Approximate Nearest Neighbor Research Articles

Deep reinforcement hashing with redundancy elimination for effective image retrieval

Hashing is one of the most promising techniques in approximate nearest neighbor search due to its time efficiency and low cost in memory. Recently, with the help of deep learning, deep supervised hashing can perform representation learning and compact hash code learning jointly in an end-to-end style, and obtains better retrieval accuracy compared to non-deep methods. However, most deep hashing methods are trained with a pair-wise loss or triplet loss in a mini-batch style, which makes them inefficient at data sampling and cannot preserve the global similarity information. Besides that, many existing methods generate hash codes with redundant or even harmful bits, which is a waste of space and may lower the retrieval accuracy. In this paper, we propose a novel deep reinforcement hashing model with redundancy elimination called Deep Reinforcement De-Redundancy Hashing (DRDH), which can fully exploit large-scale similarity information and eliminate redundant hash bits with deep reinforcement learning. DRDH conducts hash code inference in a block-wise style, and uses Deep Q Network (DQN) to eliminate redundant bits. Very promising results have been achieved on four public datasets, i.e., CIFAR-10, NUS-WIDE, MS-COCO, and Open-Images-V4, which demonstrate that our method can generate highly compact hash codes and yield better retrieval performance than those of state-of-the-art methods.

Scalable Approximate FRNN-OWA Classification

Fuzzy rough nearest neighbor classification with ordered weighted averaging operators (FRNN-OWA) is an algorithm that classifies unseen instances according to their membership in the fuzzy upper and lower approximations of the decision classes. Previous research has shown that the use of OWA operators increases the robustness of this model. However, calculating membership in an approximation requires a nearest neighbor search. In practice, the query time complexity of exact nearest neighbor search algorithms in more than a handful of dimensions is near linear, which limits the scalability of FRNN-OWA. Therefore, we propose approximate FRNN-OWA, a modified model that calculates upper and lower approximations of decision classes using the approximate nearest neighbors returned by hierarchical navigable small worlds (HNSW), a recent approximative nearest neighbor search algorithm with logarithmic query time complexity at constant near-100% accuracy. We demonstrate that approximate FRNN-OWA is sufficiently robust to match the classification accuracy of exact FRNN-OWA while scaling much more efficiently. We test four parameter configurations of HNSW and evaluate their performance by measuring classification accuracy and construction and query times for samples of various sizes from three large datasets. We find that with two of the parameter configurations, approximate FRNN-OWA achieves near-identical accuracy to exact FRNN-OWA for most sample sizes within query times that are up to several orders of magnitude faster.

Supervised Robust Discrete Multimodal Hashing for Cross-Media Retrieval

The hashing-based approximate nearest neighbors search is able to reduce the storage cost and improve query speed. Therefore, they have attracted much attention in these years. Moreover, some hashing methods have been proposed for cross-modal retrieval tasks. However, there are still some issues that need to be further addressed. For example, some of them only construct a simple similarity matrix when learning hash functions or binary codes, which may lose some useful information. Some of them solve the hard discrete optimization problem by relaxing the binary constraints and quantizing the solution to obtain the final results, which may generate large quantization errors. To address these challenges, we present a new supervised cross-modal hashing method, named supervised robust discrete multimodal hashing (SRDMH). Specifically, it incorporates full label information into the hash functions learning to preserve the similarity in the original space. In addition, instead of relaxing the binary constraints, it is able to learn the binary codes and hash functions simultaneously. Moreover, it adopts a flexible ℓ <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2,p</sub> loss with nonlinear kernel embedding and introduces an intermediate presentation of the binary codes. In light of this, it becomes more robust and easier to solve by an iterative algorithm presented in this paper. To evaluate its performance, we conduct extensive experiments on three benchmark datasets. The results verify that SRDMH outperforms seven state-of-the-art cross-modal hashing methods. In addition, we also extend it to the classification task. Compared with other hashing methods, SRDMH also obtains better results when its binary codes are used for classification.

