Fuzzy Pooling

Convolutional Neural Networks (CNNs) are artificial learning systems typically based on two operations: convolution, which implements feature extraction through filtering, and pooling, dimensionality reduction. The impact of pooling in the classification performance CNNs has been highlighted several previous works, a variety alternative operators have proposed. However, only few them tackle with uncertainty that is naturally propagated from input layer to maps hidden layers convolutions. In this paper we present novel operation (type-1) fuzzy sets cope local imprecision maps, investigate its context image classification. Fuzzy performed by fuzzification, aggregation defuzzification map neighborhoods. It used for construction can be applied as drop-in replacement current, crisp, CNN architectures. Several experiments using publicly available datasets show proposed approach enhance CNN. A comparative evaluation shows it outperforms state-of-the-art approaches.