• A submitted manuscript is the author's version of the article upon submission and before peer-review. There can be important differences between the submitted version and the official published version of record. People interested in the research are advised to contact the author for the final version of the publication, or visit the DOI to the publisher's website. • The final author version and the galley proof are versions of the publication after peer review. • The final published version features the final layout of the paper including the volume, issue and page numbers.

s of Keynotes Interactive Exploration using Hypergraphs

of Doctoral Disseration Academic Year 2016 A Multicontext-adaptive Query Creation and Search System for Large-scale Image and Video Data by Nguyen Thi Ngoc Diep

k-nearest neighbor graph is a fundamental data structure in many disciplines such as information retrieval, data-mining, pattern recognition, and machine learning, etc. In the literature, considerable research has been focusing on how to efficiently build an approximate k-nearest neighbor graph (k-NN graph) for a fixed dataset. Unfortunately, a closely related issue of how to merge two existing k-NN graphs has been overlooked. In this paper, we address the issue of k-NN graph merging in two different scenarios. In the first scenario, a symmetric merge algorithm is proposed to combine two approximate k-NN graphs. The algorithm facilitates large-scale processing by efficient merging of k-NN graphs that are produced in parallel. In the second scenario, a joint merge algorithm is proposed to expand an existing k-NN graph with a raw dataset. The algorithm enables incremental construction of a hierarchical approximate k-NN graph. Superior performance is attained when leveraging the hierarchy for NN search of various data types, dimensionality and distance measures.

k-Nearest Neighbors (kNN) algorithm is a method to find the closest points in a dataset to a query point. The result of kNN can be used for classification and regression, both of which are commonly used in data mining and machine learning. In this paper, Enhanced Vote Count (EVC) circuit, which uses hardware to compare the quantized projected values of query and training/reference vectors instead of the vectors themselves, is considered to approximate the kNN search to provide a low complexity search solution. To improve the performance of EVC with limited projection number because projection number is directly related to implementation cost of EVC circuit, the concept of weak bit is considered and only reliable binary pattern matching is evaluated. The implementation of weak bit based on EVC circuit is also described. Simulation results show that, the performance of EVC can be significantly improved with weak bit implementation under limited projection number.

his paper, provide survey of the proposed namespace management schemes for file system. Namespace management can be used to reduce exhaustive search over all directories. Namespace using semantic correlation can also increase search ability. File system namespace as an information organizing infrastructure is a help to improve system's quality of service such as performance, scalability, and ease of use. This paper discusses the improvement to be made for future proposed namespace schemes. This paper provides reader with the basis for research in namespace schemes for file system.

\kdtree \citefriedman1976algorithm has long been deemed unsuitable for exact nearest-neighbor search in high dimensional data. The theoretical guarantees and the empirical performance of \kdtree do not show significant improvements over brute-force nearest-neighbor search in moderate to high dimensions. \kdtree has been used relatively more successfully for approximate search \citemuja2009flann but lack theoretical guarantees. In the article, we build upon randomized-partition trees \citedasgupta2013randomized to propose \kdtree based approximate search schemes with $O(d łog d + łog n)$ query time for data sets with n points in d dimensions and rigorous theoretical guarantees on the search accuracy. We empirically validate the search accuracy and the query time guarantees of our proposed schemes, demonstrating the significantly improved scaling for same level of accuracy.

[See the paper for the full abstract.] We show tight upper and lower bounds for time-space trade-offs for the $c$-Approximate Near Neighbor Search problem. For the $d$-dimensional Euclidean space and $n$-point datasets, we develop a data structure with space $n^{1 + \rho_u + o(1)} + O(dn)$ and query time $n^{\rho_q + o(1)} + d n^{o(1)}$ for every $\rho_u, \rho_q \geq 0$ such that: \begin{equation} c^2 \sqrt{\rho_q} + (c^2 - 1) \sqrt{\rho_u} = \sqrt{2c^2 - 1}. \end{equation} This is the first data structure that achieves sublinear query time and near-linear space for every approximation factor $c > 1$, improving upon [Kapralov, PODS 2015]. The data structure is a culmination of a long line of work on the problem for all space regimes; it builds on Spherical Locality-Sensitive Filtering [Becker, Ducas, Gama, Laarhoven, SODA 2016] and data-dependent hashing [Andoni, Indyk, Nguyen, Razenshteyn, SODA 2014] [Andoni, Razenshteyn, STOC 2015]. Our matching lower bounds are of two types: conditional and unconditional. First, we prove tightness of the whole above trade-off in a restricted model of computation, which captures all known hashing-based approaches. We then show unconditional cell-probe lower bounds for one and two probes that match the above trade-off for $\rho_q = 0$, improving upon the best known lower bounds from [Panigrahy, Talwar, Wieder, FOCS 2010]. In particular, this is the first space lower bound (for any static data structure) for two probes which is not polynomially smaller than the one-probe bound. To show the result for two probes, we establish and exploit a connection to locally-decodable codes.

