The output of the convolutional layer is normally handed with the ReLU activation purpose to bring non-linearity to the model. It will require the characteristic map and replaces all the detrimental values with zero. Zero-padding—It allows us to manage the spatial size from the output volume by padding zeros https://financefeeds.com/arcana-network-launches-chain-abstraction-sdk-to-unify-multi-chain-experience/