Searching Privately by Imperceptible Lying: A Novel Private Hashing Method with Differential Privacy

Abstract

In the big data era, with the increasing amount of multi-media data, approximate nearest neighbor (ANN) search has been an important but challenging problem. As a widely applied large-scale ANN search method, hashing has made great progress, and achieved sublinear search time with low memory space. However, the advances in hashing are based on the availability of large and representative datasets, which often contain sensitive information. Typically, the privacy of this individually sensitive information is compromised. In this paper, we tackle this valuable yet challenging problem and formulate a task termed as private hashing, which takes into account both searching performance and privacy protection. Specifically, we propose a novel noise mechanism, i.e., Random Flipping, and two private hashing algorithms, i.e., PHashing and PITQ, with the refined analysis within the framework of differential privacy, since differential privacy is a well-established technique to measure the privacy leakage of an algorithm. Random Flipping targets binary scenarios and leverages the "Imperceptible Lying" idea to guarantee ϵ-differential privacy by flipping each datum of the binary matrix (noise addition). To preserve ϵ-differential privacy, PHashing perturbs and adds noise to the hash codes learned by non-private hashing algorithms using Random Flipping. However, the noise addition for privacy in PHashing will cause severe performance drops. To alleviate this problem, PITQ leverages the power of alternative learning to distribute the noise generated by Random Flipping into each iteration while preserving ϵ-differential privacy. Furthermore, to empirically evaluate our algorithms, we conduct comprehensive experiments on the image search task and demonstrate that proposed algorithms achieve equal performance compared with non-private hashing methods.

Locality Sensitive Hashing with Extended Differential Privacy

“…Finally, we note that some studies have proposed privacy-preserving LSH [4,9,18,57,65]. However, some of them [4,18,57] only apply LSH and claim that it protects user privacy because LSH is a kind of non-invertible transformation.…”

“…In Section 3, we showed that the lack of rigorous guarantees can lead to privacy breach. Wang et al [65] flipped each element of a hash with probability 𝛽 = 𝑒 −𝜀 and claimed that it provides 𝜀-DP. However, this claim is wrong; e.g., if 𝛽 = 1, the flipped hash completely reveals the information about the original hash; in addition, 𝜀 must increase with increase in the length of the hash.…”

Video-Text Retrieval by Supervised Sparse Multi-Grained Learning

“…Video-Text Retrieval (VTR), which involves crossmodal alignment and abstract understanding of temporal images (videos), has been a popular and fundamental task of language-grounding problems (Wang et al, 2020a(Wang et al, ,c, 2021Yu et al, 2023). Most existing conventional video-text retrieval frameworks (Yu et al, 2017;Dong et al, 2019;Zhu and Yang, 2020;Miech et al, 2020;Gabeur et al, 2020;Dzabraev et al, 2021;Croitoru et al, 2021) focus on learning powerful representations for video and text and extracting separated representations.…”

InvGC: Robust Cross-Modal Retrieval by Inverse Graph Convolution

“…For example, in text-to-video retrieval, the objective is to rank gallery videos based on the features of the query text. Recently, inspired by the success in self-supervised learning (Radford et al, 2021), significant progress has been made in CMR, including image-text retrieval (Radford et al, 2021;Li et al, 2020;Wang et al, 2020a), video-text retrieval (Chen et al, 2020;Cheng et al, 2021;Gao et al, 2021;Lei et al, 2021;Ma et al, 2022;Park et al, 2022;Wang et al, 2022a,b;Zhao et al, 2022;Wang and Shi, 2023;, and audiotext retrieval (Oncescu et al, 2021), with satisfactory retrieval performances.…”