Investigating a spline-based sigmoid activation to mitigate the vanishing-gradient problem in deep MLPs, with reproducible experiments and auto-generated comparisons against ReLU and Sigmoid.
