Visualizing loss functions to better understand why Neural Accumulators/Arithmetic Logic Units perform better at certain tasks compared to vanilla MLPs.