For example, if we have a (1K, 1K) image in RGB, then our input layer would take about 3 million inputs. If we have 1K units in the second layer, the corresponding weights will have (1K, 3M) dimensions. Training 3B parameters requires a lot of computational power and a large amount of data to be used for training. Convolutional neural networks incorporate convolution operation on the input layer to reduce the dimensionality and provide the neural network with more valuable features of the data.