AOLNSE 2013 - Special Session on Adaptive and Online Learning in Non-Stationary Environments

The computerization of many life activities and the advances in data collection and storage technology lead to obtain mountains of data. They are collected to capture information about a phenomena or a process behavior. These data are rarely of direct benefit. Thus, a set of techniques and tools are used to extract useful information for decision support, prediction, exploration and understanding of phenomena governing the data sources. The information is mostly provided in terms of system models describing the behaviour of the actual system or application under examination.
Whenever dynamic process changes occur due to changing system states, varying operation modes, or environmental conditions, the information content extracted from older (batch off-line) data sources needs to be adjusted; otherwise, the models may deteriorate significantly in performance. In on-line settings, this circumstance requires permanent updates of model components and parameters, in offline applications a transfer of old models to new states.
Therefore, adaptive and dynamic data-driven learning methodologies play an important role, as they are able to cope with dynamically and continuously evolving environments in order to keep the quality of the system models permanently up-to-date and on a high level. In particular, the methodologies typically employed are able to adjust the models to new system states and operation modes on-the-fly.
Incremental learning concepts and evolution of model components play a key role during model adaptation in order to avoid time-intensive re-training phases. Thus, the models equipped with thesetechnologies are also often called evolving models or in a broader sense evolving intelligent systems.
Important issues in these evolving learning mechanisms are dealing with upcoming drifts appropriately (achieving a reasonable balance between continuous learning and “forgetting”), keeping the supervision effort of operators at a low level, dealing with high-dimensional learning problems (omitting curse of dimensionality) as well as allowing a fast processing (by e.g. keeping the complexity of the models low).
This special session looks to gather and discuss efficient techniques, methods and tools able to manage, to exploit and to interpret correctly the increasing amount of data in environments that are continuously changing. The goal is to build models for predicting the future system behavior, able to tackle and to govern the high variability of complex non-stationary systems.
TOPICS
This session would solicit original research papers including but not limited to the following:
? Incremental learning methods,
? Adaptive, life-long and sequential learning,
? On-line classification and regression methods,
? Evolving structural components and systems modelling
? Incremental, evolving un-supervised methods? Incremental/on-line Dimension reduction methods
? Concepts to address drifts and shifts in data streams (weighting, gradual forgetting etc.)
? On-line complexity reduction, merging and splitting concepts
? On-line/Incremental Active and Semi-supervised learning concepts
? On-line Human-machine interaction and the incorporation of background knowledge
? Transfer learning
? Adaptive data pre-processing and knowledge discovery
? Applications in the field of dynamic, on-line, incremental learning