Skip to main content
Fig. 2 | Biology Direct

Fig. 2

From: Efficient differentially private learning improves drug sensitivity prediction

Fig. 2

The effect of bounding data for differentially private learning of a regression model. Top: Bounding the data increasingly tightly (by B; green square) brings 1D robust private linear regression models (blue lines illustrating the distribution of results of the privacy-preserving algorithm) closer to the non-private model (black line) as less noise needs to be injected. Blue points: data. Bottom: The data are bounded in robust private linear regression by projecting outliers within the bounds (red lines; shown only for a subset of the points)

Back to article page