ExpO

O -regularized model as well (this differs from the'support2' binary classification problem where we explain each logit The results are in Table 5. SENN did produce a more interpretable model. O or SENN model is better. But, looking at the MAPLE-NF metric, we can see that its explanations have a standard error of around 4% relative to the model's predicted probability. O-regularized model is more interpretable than SENN. Notice that the Taylor approximation-based explanations are more strongly influenced by local variations in the function. These regularizers may make the network simpler (due to sparser weights) or smoother, which may make it more amenable to local explanation.