Fig. 4From: A multi-task CNN learning model for taxonomic assignment of human virusesThe sliding window and ensembling strategy. For input reads that are longer than 50 bp, the read is split into overlapping 50-mers using a 1-bp sliding window. The predictions of each 50-mer sequences will be ensembled by a hard-voting system with a pre-defined threshold for the final predicted labelsBack to article page