Post date: Aug 26, 2015 8:06:52 PM
Learning in non-stationary environments
Most of machine learning applications assume the stationarity hypothesis for the process generating the data. This amenable assumption is so widely –and implicitly- accepted that sometimes we even forget that it does not generally hold in the practice due to concept drift (i.e., a structural change in the process generating the acquired
datastreams). The ability to detect concept drift and react accordingly is hence a major achievement for intelligent learning machines and constitutes one of the hottest research topics for embedded systems. This ability allows the machine for actively tuning the application to maintain high performance, changing online the operational strategy, detecting and isolating possible occurring faults to name a few relevant tasks. The talk will focus on “Learning in a non-stationary environments”, by introducing both passive and active approaches. The active approach will be deepened by presenting triggering mechanisms based on Change point methods and Change detection tests. Finally, the just-in-time detect&react mechanism is introduced where, following a detected change, the system immediately reacts with a strategy depending on the available information.
When and Where
Ryerson University
George Vari Engineering and Computing Center
Room ENG287
11-12pm, October 6th, 2015