Optimizing Extreme Learning Machines via Ridge Regression and Batch Intrinsic Plasticity