WikiAINeural NetworksConvolutionalNormalisationNormalisationTo keep sensible layer by layerApply kernel to same location of all channelsPixels in window divided by sum of pixel within volume across channelsMax PoolingUpConv