Biblio

Vehicular Named Data Network (VNDN) uses NDN as an underlying communication paradigm to realize intelligent transportation system applications. Content communication is the essence of NDN, which is primarily carried out through content naming, forwarding, intrinsic content security, and most importantly the in-network caching. In vehicular networks, vehicles on the road communicate with other vehicles and/or infrastructure network elements to provide passengers a reliable, efficient, and infotainment-rich commute experience. Recently, different aspects of NDN have been investigated in vehicular networks and in vehicular social networks (VSN); however, in this paper, we investigate the in-network caching, realized in NDN through the content store (CS) data structure. As the stale contents in CS do not just occupy cache space, but also decrease the overall performance of NDN-driven VANET and VSN applications, therefore the size of CS and the content lifetime in CS are primary issues in VNDN communications. To solve these issues, we propose a simple yet efficient multi-metric CS management mechanism through cache replacement (M2CRP). We consider the content popularity, relevance, freshness, and distance of a node to devise a set of algorithms for selection of the content to be replaced in CS in the case of replacement requirement. Simulation results show that our multi-metric strategy outperforms the existing cache replacement mechanisms in terms of Hit Ratio.

We propose a novel application of coded computing to the problem of the nearest neighbor estimation using MatDot Codes (Fahim et al., Allerton'17) that are known to be optimal for matrix multiplication in terms of recovery threshold under storage constraints. In approximate nearest neighbor algorithms, it is common to construct efficient in-memory indexes to improve query response time. One such strategy is Multiple Random Projection Trees (MRPT), which reduces the set of candidate points over which Euclidean distance calculations are performed. However, this may result in a high memory footprint and possibly paging penalties for large or high-dimensional data. Here we propose two techniques to parallelize MRPT that exploit data and model parallelism respectively by dividing both the data storage and the computation efforts among different nodes in a distributed computing cluster. This is especially critical when a single compute node cannot hold the complete dataset in memory. We also propose a novel coded computation strategy based on MatDot codes for the model-parallel architecture that, in a straggler-prone environment, achieves the storage-optimal recovery threshold, i.e., the number of nodes that are required to serve a query. We experimentally demonstrate that, in the absence of straggling, our distributed approaches require less query time than execution on a single processing node, providing near-linear speedups with respect to the number of worker nodes. Our experiments on real systems with simulated straggling, we also show that in a straggler-prone environment, our strategy achieves a faster query execution than the uncoded strategy.

Face recognition is a fast-expanding field of research. Countless classification algorithms have found use in face recognition, with more still being developed, searching for better performance and accuracy. For high-dimensional data such as images, the K-Nearest-Neighbours classifier is a tempting choice. However, it is very computationally-intensive, as it has to perform calculations on all items in the stored dataset for each classification it makes. Fortunately, there is a way to speed up the process by performing some of the calculations in parallel. We propose a parallel CUDA implementation of the KNN classifier and then compare it to a serial implementation to demonstrate its performance superiority.

Nearest-Neighbor Search (NNS) arises as a key component of sampling-based motion planning algorithms and it is known as their asymptotic computational bottleneck. Algorithms for exact Nearest-Neighbor Search rely on explicit distance comparisons to different extents. However, in motion planning, evaluating distances is generally a computationally demanding task, since the metric is induced by the minimum cost of steering a dynamical system between states. In the presence of driftless nonholonomic constraints, we propose efficient pruning techniques for the k-d tree algorithm that drastically reduce the number of distance evaluations performed during a query. These techniques exploit computationally convenient lower and upper bounds to the geodesic distance of the corresponding sub-Riemannian geometry. Based on asymptotic properties of the reachable sets, we show that the proposed pruning techniques are optimal, modulo a constant factor, and we provide experimental results with the Reeds-Shepp vehicle model.

K-nearest neighbor (kNN) search has wide applications in many areas, including data mining, machine learning, statistics and many applied domains. Inspired by the success of ensemble methods and the flexibility of tree-based methodology, we propose random projection forests, rpForests, for kNN search. rpForests finds kNNs by aggregating results from an ensemble of random projection trees with each constructed recursively through a series of carefully chosen random projections. rpForests achieves a remarkable accuracy in terms of fast decay in the missing rate of kNNs and that of discrepancy in the kNN distances. rpForests has a very low computational complexity. The ensemble nature of rpForests makes it easily run in parallel on multicore or clustered computers; the running time is expected to be nearly inversely proportional to the number of cores or machines. We give theoretical insights by showing the exponential decay of the probability that neighboring points would be separated by ensemble random projection trees when the ensemble size increases. Our theory can be used to refine the choice of random projections in the growth of trees, and experiments show that the effect is remarkable.

