economics. In the recent past, human beings have had the laborious job of manually examining the collected data for events of interest. With the emergence of computers, many efforts have been made to replace manual inspection with an automated process. Data, however, have become increasingly complex in recent years. Multivariate records, images, video footage, audio recordings, spatial and spatio-temporal data, text documents, and even relational data are now routinely collected. One might expect that existing work in machine learning would be well-suited for this task. However, in practice, the peculiarities of the application often grossly violate the standard assumptions of machine learning. As a result, new algorithms need to be created in order to address these issues and fill an important gap in machine learning research which would impact many of the most pressing real-world applications being studied today. The focus of this workshop will be on machine learning algorithms for surveillance and event detection in complex forms of data, novel application areas for event detection, and new directions for this area of research.
Statistical Network Analysis: Models, Issues and New Directions.
This workshop focuses on probabilistic methods for network analysis, paying special attention to model design and computational issues of learning and inference. The PC reflects the breadth of interest in this topic from machine learning to statistics, and in applications from biology to sociology and physics. Participants will focus on descriptive models for (multiple) relationships, motifs and other features of the networks, as well as generative models arising from new machine learning approaches such as hierarchical mixtures. We aim at attracting a diverse group of modelers to ICML interested in developing new approaches to very large complex networks, and to introduce researchers from the machine learning community to an exciting new set of problems and approaches. To this end we will bring together statistical network modeling researchers from different communities (computer science, statistics, physics, biology, and social sciences), thereby fostering collaborations and intellectual exchange.
Reinforcement Learning Competition and Benchmarking Event.
In the last two years, the reinforcement learning community has moved towards the establishment of standard benchmark problems. It has been widely acknowledge that the community would benefit not only from having a collection of such problems, but also from having competitive events. This is the first RL competition, aiming to build on the momentum of the benchmarking event held at NIPS'2005. The competition will focus on controlling an octopus arm, with additional evaluations being held for blackjack, cat-mouse, mountain-car and cart-pole. We will use a client-server architecture, with the environment server running at McGill. We will evaluate both performance during training, as well as the performance of the learned policy. The workshop will consist of a disucssion of the results, presentations from the participating teams and a discussion of similar future events.
Structural Knowledge Transfer for Machine Learning. Transfer learning concerns applying knowledge learned in one (or a set of) task or domain (the source) to improve learning on another (or a set of) related task or domain (the target). Human learning greatly benefits from transfer, but its application in machine learning has received significant attention only recently. Although the philosophy of transfer has underlined much of AI and ML research, there has been recent interest in the machine learning community to overtly exploit knowledge transfer for learning faster and possibly better solutions. Transfer can be especially effective when the learned knowledge is suitably structured, for example, in a relational or hierarchical fashion. This workshop is devoted to transfer learning in all subareas of machine learning, including but not limited to, concept learning, clustering, reinforcement learning, analogy, and to applications and evaluation methodologies, with an emphasis on how the learned knowledge is structured and exploited.

Call for workshop proposals is here.