A graphical heuristic for reduction and partitioning of large datasets for scalable supervised training

A scalable graphical method is presented for selecting and partitioning datasets for the training phase of a classification task. For the heuristic, a clustering algorithm is required to get its computation cost in a reasonable proportion to the task itself. This step is succeeded by construction of an information graph of the underlying classification patterns using approximate nearest neighbor methods. The presented method consists of two approaches, one for reducing a given training set, and another for partitioning the selected/reduced set. The heuristic targets large datasets, since the primary goal is a significant reduction in training computation run-time without compromising prediction accuracy. Test results show that both approaches significantly speed-up the training task when compared against that of state-of-the-art shrinking heuristics available in LIBSVM. Furthermore, the approaches closely follow or even outperform in prediction accuracy. A network design is also presented for a partitioning based distributed training formulation. Added speed-up in training run-time is observed when compared to that of serial implementation of the approaches.

Efficient nearest neighbor search in high dimensional hamming space

Fast approximate nearest neighbor search has been well studied for real-valued vectors, however, the methods for binary descriptors are less developed. The paper addresses this problem by resorting to the well established techniques in Euclidean space. To this end, the binary descriptors are firstly mapped into low dimensional float vectors under the condition that the neighborhood information in the original Hamming space could be preserved in the mapped Euclidean space as much as possible. Then, KD-Tree is used to partitioning the mapped Euclidean space in order to quickly find approximate nearest neighbors for a given query point. This is identical to filter out a subset of nearest neighbor candidates in the original Hamming space due to the property of neighborhood preserving. Finally, Hamming ranking is applied to the small number of candidates to find out the approximate nearest neighbor in the original Hamming space, with only a fraction of running time compared to the bruteforce linear scan. Our experiments demonstrate that the proposed method significantly outperforms the state of the arts, obtaining improved search accuracy at various speed up factors, e.g., at least 16% improvement of search accuracy over previous methods (from 67.7% to 83.7%) when the search speed is 200 times faster than the linear scan for a one million database.

Neighborhood Pyramid Preserving Hashing

In this paper, we devote our efforts to the approximate nearest neighbour (ANN) search problem and propose a new unsupervised binary hashing method, i.e., Neighbourhood Pyramid preserving Hashing (NPH). We represent the nearest neighbours of each data point in a pyramid, and as the learning objective, we impose that the pyramid neighbourhood in each level is consistently preserved across the original Euclidean space and the transformed Hamming space. The neighbourhood is quantitatively characterized by its size, defined as the average distance from the involved nearest neighbours to the referred data point. Our approach is consistent with the distance-preserving principle of binary hashing and achieves stricter neighbourhood structure preserving over previous graph hashing algorithms. The experiments on several large-scale benchmark datasets demonstrate that NPH achieves promising performances compared with those of the existing state-of-the-art unsupervised binary hashing methods.

Fast image similarity search by distributed locality sensitive hashing

Approximate Nearest Neighbor (ANN) search approaches that use possible neighbors instead of exact neighbors are widely investigated by researchers in recent years. ANN approaches are usually applied in a centralized manner. However, in real world applications data is usually stored in a distributed manner. This situation led to the need for implementing ANN methods in a distributed way. In this study, our goal is to perform fast and accurate search on large size image datasets by using distributed environments. For this purpose, we propose an approach called as Randomized Distributed Hashing (RDH) which uses Locality Sensitive Hashing (LSH) in a distributed scheme. In this approach, we have randomly distributed data to different nodes on a cluster. After the distribution of data, in each node we have used same randomized hash function set for indexing the local data. Then at the query stage, the query sample is locally searched in different nodes. By exploiting from parallelism, the query time performance is significantly increased. We have a speed up of 8 for the query performance in the distributed scheme with 10 nodes. The level of Mean Average Precision (MAP) scores are quite high which are comparable to other methods. We have also investigated the usage of different and selected randomized hash functions in different nodes rather than using same indexing. We create selected hash functions according to their data division property before indexing. Since LSH is data independent method, we have obtained similar results with using same hash functions. We compared our experimental results with state-of-the-art methods given in a recent study. The proposed distributed scheme is promising for searching images in large datasets with multiple nodes.

