Skip to Main Content
We propose a context recognizing incremental adaptive network to classify user contexts. The preferred approach is to evaluate the previous insertion by the observation of the error. Because each insertion influences the local behavior, the observed error should also be a local measurement and not the average error on the task. Such an insertion-evaluation cycle allows a local optimization, but decreases the ability to allocate new nodes if the previous insertions were not successful. In a non-stationary environment with an unknown amount of data, it would cause a severe problem with gradually developing. So, in parallel to the error-driven insertion an additional distance-driven insertion is considered. The proposed method is applicable to providing context-aware services in ubiquitous environment.