Machine learning (ML) is playing an increasing role in decision-making tasks that directly affect individuals, e.g., loan approvals, or job applicant screening. Significant concerns arise that, without special provisions, individuals from under-privileged backgrounds may not get equitable access to services and opportunities. Existing research studies fairness with respect to protected attributes such as gender, race or income, but the impact of location data on fairness has been largely overlooked. With the widespread adoption of mobile apps, geospatial attributes are increasingly used in ML, and their potential to introduce unfair bias is significant, given their high correlation with protected attributes. We propose techniques to mitigate location bias in machine learning. Specifically, we consider the issue of miscalibration when dealing with geospatial attributes. We focus on spatial group fairness and we propose a spatial indexing algorithm that accounts for fairness. Our KD-tree inspired approach significantly improves fairness while maintaining high learning accuracy, as shown by extensive experimental results on real data.
Large-scale three-dimensional spatial data has gained increasing attention with the development of self-driving, mineral exploration, CAD, and human atlases. Such 3D objects are often represented with a polygonal model at high resolution to preserve accuracy. This poses major challenges for 3D data management and spatial queries due to the massive amounts of 3D objects, e.g., trillions of 3D cells, and the high complexity of 3D geometric computation. Traditional spatial querying methods in the Filter-Refine paradigm have a major focus on indexing-based filtering using approximations like minimal bounding boxes and largely neglect the heavy computation in the refinement step at the intra-geometry level, which often dominates the cost of query processing. In this paper, we introduce 3DPro, a system that supports efficient spatial queries for complex 3D objects. 3DPro uses progressive compression of 3D objects preserving multiple levels of details, which significantly reduces the size of the objects and has the data fit into memory. Through a novel Filter-Progressive-Refine paradigm, 3DPro can have query results returned early whenever possible to minimize decompression and geometric computations of 3D objects in higher resolution representations. Our experiments demonstrate that 3DPro out-performs the state-of-the-art 3D data processing techniques by up to an order of magnitude for typical spatial queries.
Millions of videos are ubiquitously generated and shared everyday. Releasing videos would be greatly beneficial to social interactions and the community but may result in severe privacy concerns. To the best of our knowledge, most of the existing privacy preserving techniques for video data focus on detecting and blurring the sensitive regions in the video. Such simple privacy models have two major limitations: (1) they cannot quantify and bound the privacy risks, and (2) they cannot address the inferences drawn from the background knowledge on the involved objects in the videos. In this paper, we first define a novel privacy notion ϵ-Object Indistinguishability for all the predefined sensitive objects (e.g., humans and vehicles) in the video, and then propose a video sanitization technique VERRO that randomly generates utility-driven synthetic videos with indistinguishable objects. Therefore, all the objects can be well protected in the generated utility-driven synthetic videos which can be disclosed to any untrusted video recipient. We have conducted extensive experiments on three real videos captured for pedestrians on the streets. The experimental results demonstrate that the generated synthetic videos lie close to the original video for retaining good utility while ensuring rigorous privacy guarantee.
The concept of similarity is used as the basis for many data exploration and data mining tasks. Nearest Neighbor (NN) queries identify the most similar items, or in terms of distance the closest points to a query point. Similarity is traditionally characterized using a distance function between multi-dimensional feature vectors. However, when the data is high-dimensional, traditional distance functions fail to significantly distinguish between the closest and furthest points, as few dissimilar dimensions dominate the distance function. Localized similarity functions, i.e. functions that only consider dimensions close to the query, quantize each dimension independently and only compute similarity for the dimensions where the query and the points fall into the same bin. These quantizations are query-agnostic. There is potential to improve accuracy when a query-dependent quantization is used. In this paper we propose a Query dependent Equi-Depth (QED) on-the-fly quantization method to improve high-dimensional similarity searches. The quantization is done for each dimension at query time and localized scores are generated for the closest p fraction of the points while a constant penalty is applied for the rest of the points. QED not only improves the quality of the distance metric, but also improves query time performance by filtering out non relevant data. We propose a distributed indexing and query algorithm to efficiently compute QED. Our experimental results show improvements in classification accuracy as well as query performance up to one order of magnitude faster than Manhattan-based sequential scan NN queries over datasets with hundreds of dimensions.
Many modern applications involve collecting large amounts of data from multiple sources, and then aggregating and manipulating it in intricate ways. The complexity of such applications, combined with the size of the collected data, makes it difficult to understand the application logic and how information was derived. Data provenance has been proven helpful in this respect in different contexts; however, maintaining and presenting the full and exact provenance may be infeasible, due to its size and complex structure. For that reason, we introduce the notion of approximated summarized provenance, where we seek a compact representation of the provenance at the possible cost of information loss. Based on this notion, we have developed PROX, a system for the management, presentation and use of data provenance for complex applications. We propose to demonstrate PROX in the context of a movies rating crowd-sourcing system, letting participants view provenance summarization and use it to gain insights on the application and its underlying data.
