Parameter Efficient Deep Learning Architectures in Vision