Learning binary code for fast nearest subspace search

Subspace is widely used to represent objects under different viewpoints, illuminations, identities, and more. Due to the growing amount and dimensionality of visual contents, fast search in a large-scale database with high-dimensional subspaces is an important task in many applications, such as image retrieval, clustering, video retrieval, and visual recognition. This can be facilitated by approximate nearest subspace (ANS) search which requires effective subspace representation. All existing methods for this problem represent a subspace by a point in the Euclidean or the Grassmannian space before applying the approximate nearest neighbor (ANN) search. However, the efficiency of these methods is not guaranteed because the subspace representation step can be very time consuming when coping with high-dimensional data. Moreover, the subspace to point transforming process may cause subspace structural information loss which influences the search accuracy. In this paper, we present a new approach for hashing-based ANS search which can directly binarize a subspace without transforming it into a vector. The proposed method learns the binary codes for subspaces following a similarity preserving criterion, and simultaneously leverages the learned binary codes to train matrix classifiers as hash functions. Experiments on face and action recognition and video retrieval applications show that our method outperforms several state-of-the-art methods in both efficiency and accuracy. Moreover, we also compare our method with vector-based hashing methods. Results also show the superiority of our subspace matrix based search scheme.

Siamese Dilated Inception Hashing With Intra-Group Correlation Enhancement for Image Retrieval

For large-scale image retrieval, hashing has been extensively explored in approximate nearest neighbor search methods due to its low storage and high computational efficiency. With the development of deep learning, deep hashing methods have made great progress in image retrieval. Most existing deep hashing methods cannot fully consider the intra-group correlation of hash codes, which leads to the correlation decrease problem of similar hash codes and ultimately affects the retrieval results. In this article, we propose an end-to-end siamese dilated inception hashing (SDIH) method that takes full advantage of multi-scale contextual information and category-level semantics to enhance the intra-group correlation of hash codes for hash codes learning. First, a novel siamese inception dilated network architecture is presented to generate hash codes with the intra-group correlation enhancement by exploiting multi-scale contextual information and category-level semantics simultaneously. Second, we propose a new regularized term, which can force the continuous values to approximate discrete values in hash codes learning and eventually reduces the discrepancy between the Hamming distance and the Euclidean distance. Finally, experimental results in five public data sets demonstrate that SDIH can outperform other state-of-the-art hashing algorithms.

Two-Stream Deep Hashing With Class-Specific Centers for Supervised Image Search.

Hashing has been widely used for large-scale approximate nearest neighbor search due to its storage and search efficiency. Recent supervised hashing research has shown that deep learning-based methods can significantly outperform nondeep methods. Most existing supervised deep hashing methods exploit supervisory signals to generate similar and dissimilar image pairs for training. However, natural images can have large intraclass and small interclass variations, which may degrade the accuracy of hash codes. To address this problem, we propose a novel two-stream ConvNet architecture, which learns hash codes with class-specific representation centers. Our basic idea is that if we can learn a unified binary representation for each class as a center and encourage hash codes of images to be close to the corresponding centers, the intraclass variation will be greatly reduced. Accordingly, we design a neural network that leverages label information and outputs a unified binary representation for each class. Moreover, we also design an image network to learn hash codes from images and force these hash codes to be close to the corresponding class-specific centers. These two neural networks are then seamlessly incorporated to create a unified, end-to-end trainable framework. Extensive experiments on three popular benchmarks corroborate that our proposed method outperforms current state-of-the-art methods.