Within the IoT-cloud, security has a very significant role to play. One of the best means by which the security and privacy of an image may be safeguarded confidentially is through encryption. However, this methodological process engenders a disadvantage in that it is difficult to search through encrypted images. A number of different means by which encrypted image can be searched have been devised, however, certain security solutions may not be used for smart devices within an IoTcloud due to the fact that such solutions are not lightweight. We present a lightweight scheme that is able to provide a contentbased search through images that have been encrypted. More specifically, images are represented using local features. A similar methodology further described in [1] is also used for image similarity discrimination. In addition, we use a hashing method concerning a locality sensitive hash (LSH) so that the searchable index can be devised. The use of the LSH index means that the proficiency and effectiveness of the system is increased, which allows the retrieval of only relevant images with a minimum number of distance evaluations. Refining vector techniques are used to refine relevant results efficiently and securely. Our index construction process ensures that stored data and trapdoors are kept private.

With vast amount of content online, it is not surprising that unscrupulous entities "borrow" from the web to provide content for advertisements, link farms, and spam. Our insight is that cryptographic hashing and fingerprinting can efficiently identify content reuse for web-size corpora. We develop two related algorithms, one to automatically *discover* previously unknown duplicate content in the web, and the second to *precisely detect* copies of discovered or manually identified content. We show that *bad neighborhoods*, clusters of pages where copied content is frequent, help identify copying in the web. We verify our algorithm and its choices with controlled experiments over three web datasets: Common Crawl (2009/10), GeoCities (1990s–2000s), and a phishing corpus (2014). We show that our use of cryptographic hashing is much more precise than alternatives such as locality-sensitive hashing, avoiding the thousands of false-positives that would otherwise occur. We apply our approach in three systems: discovering and detecting duplicated content in the web, searching explicitly for copies of Wikipedia in the web, and detecting phishing sites in a web browser. We show that general copying in the web is often benign (for example, templates), but 6–11% are commercial or possibly commercial. Most copies of Wikipedia (86%) are commercialized (link farming or advertisements). For phishing, we focus on PayPal, detecting 59% of PayPal-phish even without taking on intentional cloaking.

With the wide application of biometrics technology, the scale of biometrics databases is increasing rapidly. In this situation, fast retrieval technology is more and more necessary for large-scale biometrics retrieval and recognition. Palmprint recognition is one of the emerging biometrics technologies. However, the research on fast palmprint retrieval algorithm is still preliminary. Hashing is one of the most popular image retrieval technologies due to its fast speed and low storage cost. In this paper, we propose a new deep palmprint hashing method, which integrates classification loss, pairing loss and quantization loss in a unified deep learning framework. Experimental results show that the proposed deep multi-loss hashing method has better performance for palmprint recognition and retrieval than other existing classic hashing methods.

With the volume of data grows rapidly, the cost of detecting duplication entities has increased significantly in data cleaning. However, some real-time applications only need to identify as many duplicate entities as possible in a limited time, rather than all of them. The existing works adopt the sorting method to divide similar records into blocks, and arrange the processing order of blocks to detect duplicate entity progressively. However, this method only works well when the attributes of records are suitable for sorting. Therefore, this paper proposes a novel progressive de-duplicate method for records that can't be sorted by their attributes. The method distributes records into different blocks based on their features and generates a modified bloom filter index for each block. Then it uses the bloom filter to predict the probability of duplicate entities in this block, which determines the processing order of blocks to detect the duplicate entities more quickly. The comprehensive experiment shows that the number of duplicate detection by this algorithm in the finite time is far more efficient than other algorithms involved in the related works.

With the vast amount of accessible, online content, it is not surprising that unscrupulous entities “borrow” from the web to provide filler for advertisements, link farms, and spam and make a quick profit. Our insight is that cryptographic hashing and fingerprinting can efficiently identify content reuse for web-size corpora. We develop two related algorithms, one to automatically discover previously unknown duplicate content in the web, and the second to detect copies of discovered or manually identified content in the web. Our detection can also bad neighborhoods, clusters of pages where copied content is frequent. We verify our approach with controlled experiments with two large datasets: a Common Crawl subset the web, and a copy of Geocities, an older set of user-provided web content. We then demonstrate that we can discover otherwise unknown examples of duplication for spam, and detect both discovered and expert-identified content in these large datasets. Utilizing an original copy of Wikipedia as identified content, we find 40 sites that reuse this content, 86% for commercial benefit.

