arXiv subjects

Vasyl Pihur

Publications and source records attributed to Vasyl Pihur.

7 recordsLinked to original sources

arXiv2022

Fuzzy Substring Matching: On-device Fuzzy Friend Search at Snapchat

About 50% of all queries on Snapchat app are targeted at finding the right friend to interact with. Since everyone has a unique list of friends and that list is not very large (maximum a few thousand), it makes sense to perform this search locally, on users' devices. In addition, the friend list is already available for other purposes, such as showing the chat feed, and the latency savings can be significant by avoiding a server round-trip call. Historically, we resorted to substring matching, ranking prefix matches at the top of the result list. Introducing the ability to perform fuzzy search on a resource-constrained device and in the environment where typo's are prevalent is both prudent and challenging. In this paper, we describe our efficient and accurate two-step approach to fuzzy search, characterized by a skip-bigram retrieval layer and a novel local Levenshtein distance computation used for final ranking.

cs.IR↗

arXiv2022

A Stochastic Model for Estimating the Number of Offenders and Targets on Snapchat Platform

Snapchat, like many other social platforms, provides mechanisms for its users to report content and private/public interactions that violate their sense of safety and decency. From our experience and common sense, we can safely assume that not everybody makes an effort to report, leaving potentially a large number of offending users and content unnoticed. The goal of this work is to directly estimate the probability of someone reporting on Snapchat using current in-app reporting options and, thereby, to provide estimates of the total prevalence (count) of offenders and users subjected to their unwanted, unwelcome or unsafe interactions.

cs.SI↗

arXiv2022

Query Processing at Snapchat: How we Handle Query Completion, Suggestion and Localization

From the Publisher:Software is a commodity being sold across diverse language and cultural groups, whether in the commercial marketplace, or as customized applications. Developers must structure their applications so that they can be readily and cheaply localized for sale in this range of markets. Obvious differences such as scripts and languages must be understood as well as a range of more subtle cultural conventions. Further topics covered include: the overall architecture for internationalized products and an outline of an internationalization API; the use of computational linguistics methods; quality assurance, testing and documentation. Appendices contain summaries of the facilities available for localization on major platforms, characteristics of European languages, commercial tools and further reading. The book is aimed at small and medium sized software producers, and the IT departments of multinational corporations.

cs.IR↗

arXiv2019

The Podium Mechanism: Improving on the Laplace and Staircase Mechanisms

The Podium mechanism guarantees ($ε, 0$)-differential privacy by sampling noise from a \emph{finite} mixture of three uniform distributions. By carefully constructing such a mixture distribution, we trivially guarantee privacy properties, while minimizing the variance of the noise added to our continuous outcome. Our gains in variance control are due to the "truncated" nature of the Podium mechanism where support for the noise distribution is maintained as close as possible to the sensitivity of our data collection, unlike the \emph{infinite} support that characterizes both the Laplace and Staircase mechanisms. In a high-privacy regime ($ε< 1$), the Podium mechanism outperforms the other two by 50-70\% in terms of the noise variance reduction, while in a low privacy regime ($ε\to \infty$), it asymptotically approaches the Staircase mechanism.

cs.CR↗

arXiv2018

Differentially-Private "Draw and Discard" Machine Learning

In this work, we propose a novel framework for privacy-preserving client-distributed machine learning. It is motivated by the desire to achieve differential privacy guarantees in the local model of privacy in a way that satisfies all systems constraints using asynchronous client-server communication and provides attractive model learning properties. We call it "Draw and Discard" because it relies on random sampling of models for load distribution (scalability), which also provides additional server-side privacy protections and improved model quality through averaging. We present the mechanics of client and server components of "Draw and Discard" and demonstrate how the framework can be applied to learning Generalized Linear models. We then analyze the privacy guarantees provided by our approach against several types of adversaries and showcase experimental results that provide evidence for the framework's viability in practical deployments.

cs.CR↗

arXiv2015

Building a RAPPOR with the Unknown: Privacy-Preserving Learning of Associations and Data Dictionaries

Techniques based on randomized response enable the collection of potentially sensitive data from clients in a privacy-preserving manner with strong local differential privacy guarantees. One of the latest such technologies, RAPPOR, allows the marginal frequencies of an arbitrary set of strings to be estimated via privacy-preserving crowdsourcing. However, this original estimation process requires a known set of possible strings; in practice, this dictionary can often be extremely large and sometimes completely unknown. In this paper, we propose a novel decoding algorithm for the RAPPOR mechanism that enables the estimation of "unknown unknowns," i.e., strings we do not even know we should be estimating. To enable learning without explicit knowledge of the dictionary, we develop methodology for estimating the joint distribution of two or more variables collected with RAPPOR. This is a critical step towards understanding relationships between multiple variables collected in a privacy-preserving manner.

cs.CR↗

arXiv2014

RAPPOR: Randomized Aggregatable Privacy-Preserving Ordinal Response

Randomized Aggregatable Privacy-Preserving Ordinal Response, or RAPPOR, is a technology for crowdsourcing statistics from end-user client software, anonymously, with strong privacy guarantees. In short, RAPPORs allow the forest of client data to be studied, without permitting the possibility of looking at individual trees. By applying randomized response in a novel manner, RAPPOR provides the mechanisms for such collection as well as for efficient, high-utility analysis of the collected data. In particular, RAPPOR permits statistics to be collected on the population of client-side strings with strong privacy guarantees for each client, and without linkability of their reports. This paper describes and motivates RAPPOR, details its differential-privacy and utility guarantees, discusses its practical deployment and properties in the face of different attack models, and, finally, gives results of its application to both synthetic and real-world data.

cs.CR↗