SPARQL is a standard query language for knowledge graphs (KGs). However, it is hard to find correct answer if KGs are incomplete or incorrect. Knowledge graph embedding (KGE) enables answering queries on such KGs by inferring unknown knowledge and removing incorrect knowledge. Hence, our long-term goal in this line of research is to propose a new framework that integrates KGE and SPARQL, which opens various research problems to be addressed. In this paper, we solve one of the most critical problems, that is, optimizing the performance of nearest neighbor (NN) search. In our evaluations, we demonstrate that the search time of state-of-the-art NN search algorithms is improved by 40% without sacrificing answer accuracy.

Nowadays, with such a huge amount of information available online, one key challenge is how to retrieve target data efficiently. A recent state-of-art solution, k-means hashing (KMH), codes data via a string of binary code obtained by iterative k-means clustering and binary code optimizing. To deal with high dimensional data, KMH divides the space into low-dimensional subspaces, places a hypercube in each subspace and finds its proper location by the mentioned optimizing process. However, the complexity of the optimization increases rapidly when the dimension of the hypercube increases. To address this issue, we propose an improved hashing method stacked k-means hashing (SKMH). The main idea is to increase the approximation by a coarse-to-fine multi-layer lower-dimensional cubes. With these kinds of lower-dimensional cubes, SKMH can achieve a similar approximation ability via a less optimizing time, compared with KMH method using higher-dimensional cubes. Extensive experiments have been conducted on two public databases, demonstrating the performance of our method by some common metrics in fast nearest neighbor search.

This paper proposes an FPGA-based nearest neighbor search engine for high-dimensional data, in which nearest neighbor search is performed based on distance-based hashing. The proposed hardware search engine implements a nearest neighbor search algorithm based on an extension of flexible distance-based hashing (FDH, for short), which finds an exact solution with high probability. The proposed engine is a parallel processing and pipelined circuit so that search results can be obtained in a short execution time. Experimental results show the effectiveness and efficiency of the proposed engine.

Nearest neighbor search algorithm plays a very important role in computer image algorithm. When the search data is large, we need to use fast search algorithm. The current fast retrieval algorithms are tree based algorithms. The efficiency of the tree algorithm decreases sharply with the increase of the data dimension. In this paper, a local integral hash nearest neighbor algorithm of the spatial space is proposed to construct the tree structure by changing the way of the node of the access tree. It is able to express data distribution characteristics. After experimental testing, this paper achieves more efficient performance in high dimensional data.

Image registration is the process of aligning images of the same scene taken at different instances, from different viewpoints or by heterogeneous sensors. This can be achieved either by area based or by feature based image matching techniques. Feature based image registration focuses on detecting relevant features from the input images and attaching descriptors to these features. Matching visual descriptions of two images is a major task in image registration. This feature matching is currently done using Exhaustive Search (or Brute-Force) and Nearest Neighbour Search. The traditional method used for nearest neighbour search is by representing the data as k-d trees. This nearest neighbour search can also be performed using combinatorial optimization algorithms such as Simulated Annealing. This work proposes a method to perform image feature matching by nearest neighbour search done based on Threshold Accepting, a faster version of Simulated Annealing.The experiments performed suggest that the proposed algorithm can produce better results within a minimum number of iterations than many existing algorithms.

Cloud computing has become an important and popular infrastructure for data storage and sharing. Typically, data owners outsource their massive data to a public cloud that will provide search services to authorized data users. With privacy concerns, the valuable outsourced data cannot be exposed directly, and should be encrypted before outsourcing to the public cloud. In this paper, we focus on k-Nearest Neighbor (k-NN) search over encrypted data. We propose efficient and secure k-NN search schemes based on matrix similarity to achieve efficient and secure query services in public cloud. In our basic scheme, we construct the traces of two diagonal multiplication matrices to denote the Euclidean distance of two data points, and perform secure k-NN search by comparing traces of corresponding similar matrices. In our enhanced scheme, we strengthen the security property by decomposing matrices based on our basic scheme. Security analysis shows that our schemes protect the data privacy and query privacy under attacking with different levels of background knowledge. Experimental evaluations show that both schemes are efficient in terms of computation complexity as well as computational cost.