Optimized high order product quantization for approximate nearest neighbors search

Product quantization is now considered as an effective approach to solve the approximate nearest neighbor (ANN) search. A collection of derivative algorithms have been developed. However, the current techniques ignore the intrinsic high order structures of data, which usually contain helpful information for improving the computational precision. In this paper, aiming at the complex structure of high order data, we design an optimized technique, called optimized high order product quantization (O-HOPQ) for ANN search. In O-HOPQ, we incorporate the high order structures of the data into the process of designing a more effective subspace decomposition way. As a result, spatial adjacent elements in the high order data space are grouped into the same sub-space. Then, O-HOPQ generates its spatial structured code-book, by optimizing the quantization distortion. Starting from the structured codebook, the global optimum quantizers can be obtained effectively and efficiently. Experimental results show that appropriate utilization of the potential information that exists in the complex structure of high order data will result in significant improvements to the performance of the product quantizers. Besides, the high order structure based approaches are effective to the scenario where the data have intrinsic complex structures.

Semantic Correlation Based Deep Cross-Modal Hashing For Faster Retrieval

Due to growth of multi-modal data, large amount of data is being generated. Nearest Neighbor (NN) search is used to retrieve information but it suffers when there is high-dimensional data. However Approximate Nearest Neighbor (ANN) is a searching method which is extensively used by the researchers where data is represented in form of binary code using semantic hashing. Such representation reduces the storage cost and retrieval speed. In addition, deep learning has shown good performance in information retrieval which efficiently handle scalability problem. The multi-modal data has different statistical properties so there is a need to have method which finds semantic correlation between them. In this paper, experiment is performed using correlation methods like CCA, KCCA and DCCA on NMIST dataset. NMIST dataset is multi-view dataset and result proves that DCCA outperforms over CCA and KCCA by learning representations with higher correlations. However, due to flexible requirements of users, cross-modal retrieval plays very important role which works across the modalities. Traditional cross-modal hashing techniques are based on the hand-crafted features. So performance is not satisfactory as feature learning and binary code generation is independent process. In addition, traditional cross-modal hashing techniques fail to bridge the heterogeneous gap over various modalities. So many deep-based cross-modal hashing techniques were proposed which improves the performance in comparison with non-deep cross-modal techniques. Inside the paper, we presented a comprehensive survey of hashing techniques which works across the modalities.

A distributed approximate nearest neighbors algorithm for efficient large scale mean shift clustering

Mean Shift clustering, as a generalization of the well-known k-means clustering, computes arbitrarily shaped clusters as defined as the basins of attraction to the local modes created by the density gradient ascent paths. Despite its potential for improved clustering accuracy, the Mean Shift approach is a computationally expensive method for unsupervised learning. We introduce two contributions aiming to provide approximate Mean Shift clustering, based on scalable procedures to compute the density gradient ascent and cluster labeling, with a linear time complexity, as opposed to the quadratic time complexity for the exact clustering. Both propositions are based on Locality Sensitive Hashing (LSH) to approximate nearest neighbors. When implemented on a serial system, these approximate methods can be used for moderate sized datasets. To facilitate the analysis of Big Data, a distributed implementation, written for the Spark/Scala ecosystem is proposed. An added benefit is that our proposed approximations of the density gradient ascent, when used as a pre-processing step in other clustering methods, can also improve the clustering accuracy of the latter. We present experimental results illustrating the effect of tuning parameters on cluster labeling accuracy and execution times, as well as the potential to solve concrete problems in Big Clustering.

Extremely Fast and Accurate Open Modification Spectral Library Searching of High-Resolution Mass Spectra Using Feature Hashing and Graphics Processing Units.

