This page will contained the agenda of the ECAI 2010 'Active and Incremental' Workshop.
The 4 papers have 20 minutes for the oral presentations + 10 minutes for the questions and discussions.
- 14:00 - 15:00 - Invited Talk - Manuel Roveri (http://home.dei.polimi.it/roveri/)
- Title : Learning models in nonstationary environments: the Just-In-Time approach
Abstract: Most machine learning techniques assume, either explicitly or implicitly, that the data-generating process is stationary. This assumption guarantees that the model learnt during the initial training phase remains valid over time and that its performance is in line with our expectations. Unfortunately, this assumption does not truly hold in the real world representing, in many cases, a simplistic approximation of the reality. The talk will describe the Just-In-Time (JIT) approach that is a flexible tool implementing the detection/adaptation paradigm to cope with evolving processes. Solutions following this approach improve the knowledge about the model in stationary conditions by exploiting additional information coming from the field during the operational life. Differently, in nonstationary conditions, as soon as a change in the data-generating process is detected, the learnt model is discarded and a suitable one activated to keep the performance. As a valuable and challenging application of the proposed approach, JIT classifiers for concept drift will be detailed and discussed.
- 15:00-15:30 - Paper 1 : Episodic Clustering of Data Streams Using a Topology-Learning Neural Network, Marko Tscherepanow, Sina Kühnel and Sören Riechers, Applied Informatics, Faculty of Technology, Bielefeld University (Germany)
Abstract : In this paper, an extension of
the unsupervised topologylearning
TopoART neural network is presented. Like TopoART, it is capable of
stable incremental on-line clustering of real-valued data. However, it
incorporates temporal information in such a way that consecutive input
vectors with a low distance in the input space are summarised to
episode-like clusters. Inspired by natural memory systems, we propose
two recall methods enabling the selection and retrieval
of these episodes. They are demonstrated at the example of a
video stream recorded in a natural environment.
- 15:30-16:00 - Paper 2 : Online Active Constraint Selection For Semi-Supervised Clustering, Caiming Xiong, David Johnson and Jason Corso - Suny at Buffalo (United States)
Abstract : Due to strong demand for the
ability to enforce top-down structure on clustering results,
semi-supervised clustering methods using pairwise constraints as side
information have received increasing attention in recent years. However,
most current methods are passive in the sense that the side information
is provided beforehand and selected randomly. This may lead to the use
of constraints that are redundant, unnecessary, or even harmful to the
clustering results. To overcome this, we present an active clustering
framework which selects pairwise constraints online as clustering
proceeds, and propose an online constraint selection method that
actively selects pairwise constraints by identifying uncertain nodes in
the data. We also propose two novel methods for computing node
uncertainty: one global and parametric and the other one local and
nonparametric.We evaluate our active constraint selection method with
two different semisupervised clustering algorithms on UCI, digits, gene
and image datasets, and achieve results superior to current state of
the art active techniques.
- 16:00-16:15 - Coffee Break
- 16:45-17:15 - Paper 4 : An Incremental On-line Classifier for Imbalanced, Incomplete, and Noisy Data, Marko Tscherepanow and Sören Riechers - Applied Informatics, Faculty of Technology, Bielefeld University (Germany)
Abstract : Incremental on-line learning is a research topic gaining
increasing interest in the machine learning community. Such learning
methods are highly adaptive, not restricted to distinct training and application
phases, and applicable to large volumes of data. In this paper,
we present a novel classifier based on the unsupervised topologylearning
neural network TopoART. We demonstrate that this classifier
is capable of fast incremental on-line learning and achieves excellent
results on standard datasets.We further show that it can successfully
process imbalanced, incomplete, and noisy data. Due to these
properties, we consider it a promising component for constructing
artificial agents operating in real-world environments.
- 17:15 – 18:00 Participants Discussion