A plethora of recent work has shown that convolutional networks are not robust to adversarial images: images that are created by perturbing a sample from the data distribution as to maximize the loss on the perturbed example. In this work, we hypothesize that adversarial perturbations move the image away from the image manifold in the sense that there exists no physical process that could have produced the adversarial image. This hypothesis suggests that a successful defense mechanism against adversarial images should aim to project the images back onto the image manifold. We study such defense mechanisms, which approximate the projection onto the unknown image manifold by a nearest-neighbor search against a web-scale image database containing tens of billions of images. Empirical evaluations of this defense strategy on ImageNet suggest that it very effective in attack settings in which the adversary does not have access to the image database. We also propose two novel attack methods to break nearest-neighbor defense settings and show conditions under which nearest-neighbor defense fails. We perform a series of ablation experiments, which suggest that there is a trade-off between robustness and accuracy between as we use features from deeper in the network, that a large index size (hundreds of millions) is crucial to get good performance, and that careful construction of database is crucial for robustness against nearest-neighbor attacks.

Cross-modal hashing, which searches nearest neighbors across different modalities in the Hamming space, has become a popular technique to overcome the storage and computation barrier in multimedia retrieval recently. Although dozens of cross-modal hashing algorithms are proposed to yield compact binary code representation, applying exhaustive search in a large-scale dataset is impractical for the real-time purpose, and the Hamming distance computation suffers inaccurate results. In this paper, we propose a novel index scheme over binary hash codes in cross-modal retrieval. The proposed indexing scheme exploits a few binary bits of the hash code as the index code. Based on the index code representation, we construct an inverted index structure to accelerate the retrieval efficiency and train a neural network to improve the indexing accuracy. Experiments are performed on two benchmark datasets for retrieval across image and text modalities, where hash codes are generated by three cross-modal hashing methods. Results show the proposed method effectively boosts the performance over the benchmark datasets and hash methods.

Product quantization (PQ) and its variations are popular and attractive in approximate nearest neighbor search (ANN) due to their lower memory usage and faster retrieval speed. PQ decomposes the high-dimensional vector space into several low-dimensional subspaces, and quantizes each sub-vector in their subspaces, separately. Thus, PQ can generate a codebook containing an exponential number of codewords or indices by a Cartesian product of the sub-codebooks from different subspaces. However, when there is large variance in the average amplitude of the components of the data points, directly utilizing the PQ on the data points would result in poor performance. In this paper, we propose a new approach, namely, mean-removed product quantization (MRPQ) to address this issue. In fact, the average amplitude of a data point or the mean of a date point can be regarded as statistically independent of the variation of the vector, that is, of the way the components vary about this average. Then we can learn a separate scalar quantizer of the means of the data points and apply the PQ to their residual vectors. As shown in our comprehensive experiments on four large-scale public datasets, our approach can achieve substantial improvements in terms of Recall and MAP over some known methods. Moreover, our approach is general which can be combined with PQ and its variations.

This paper introduces a biologically inspired associative data model and structure for finding nearest neighbors and similar patterns. The method can be used as an alternative to the classical approaches to accelerate the search for such patterns using various priorities for attributes according to the Sebestyen measure. The presented structure, together with algorithms developed in this paper can be useful in various computational intelligence tasks like pattern matching, recognition, clustering, classification, multi-criterion search etc. This approach is particularly useful for the on-line operation of associative neural network graphs. Graphs that dynamically develop their structure during learning on training data. The results of experiments show that the associative approach can substantially accelerate the nearest neighbor search and that associative structures can also be used as a model for KNN tasks. Finally, this paper presents how the associative structures can be used to self-organize data and represent knowledge about them in the associative way, which yields new search approaches described in this paper.

