| | |
Summary:
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
SPECIAL ISSUE ON LEARNING IN NONSTATIONARY AND EVOLVING ENVIRONMENTS
Using a computational model to learn under various environments has been a well-researched field that produced
relevant results; unfortunately, the majority of these efforts rely on three fundamental assumptions: i) there is a
sufficient and representative data set to configure and assess the model performance; ii) data are drawn from a fi-
xed albeit unknown distribution; and iii) samples are mostly supposed to be independent. Alas, all these as-
sumptions often do not hold in many real-world applications, such as in the analysis of climate or financial data,
network intrusion, spam and fraud detection, electricity demand and industrial quality inspection among many
others. Recent efforts towards incremental and online learning allow us to relax the "sufficiency" requirement by
continuously updating a model to learn from small batches or online data, yet, the data that become available are
still assumed to be drawn from a fixed distribution. More recently, approaches commonly called concept drift
and to some extend domain adaptation algorithms, possibly in collaboration with change detection tests, have
attempted to remove this assumption, by accommodating a stream or batches of data whose underlying distribu-
tion changes over time. However, early efforts have made other assumptions, such as restricting the type of faults
or changes affecting the system or the distribution and are primarily heuristic in nature with several free parame-
ters to be fine-tuned.
Against this background, the need for a general framework to learn from and adapt to a changing environ-
ment can be hardly overstated. A special issue that discusses the state-of-the-art and latest results on detecting and
adapting to changes in underlying data distributions is very timely.
|
