Enhanced Self Attention In Deep Learning