Convolutional neural networks have strong capabilities in image recognition and typically use coarse-graining schemes that reduce the numbers of pixels in the image. They perform discrete convolutions accompanied by information reduction and can be very successful in probabilistic classification. The recurrent neural networks in addition to passing forward information they retain information from previous steps. This makes them particularly useful in time-series predictions. The long short-term memory network has a cell that can retain both short and longer term memory.