Sampling near neighbors in search for fairness

Similarity search is a fundamental algorithmic primitive, widely used in many computer science disciplines. Given set of points S and radius parameter r > 0, the -near neighbor ( -NN) problem asks for data structure that, given any query point q , returns p within distance at most from q. In this paper, we study -NN light individual fairness providing equal opportunities: all that are should have same probability to be returned. The special interest high dimensions, where Locality Sensitive Hashing (LSH), theoretically leading approach similarity search, does not provide guarantee. work, show LSH-based algorithms can made fair, without significant loss efficiency. We propose several efficient structures exact approximate variants fair NN problem. Our works more generally sampling uniformly sub-collection sets collection few other applications. also carried out an experimental evaluation highlights inherent unfairness existing structures.