On InstaHide, Phase Retrieval, and Sparse Matrix Factorization
arXiv:2011.11181
Abstract
In this work, we examine the security of InstaHide, a scheme recently proposed by [Huang, Song, Li and Arora, ICML'20] for preserving the security of private datasets in the context of distributed learning. To generate a synthetic training example to be shared among the distributed learners, InstaHide takes a convex combination of private feature vectors and randomly flips the sign of each entry of the resulting vector with probability 1/2. A salient question is whether this scheme is secure in any provable sense, perhaps under a plausible hardness assumption and assuming the distributions generating the public and private data satisfy certain properties. We show that the answer to this appears to be quite subtle and closely related to the average-case complexity of a new multi-task, missing-data version of the classic problem of phase retrieval. Motivated by this connection, we design a provable algorithm that can recover private vectors using only the public vectors and synthetic vectors generated by InstaHide, under the assumption that the private and public vectors are isotropic Gaussian.
arXiv comment30 pages, to appear in ICLR 2021, v2: updated discussion of follow-up work
References in corpus (5)
- An algebraic characterization of injectivity in phase retrieval
- Alternating Minimization for Mixed Linear Regression
- Learning Non-overlapping Convolutional Neural Networks with Multiple Kernels
- Superpolynomial Lower Bounds for Learning One-Layer Neural Networks using Gradient Descent
- InstaHide's Sample Complexity When Mixing Two Private Images