With the recent trend of promoting Field-Programmable Gate Arrays (FPGAs) to first-class citizens in accelerating compute-intensive applications in networking, cloud services and artificial intelligence, FPGAs face two major challenges in sustaining competitive advantages in performance and energy efficiency for diverse cloud workloads: 1) limited configuration capability for supporting light-weight computations/on-chip data storage to accelerate emerging search-/data-intensive applications. 2) lack of architectural support to hide reconfiguration overhead for assisting virtualization in a cloud computing environment. In this paper, we propose a reconfigurable memory-oriented computing fabric, namely Liquid Silicon-Monona (L-Si), enabled by emerging nonvolatile memory technology i.e. RRAM, to address these two challenges. Specifically, L-Si addresses the first challenge by virtue of a new architecture comprising a 2D array of physically identical but functionally-configurable building blocks. It, for the first time, extends the configuration capabilities of existing FPGAs from computation to the whole spectrum ranging from computation to data storage. It allows users to better customize hardware by flexibly partitioning hardware resources between computation and memory, greatly benefiting emerging search- and data-intensive applications. To address the second challenge, L-Si provides scalable multi-context architectural support to minimize reconfiguration overhead for assisting virtualization. In addition, we provide compiler support to facilitate the programming of applications written in high-level programming languages (e.g. OpenCL) and frameworks (e.g. TensorFlow, MapReduce) while fully exploiting the unique architectural capability of L-Si. Our evaluation results show L-Si achieves 99.6% area reduction, 1.43× throughput improvement and 94.0% power reduction on search-intensive benchmarks, as compared with the FPGA baseline. For neural network benchmarks, on average, L-Si achieves 52.3× speedup, 113.9× energy reduction and 81% area reduction over the FPGA baseline. In addition, the multi-context architecture of L-Si reduces the context switching time to - 10ns, compared with an off-the-shelf FPGA (∼100ms), greatly facilitating virtualization.

With the rapid increasing popularity of web video, video copy detection (VCD) has become a hot research field in content-based video retrieval. Many efforts have been carried out to improve the effectiveness and efficiency of copy detection. In order to solve two challenging problems: video feature extraction and fast search in large database, a novel efficient video copy detection approach was proposed in this paper. A compact video signature was computed based on spatial-temporal features of video sequences. The video similarity is measured by the distance of video signature. For the scalable computing requirement, a new search method via clustering index table was presented by index clustering. The experimental results from query tests in large database show this method is efficient with high recall and precision rate.

With the rapid growth of mobile devices and applications, geo-tagged data has become a major workload for big data storage systems. In order to achieve scalability, existing solutions build an additional index layer above general purpose distributed data stores. Fulfilling the semantic level need, this approach, however, leaves a lot to be desired for execution efficiency, especially when users query for moving objects within a high resolution geometric area, which we call geometry queries. Such geometry queries translate to a much larger set of range scans, forcing the backend to handle orders of magnitude more requests. Moreover, spatial-temporal applications naturally create dynamic workload hotspots, which pushes beyond the design scope of existing solutions. This paper presents Pyro, a spatial-temporal bigdata storage system tailored for high resolution geometry queries and dynamic hotspots. Pyro understands geometries internally, which allows range scans of a geometry query to be aggregately optimized. Moreover, Pyro employs a novel replica placement policy in the DFS layer that allows Pyro to split a region without losing data locality benefits. Our evaluations use NYC taxi trace data and an 80-server cluster. Results show that Pyro reduces the response time by 60× on 1km×1km rectangle geometries compared to the state-of-the-art solutions. Pyro further achieves 10× throughput improvement on 100m×100m rectangle geometries.

With the rapid development of web video application, video similarity search has become a hot research field in content-based video retrieval. Many efforts have been carried out to improve the effectiveness and efficiency of similarity search in large database. In order to solve two challenging problems: similarity measurement and search method, a novel efficient VSS approach is proposed in this paper. A compact video signature was computed based on Spatial-temporal features of video sequences. The video similarity was measured by an efficient computation of the distance of video signature in feature space. For the scalable computing requirement, a new search method via Signature Index Table was presented by index clustering. The experimental results from the query tests in large database show this method is highly efficient for similar video search.