Continuous growth in sensor data and other temporal sequence data necessitates efficient retrieval and similarity search support on these big time series datasets. However, finding exact similarity results, especially at the granularity of subsequences, is known to be prohibitively costly for large data sets. In this paper, we thus propose an efficient framework for solving this exact subsequence similarity match problem, called TINN (TIme series Nearest Neighbor search). Exploiting the range interval diversity properties of time series datasets, TINN captures similarity at two levels of abstraction, namely, relationships among subsequences within each long time series and relationships across distinct time series in the data set. These relationships are compactly organized in an augmented relationship graph model, with the former relationships encoded in similarity vectors at TINN nodes and the later captured by augmented edge types in the TINN Graph. Query processing strategy deploy novel pruning techniques on the TINN Graph, including node skipping, vertical and horizontal pruning, to significantly reduce the number of time series as well as subsequences to be explored. Comprehensive experiments on synthetic and real world time series data demonstrate that our TINN model consistently outperforms state-of-the-art approaches while still guaranteeing to retrieve exact matches.

A fundamental recurring task in many machine learning applications is the search for the Nearest Neighbor in high dimensional metric spaces. Towards answering queries in large scale problems, state-of-the-art methods employ Approximate Nearest Neighbors (ANN) search, a search that returns the nearest neighbor with high probability, as well as techniques that compress the dataset. Product-Quantization (PQ) based ANN search methods have demonstrated state-of-the-art performance in several problems, including classification, regression and information retrieval. The dataset is encoded into a Cartesian product of multiple low-dimensional codebooks, enabling faster search and higher compression. Being intrinsically parallel, PQ-based ANN search approaches are amendable for hardware acceleration. This paper proposes a novel Hierarchical PQ (HPQ) based ANN search method as well as an FPGA-tailored architecture for its implementation that outperforms current state of the art systems. HPQ gradually refines the search space, reducing the number of data compares and enabling a pipelined search. The mapping of the architecture on a Stratix 10 FPGA device demonstrates over ×250 speedups over current state-of-the-art systems, opening the space for addressing larger datasets and/or improving the query times of current systems.

A main issue in approximate nearest neighbor search is to achieve an excellent tradeoff between search accuracy and computation cost. In this paper, we address this issue by leveraging k-nearest neighbor graph and hill-climbing to accelerate vector quantization in the query assignment process. A modified hill-climbing algorithm is proposed to traverse k-nearest neighbor graph to find closest centroids for a query, rather than calculating the query distances to all centroids. Instead of using random seeds in the original hill-climbing algorithm, we generate high-quality seeds based on the hashing technique. It can boost the query assignment efficiency due to a better start-up in hill-climbing. We evaluate the experiment on the benchmarks of SIFT1M and GIST1M datasets, and show the proposed hashing-based seed generation effectively improves the search performance.

Fault detection in a machine at earlier stage can prevent severe damage and loss to the industries. Fault detection techniques are broadly classified into three categories; signature extraction-based, model-based and knowledge-based approach. Model-based techniques are efficient for raising an alarm signal if there is any fault in the machine. This paper focuses on one such model based-technique to identify the internal faults of induction machine. The model developed is deployed in the end to make it feasible to use in real time. K-Nearest Neighbors (KNN) and grid search cross validation (CV) have been used to train and optimize the model to give the best results. The advantage of proposed algorithm is the accuracy in prediction which has been seen to be 80%. Finally, a user friendly interface has been built using Flask, a python web framework.

Large graphs are increasingly prevalent in mobile networks, social networks, traffic networks and biological networks. These graphs are often uncertain, where edges are augmented with probabilities that indicates the chance to exist. Recently k-nearest neighbor search has been studied within the field of uncertain graphs, but the scalability and efficiency issues are not well solved. Moreover, solutions are implemented on a single machine and thus cannot fit large uncertain graphs. In this paper, we develop a framework, called DURS, to distribute k-nearest neighbor search into several machines and re-partition the uncertain graphs to balance the work loads and reduce the communication costs. Evaluation results show that DURS is essential to make the system scalable when answering k-nearest neighbor queries on uncertain graphs.

Natural language processing (NLP) have been recently used to extract clinical information from free text in Electronic Health Record (EHR). In clinical NLP one challenge is that the meaning of clinical entities is heavily affected by assertion modifiers such as negation, uncertain, hypothetical, experiencer and so on. Incorrect assertion assignment could cause inaccurate diagnosis of patients' condition or negatively influence following study like disease modeling. Thus, clinical NLP systems which can detect assertion status of given target medical findings (e.g. disease, symptom) in clinical context are highly demanded. Here in this work, we propose a deep-learning system based on word embedding, RNN and attention mechanism (more specifically: Attention-based Bidirectional Long Short-Term Memory networks) for assertion detection in clinical notes. Unlike previous state-of-art methods which require knowledge input or feature engineering, our system is a knowledge poor machine learning system and can be easily extended or transferred to other domains. The evaluation of our system on public benchmarking corpora demonstrates that a knowledge poor deep-learning system can also achieve high performance for detecting negation and assertions comparing to state-of-the-art systems.

