Edwin Lughofer,
"Key Issues of Incremental Learning in Intelligent Systems"
, FLLL-TR-0701, Fuzzy Logic Laboratorium Linz, A-4232 Hagenberg, 8-2007
Original Titel:
Key Issues of Incremental Learning in Intelligent Systems
Sprache des Titels:
Englisch
Original Kurzfassung:
In nowadays intelligent industrial systems
incremental learning can be seen as THE engine for adaptive and evolving modelling tasks during on-line operation modes.
After describing the purpose of and requirements for incremental learning in intelligent industrial systems and explaining its characteristics, some key problems for guaranteeing safe,
robust and a high-performance incremental, on-line learning procedures are discussed in this paper. These include the
following aspects: components to learn, robustness of incremental learning procedures, incorporating new system states on demand, evolving models with changing input structure,
timing of adaptation, fault and outlier treatment and dealing with drifts in data streams.