Open modification searching (OMS) is a powerful search strategy to identify peptides with any type of modification. OMS works by using a very wide precursor mass window to allow modified spectra to match against their unmodified variants, after which the modification types can be inferred from the corresponding precursor mass differences. A disadvantage of this strategy, however, is the large computational cost, because each query spectrum has to be compared against a multitude of candidate peptides.We have previously introduced the ANN-SoLo tool for fast and accurate open spectral library searching. ANN-SoLo uses approximate nearest neighbor indexing to speed up OMS by selecting only a limited number of the most relevant library spectra to compare to an unknown query spectrum. Here we demonstrate how this candidate selection procedure can be further optimized using graphics processing units. Additionally, we introduce a feature hashing scheme to convert high-resolution spectra to low-dimensional vectors. Based on these algorithmic advances, along with low-level code optimizations, the new version of ANN-SoLo is up to an order of magnitude faster than its initial version. This makes it possible to efficiently perform open searches on a large scale to gain a deeper understanding about the protein modification landscape. We demonstrate the computational efficiency and identification performance of ANN-SoLo based on a large data set of the draft human proteome.ANN-SoLo is implemented in Python and C++. It is freely available under the Apache 2.0 license at https://github.com/bittremieux/ANN-SoLo.

Deep supervised hashing network with integrated regularisation

Hashing has been widely deployed to approximate nearest neighbour search for large-scale multimedia retrieval tasks due to storage and retrieval efficiency. State-of-the-art supervised hashing methods for image retrieval construct deep structures to simultaneously learn image representation and generate good hash codes, and the key step among them is simultaneously learned feature representation and binary hash code. Existing methods use similarity and regularity loss to train deep hashing systems, but these two functions usually work together but not cooperative, which may lead to inadequate performance of the whole system. In this study, a new method for training deep hashing system to learn compact binary codes is presented. The deep supervised hashing network with integrated regularisation (DSHIR) system develop the zero division restriction as a new part of the loss function, which settles the problem of cooperatively guiding the system generate similarity preserving binary codes. DSHIR system also modifies the similarity handling loss to better extract features from image data, which promotes the performance compared to existing end-to-end deep hashing systems. Experiments show that DSHIR yields about 10 per cent higher mean average precision on CIFAR-10 dataset, and also promote on other evaluation indexes compared with state-of-the-art systems.

Deep Scalable Supervised Quantization by Self-Organizing Map

Approximate Nearest Neighbor (ANN) search is an important research topic in multimedia and computer vision fields. In this article, we propose a new deep supervised quantization method by Self-Organizing Map to address this problem. Our method integrates the Convolutional Neural Networks and Self-Organizing Map into a unified deep architecture. The overall training objective optimizes supervised quantization loss as well as classification loss. With the supervised quantization objective, we minimize the differences on the maps between similar image pairs and maximize the differences on the maps between dissimilar image pairs. By optimization, the deep architecture can simultaneously extract deep features and quantize the features into suitable nodes in self-organizing map. To make the proposed deep supervised quantization method scalable for large datasets, instead of constructing a larger self-organizing map, we propose to divide the input space into several subspaces and construct self-organizing map in each subspace. The self-organizing maps in all the subspaces implicitly construct a large self-organizing map, which costs less memory and training time than directly constructing a self-organizing map with equal size. The experiments on several public standard datasets prove the superiority of our approaches over the existing ANN search methods. Besides, as a by-product, our deep architecture can be directly applied to visualization with little modification, and promising performance is demonstrated in the experiments.

A supervised learning to index model for approximate nearest neighbor image retrieval

Image indexing plays a significant role in approximate K-nearest neighbor image retrieval. In the past decades, the most commonly used methods in image indexing are unsupervised-based. We recognize that unsupervised-based methods cannot benefit from supervised information. In this paper, supervised learning to index method is proposed to realize approximate K-nearest neighbor image retrieval. For designing the supervised learning to index algorithm, we first need to establish the supervision information of an image that should be quantified to the codeword in the codebook. Thus, we propose an image relabeling algorithm to relabel the database images so that the supervised codebook learning can be realized. Second, compared with only distance-based unsupervised image indexing method, a better idea of supervised learning to index is incorporating the distance-based similarities with the classification probabilities of samples at the image indexing stage. In order to achieve this goal, a classifier is also trained based on the relabeled images, and we finally build a constraint optimization objective function to learn the codebook and the classifier simultaneously. The proposed method is verified on the MNIST, the CIFAR-10, and the ImageNet data sets. Experimental results demonstrate the effectiveness of the proposed method.

