Reducing Bias in Deep Learning Optimization: The RSGDM Approach