With the rapid development of the Internet, nowadays tremendous amounts of data including images and videos, up to millions or billions, can be collected for training machine learning models. Inspired by this trend, this thesis is dedicated to developing large-scale machine learning techniques for the purpose of making classification and nearest neighbor search practical on gigantic databases. 1. Large Graph Construction: We present a novel graph construction approach, called Anchor Graphs, which enjoys linear space and time complexities and can thus be constructed over gigantic databases efficiently. The central idea of the Anchor Graph is introducing a few anchor points and converting intensive data-to-data affinity computation to drastically reduced data-to-anchor affinity computation. A low-rank data-to-data affinity matrix is derived using the data-to-anchor affinity matrix. We also theoretically prove that the Anchor Graph lends itself to an intuitive probabilistic interpretation by showing that each entry of the derived affinity matrix can be considered as a transition probability between two data points through Markov random walks. 2. Large-Scale Semi-Supervised Learning: We employ Anchor Graphs to develop a scalable solution for semi-supervised learning, which capitalizes on both labeled and unlabeled data to learn graph-based classification models. We propose several key methods to build scalable semi-supervised kernel machines such that real-world linearly inseparable data can be tackled. The proposed techniques take advantage of the Anchor Graph from a kernel point of view, generating a set of low-rank kernels which are made to encompass the neighborhood structure unveiled by the Anchor Graph. By linearizing these low-rank kernels, training nonlinear kernel machines in semi-supervised settings can be simplified to training linear SVMs in supervised settings, so the computational cost for classifier training is substantially reduced. We accomplish excellent classification performance by applying the proposed semi-supervised kernel machine - a linear SVM with a linearized Anchor Graph warped kernel. 3. Unsupervised Hashing: We present a novel unsupervised hashing approach based on the Anchor Graph which captures the underlying manifold structure. The Anchor Graph Hashing approach allows constant time hashing of a new data point by extrapolating graph Laplacian eigenvectors to eigenfunctions. Furthermore, a hierarchical threshold learning procedure is devised to produce multiple hash bits for each eigenfunction, thus leading to higher search accuracy. 4. Supervised Hashing: We present a novel kernel-based supervised hashing model which requires a limited amount of supervised information in the form of similar and dissimilar data pairs, and is able to achieve high hashing quality at a practically feasible training cost. The idea is to map the data to compact binary codes whose Hamming distances are simultaneously minimized on similar pairs and maximized on dissimilar pairs. Our approach is distinct from prior work in utilizing the equivalence between optimizing the code inner products and the Hamming distances. This enables us to sequentially and efficiently train the hash functions one bit at a time, yielding very short yet discriminative codes. The presented supervised hashing approach is general, allowing search of both semantically similar neighbors and metric distance neighbors. 5. Hyperplane Hashing: We present a novel hyperplane hashing technique which yields high search accuracy with compact hash codes. The key idea is a novel bilinear form used in designing the hash functions, leading to a higher collision probability than all of the existing hyperplane hash functions when using random projections. To further increase the performance, we develop a learning based framework in which the bilinear functions are directlylearned from the input data. This results in compact yet discriminative codes, as demonstrated by the superior search performance over all random projection based solutions. (Abstract shortened by UMI.)

With the rapid development of the Internet and multimedia technologies over the last decade, a huge amount of data has become available, from text corpus, to collections of online images and videos. Cheap storage cost and modern database technologies have made it possible to accumulate large-scale datasets. However, the ever-growing sizes of the datasets make it harder to search useful information from such data. A fundamental computational primitive for dealing with massive multimedia datasets is the similarity search problem. Multimedia similarity search aims to preprocess a database so that given a query object, one can quickly find its similar objects in the database. Searching similar objects from a large dataset in high-dimensional spaces is at the heart of many multimedia applications, such as near-duplicate retrieval, multimedia tagging, recommendation, and so on. Driven by its significance, lots of efforts have been made on this topic. The goal of my research is to design efficient hashing methods for large-scale multimedia search. In this paper, we first present the general framework for multimedia similarity search and discuss the latest improvements and progresses in the field. Then we describe the contributions we have made to effectively and efficiently search similar multimedia objects from large-scale databases. Finally, we discuss the future work and draw a conclusion.

With the prevalence of the commodity depth cameras, the new paradigm of user interfaces based on 3D motion capturing and recognition have dramatically changed the way of interactions between human and computers. Human action recognition, as one of the key components in these devices, plays an important role to guarantee the quality of user experience. Although the model-driven methods have achieved huge success, they cannot provide a scalable solution for efficiently storing, retrieving and recognizing actions in the large-scale applications. These models are also vulnerable to the temporal translation and warping, as well as the variations in motion scales and execution rates. To address these challenges, we propose to treat the 3D human action recognition as a video-level hashing problem and propose a novel First-Take-All (FTA) Hashing algorithm capable of hashing the entire video into hash codes of fixed length. We demonstrate that this FTA algorithm produces a compact representation of the video invariant to the above mentioned variations, through which action recognition can be solved by an efficient nearest neighbor search by the Hamming distance between the FTA hash codes. Experiments on the public 3D human action datasets shows that the FTA algorithm can reach a recognition accuracy higher than 80%, with about 15 bits per frame considering there are 65 frames per video over the datasets.