The need for performing deep neural network inferences on resource-constrained embedded devices (e.g., Internet of Things nodes) requires specialized architectures to achieve the best trade-off among performance, energy, and cost. One of the most promising architectures in this context is based on massive parallel and specialized cores interconnected by means of a Network-on-Chip (NoC). In this paper, we extensively evaluate NoC-based deep neural network accelerators by exploring the design space spanned by several architectural parameters including, network size, routing algorithm, local memory size, link width, and number of memory interfaces. We show how latency is mainly dominated by the on-chip communication whereas energy consumption is mainly accounted by memory (both on-chip and off-chip). The outcome of the analysis, thus, pushes toward a research line devoted to the optimization of the on-chip communication fabric and the memory subsystem for performance improvement and energy efficiency, respectively.

The Department of Defense (DoD) manages capabilities through the Joint Interoperability and Capability Development System (JCIDS) process. As part of this process, sponsors develop a series of DoD Architecture Framework (DoDAF) products to assist analysts understand the proposed capability and how it fits into the broader network of DoD legacy systems and systems under development. However, the Joint Staff, responsible for executing the JCIDS process, often analyzes these architectures in isolation without considering the broader network of systems. DoD leadership, the Government Accountability Organization, and others have noted the lack of the DoD's ability to manage the broader portfolio of capabilities in various reports and papers. Several efforts have proposed merging DoDAF architecture into a larger meta-architecture based on individual system architectures. This paper specifically targets the Systems View 3 (SV-3), System-to-system matrix, as an opportunity to merge multiple DoDAF architecture views into a network of system and understand the potential benefits associated with analyzing a broader perspective. The goal of merging multiple SV-3s is to better understand the interoperability of a system within the network of DoD systems as network metrics may provide insights into the relative interoperability of a DoD system. Currently, the DoD's definition of interoperability focuses on the system or capability's ability to enter and operate within the DoD Information Network (DoDIN); however, this view limits the definition of interoperability as it focuses solely on information flows and not resource flows or physical connections that should be present in a SV-3. The paper demonstrates the importance of including all forms of connections between systems in a network by comparing network metrics associated with the different types of connections. Without a complete set of DoDAF architectures for each system within the DoD and based on the potential classification of these products, the paper collates data that should be included in an SV-3 from open source, unclassified references to build the overall network of DoD systems. From these sources, a network of over 300 systems with almost 1000 connections emerges based on the documented information, resource, and physical connections between these legacy and planned DoD systems. With this network, the paper explores the quantification of individual system's interoperability through the application of nodal and network metrics from social network analysis (SNA). A SNA perspective on a network of systems provides additional insights beyond traditional network analysis because of the emphasis on the importance of nodes, systems, in the network as well as the relationship, connections, between the nodes. Finally, the paper proposes future work to explore the quantification of additional attributes of systems as well as a method for further validating the findings.

Major blackouts are due to cascading failures in power systems. These failures usually occur at vulnerable links of the network. To identify these, indicators have already been defined using complex network theory. However, most of these indicators only depend on the topology of the grid; they fail to detect the weak links. We introduce a new metric to identify the vulnerable lines, based on the load-flow equations and the grid geometry. Contrary to the topological indicators, ours is built from the electrical equations and considers the location and magnitude of the loads and of the power generators. We apply this new metric to the IEEE 118-bus system and compare its prediction of weak links to the ones given by an industrial software. The agreement is very well and shows that using our indicator a simple examination of the network and its generator and load distribution suffices to find the weak lines.

Physical Unclonable Functions (PUFs) are vulnerable to various modelling attacks. The chaotic behaviour of oscillating systems can be leveraged to improve their security against these attacks. We have integrated an Arbiter PUF implemented on a FPGA with Chua's oscillator circuit to obtain robust final responses. These responses are tested against conventional Machine Learning and Deep Learning attacks for verifying security of the design. It has been found that such a design is robust with prediction accuracy of nearly 50%. Moreover, the quality of the PUF architecture is evaluated for uniformity and uniqueness metrics and Monte Carlo analysis at varying temperatures is performed for determining reliability.