2019-06-25-beimel19a.md

abstract	section	title	layout	series	id	month	tex_title	firstpage	lastpage	page	order	cycles	bibtex_author	author	date	address	publisher	container-title	volume	genre	issued	pdf	extras
We present a private agnostic learner for halfspaces over an arbitrary finite domain $X\subset \R^d$ with sample complexity $\mathsf{poly}(d,2^{\log^*|X|})$. The building block for this learner is a differentially private algorithm for locating an approximate center point of $m>\mathsf{poly}(d,2^{\log^*|X|})$ points – a high dimensional generalization of the median function. Our construction establishes a relationship between these two problems that is reminiscent of the relation between the median and learning one-dimensional thresholds [Bun et al. FOCS ’15]. This relationship suggests that the problem of privately locating a center point may have further applications in the design of differentially private algorithms. We also provide a lower bound on the sample complexity for privately finding a point in the convex hull. For approximate differential privacy, we show a lower bound of $m=\Omega(d+\log^*|X|)$, whereas for pure differential privacy $m=\Omega(d\log|X|)$.	contributed	Private Center Points and Learning of Halfspaces	inproceedings	Proceedings of Machine Learning Research	beimel19a	0	Private Center Points and Learning of Halfspaces	269	282	269-282	269	false	Beimel, Amos and and Moran, Shay and Nissim, Kobbi and Stemmer, Uri	given family Amos Beimel given family Shay Moran given family Kobbi Nissim given family Uri Stemmer	2019-06-25		PMLR	Proceedings of the Thirty-Second Conference on Learning Theory	99	inproceedings	date-parts 2019 6 25	http://proceedings.mlr.press/v99/beimel19a/beimel19a.pdf