Improved Deep Hashing With Soft Pairwise Similarity for Multi-Label Image Retrieval

Hash coding has been widely used in the approximate nearest neighbor search for large-scale image retrieval. Recently, many deep hashing methods have been proposed and shown largely improved performance over traditional feature-learning-based methods. Most of these methods examine the pairwise similarity on the semantic-level labels, where the pairwise similarity is generally defined in a hard-assignment way. That is, the pairwise similarity is '1' if they share no less than one class label and '0' if they do not share any. However, such similarity definition cannot reflect the similarity ranking for pairwise images that hold multiple labels. In this paper, a new deep hashing method is proposed for multi-label image retrieval by re-defining the pairwise similarity into an instance similarity, where the instance similarity is quantified into a percentage based on the normalized semantic labels. Based on the instance similarity, a weighted cross-entropy loss and a minimum mean square error loss are tailored for loss-function construction, and are efficiently used for simultaneous feature learning and hash coding. Experiments on three popular datasets demonstrate that, the proposed method outperforms the competing methods and achieves the state-of-the-art performance in multi-label image retrieval.

Clustering co-abundant genes identifies components of the gut microbiome that are reproducibly associated with colorectal cancer and inflammatory bowel disease

BackgroundWhole-genome “shotgun” (WGS) metagenomic sequencing is an increasingly widely used tool for analyzing the metagenomic content of microbiome samples. While WGS data contains gene-level information, it can be challenging to analyze the millions of microbial genes which are typically found in microbiome experiments. To mitigate the ultrahigh dimensionality challenge of gene-level metagenomics, it has been proposed to cluster genes by co-abundance to form Co-Abundant Gene groups (CAGs). However, exhaustive co-abundance clustering of millions of microbial genes across thousands of biological samples has previously been intractable purely due to the computational challenge of performing trillions of pairwise comparisons.ResultsHere we present a novel computational approach to the analysis of WGS datasets in which microbial gene groups are the fundamental unit of analysis. We use the Approximate Nearest Neighbor heuristic for near-exhaustive average linkage clustering to group millions of genes by co-abundance. This results in thousands of high-quality CAGs representing complete and partial microbial genomes. We applied this method to publicly available WGS microbiome surveys and found that the resulting microbial CAGs associated with inflammatory bowel disease (IBD) and colorectal cancer (CRC) were highly reproducible and could be validated independently using multiple independent cohorts.ConclusionsThis powerful approach to gene-level metagenomics provides a powerful path forward for identifying the biological links between the microbiome and human health. By proposing a new computational approach for handling high dimensional metagenomics data, we identified specific microbial gene groups that are associated with disease that can be used to identify strains of interest for further preclinical and mechanistic experimentation.

A Meta-Algorithm for Improving Top-N Prediction Efficiency of Matrix Factorization Models in Collaborative Filtering

Matrix factorization models often reveal the low-dimensional latent structure in high-dimensional spaces while bringing space efficiency to large-scale collaborative filtering problems. Improving training and prediction time efficiencies of these models are also important since an accurate model may raise practical concerns if it is slow to capture the changing dynamics of the system. For the training task, powerful improvements have been proposed especially using SGD, ALS, and their parallel versions. In this paper, we focus on the prediction task and combine matrix factorization with approximate nearest neighbor search methods to improve the efficiency of top-N prediction queries. Our efforts result in a meta-algorithm, MMFNN, which can employ various common matrix factorization models, drastically improve their prediction efficiency, and still perform comparably to standard prediction approaches or sometimes even better in terms of predictive power. Using various batch, online, and incremental matrix factorization models, we present detailed empirical analysis results on many large implicit feedback datasets from different application domains.