Outlier or anomaly detection in large data sets is a fundamental task in data science, with broad applications. However, in real data sets with high-dimensional space, most outliers are hidden in certain dimensional combinations and are relative to a user's search space and interest. It is often more effective to give power to users and allow them to specify outlier queries flexibly, and the system will then process such mining queries efficiently. In this study, we introduce the concept of query-based outlier in heterogeneous information networks, design a query language to facilitate users to specify such queries flexibly, define a good outlier measure in heterogeneous networks, and study how to process outlier queries efficiently in large data sets. Our experiments on real data sets show that following such a methodology, interesting outliers can be defined and uncovered flexibly and effectively in large heterogeneous networks.
Differential privacy has recently emerged in private statistical data release as one of the strongest privacy guarantees. Most of the existing techniques that generate differentially private histograms or synthetic data only work well for single dimensional or low-dimensional histograms. They become problematic for high dimensional and large domain data due to increased perturbation error and computation complexity. In this paper, we propose DPCopula, a differentially private data synthesization technique using Copula functions for multi-dimensional data. The core of our method is to compute a differentially private copula function from which we can sample synthetic data. Copula functions are used to describe the dependence between multivariate random vectors and allow us to build the multivariate joint distribution using one-dimensional marginal distributions. We present two methods for estimating the parameters of the copula functions with differential privacy: maximum likelihood estimation and Kendall's τ estimation. We present formal proofs for the privacy guarantee as well as the convergence property of our methods. Extensive experiments using both real datasets and synthetic datasets demonstrate that DPCopula generates highly accurate synthetic multi-dimensional data with significantly better utility than state-of-the-art techniques.
The integration of information dispersed among multiple repositories is a crucial step for accurate data analysis in various domains. In support of this goal, it is critical to devise procedures for identifying similar records across distinct data sources. At the same time, to adhere to privacy regulations and policies, such procedures should protect the confidentiality of the individuals to whom the information corresponds. Various private record linkage (PRL) protocols have been proposed to achieve this goal, involving secure multi-party computation (SMC) and similarity preserving data transformation techniques. SMC methods provide secure and accurate solutions to the PRL problem, but are prohibitively expensive in practice, mainly due to excessive computational requirements. Data transformation techniques offer more practical solutions, but incur the cost of information leakage and false matches. In this paper, we introduce a novel model for practical PRL, which 1) affords controlled and limited information leakage, 2) avoids false matches resulting from data transformation. Initially, we partition the data sources into blocks to eliminate comparisons for records that are unlikely to match. Then, to identify matches, we apply an efficient SMC technique between the candidate record pairs. To enable efficiency and privacy, our model leaks a controlled amount of obfuscated data prior to the secure computations. Applied obfuscation relies on differential privacy which provides strong privacy guarantees against adversaries with arbitrary background knowledge. In addition, we illustrate the practical nature of our approach through an empirical analysis with data derived from public voter records.
Large data generated by scientific applications imposes challenges in storage and efficient query processing. Many queries against scientific data are analytical in nature and require super-linear computation time using straightforward methods. Spatial distance histogram (SDH) is one of the basic queries to analyze the molecular simulation (MS) data, and it takes quadratic time to compute using brute-force approach. Often, an SDH query is executed continuously to analyze the simulation system over a period of time. This adds to the total time required to compute SDH. In this paper, we propose an approximate algorithm to compute SDH efficiently over consecutive time periods. In our approach, data is organized into a Quad-tree based data structure. The spatial locality of the particles (at given time) in each node of the tree is acquired to determine the particle distribution. Similarly, the temporal locality of particles (between consecutive time periods) in each node is also acquired. The spatial distribution and temporal locality are utilized to compute the approximate SDH at every time instant. The performance is boosted by storing and updating the spatial distribution information over time. The efficiency and accuracy of the proposed algorithm is supported by mathematical analysis and results of extensive experiments using biological data generated from real MS studies.
Structured data including sets, sequences, trees and graphs, pose significant challenges to fundamental aspects of data management such as efficient storage, indexing, and similarity search. With the fast accumulation of graph databases, similarity search in graph databases has emerged as an important research topic. Graph similarity search has applications in a wide range of domains including cheminformatics, bioinformatics, sensor network management, social network management, and XML documents, among others. Most of the current graph indexing methods focus on subgraph query processing, i.e. determining the set of database graphs that contains the query graph and hence do not directly support similarity search. In data mining and machine learning, various graph kernel functions have been designed to capture the intrinsic similarity of graphs. Though successful in constructing accurate predictive and classification models for supervised learning, graph kernel functions have (i) high computational complexity and (ii) non-trivial difficulty to be indexed in a graph database. Our objective is to bridge graph kernel function and similarity search in graph databases by proposing (i) a novel kernel-based similarity measurement and (ii) an efficient indexing structure for graph data management. Our method of similarity measurement builds upon local features extracted from each node and their neighboring nodes in graphs. A hash table is utilized to support efficient storage and fast search of the extracted local features. Using the hash table, a graph kernel function is defined to capture the intrinsic similarity of graphs and for fast similarity query processing. We have implemented our method, which we have named G-hash, and have demonstrated its utility on large chemical graph databases. Our results show that the G-hash method achieves state-of-the-art performance for k-nearest neighbor (k-NN) classification. Most importantly, the new similarity measurement and the index structure is scalable to large database with smaller indexing size, faster indexing construction time, and faster query processing time as compared to state-of-the-art indexing methods such as C-tree, gIndex, and GraphGrep.
