Practical Data-Dependent Metric Compression with Provable Guarantees
We introduce a new distance-preserving compact representation of multi-dimensional point-sets. Given n points in a d-dimensional space where each coordinate is represented using B bits (i.e., dB bits per point), it produces a representation of size O( d (d B/ϵ) + n) bits per point from which one can approximate the distances up to a factor of 1 ±ϵ. Our algorithm almost matches the recent bound of indyk2017near while being much simpler. We compare our algorithm to Product Quantization (PQ) jegou2011product, a state of the art heuristic metric compression method. We evaluate both algorithms on several data sets: SIFT (used in jegou2011product), MNIST lecun1998mnist, New York City taxi time series guha2016robust and a synthetic one-dimensional data set embedded in a high-dimensional space. With appropriately tuned parameters, our algorithm produces representations that are comparable to or better than those produced by PQ, while having provable guarantees on its performance.
READ FULL TEXT