Advanced examples

Concrete-ML models

The following table summarizes the various examples in this section, along with their accuracies.

Model	Dataset	Metric	Clear	Quantized	FHE
Linear Regression	Synthetic 1D	R2	0.876	0.863	0.863
Logistic Regression	Synthetic 2D with 2 classes	accuracy	0.90	0.875	0.875
Poisson Regression	OpenML insurance (freq)	mean Poisson deviance	1.38	1.68	1.68
Decision Tree	OpenML spams	precision score	0.95	0.97	0.97*
XGBoost	Diabetes	MCC	0.48	0.52	0.52*

A * means that FHE accuracy was calculated on a subset of the validation set.

• LinearRegression.ipynb

• LogisticRegression.ipynb

• PoissonRegression.ipynb

• DecisionTreeClassifier.ipynb

• XGBClassifier.ipynb

Comparison of classifiers

• ClassifierComparison.ipynb

Deep learning

Model	Dataset	Metric	Clear	Quantized	FHE
Fully Connected NN	Iris	accuracy	0.947	0.895	0.895
Convolutional NN	Digits	accuracy	0.90	**	**

In this table, ** means that the accuracy is actually random-like, because the quantization we need to set to fullfill bitsize constraints is too strong.

• Fully Connected NN

• Convolutional NN