The optimization of large-scale SVM using nest template ant clustering in kernel space

To solve the problem that traditional multi-label support vector machine (SVM) classification algorithm adopting nonlinear kernel has been severely restricted from being used on large-scale data sets, we propose fast multi-label low-rank-linearized SVM classification algorithm based on approximate extreme points (AEML-LLSVM). First, it adopts the approximate extreme points' method to obtain representative sets from the training data set. Then, the approximate extreme points' low-rank-linearized SVM (AELLSVM) is trained on the representative sets. The AELLSVM integrates the advantages of approximate extreme points' method and LLSVM. Experimental results on three large-scale multi-label data sets have proven that the training and the testing speed of AEML-LLSVM classification algorithm are greatly improved under the premise that its classification performance is similar to that of ML-LIBSVM classification algorithm and superior to that of other fast multi-label SVM classification algorithms.

Recent years have witnessed extensive attention in binary code learning, a.k.a. hashing, for nearest neighbor search problems. It has been seen that high-dimensional data points can quantize into binary codes to give an efficient similarity approximation via Hamming distance. Among the existing schemes, ranking-based hashing is recent promising that targets at preserving ordinal relations of ranking in the Hamming space to minimize retrieval loss. However, the size of the ranking tuples that show the ordinal relations, is quadratic or cubic to the size of training samples. It is so very expensive to embed such ranking tuples in binary code learning, especially given a large-scale training data set. Besides, it remains difficult to build ranking tuples efficiently for most ranking-preserving hashing, which are deployed over an ordinal graph-based setting. To handle these problems, we propose a novel ranking-preserving hashing method, dubbed Ordinal Constraint Hashing (OCH), which efficiently learns the optimal hashing functions with a graph-based approximation to embed the ordinal relations. The core idea is to reduce the size of ordinal graph with ordinal constraint projection, which preserves the ordinal relations through a small data set (such as clusters or random samples). In particular, to learn such hash functions effectively, we further relax the discrete constraints and design a specific stochastic gradient decent algorithm for optimization. Experimental results on three large-scale visual search benchmark datasets, i.e. LabelMe, Tiny100K and GIST1M, show that the proposed OCH method can achieve superior performance over the state-of-the-arts approaches.

Blind quality assessment for screen content images by combining local and global features

Advanced medical imaging technologies have enabled biologists and other researchers in biomedicine, biochemistry and bio-informatics to gain better insight in complex, large-scale data sets. These datasets, which occupy large amounts of space, can no longer be stored on local hard drives. San Diego Supercomputer Center (SDSC) maintains a large data repository, called High Performance Storage System (HPSS), where large-scale biomedical data sets can be stored. These data sets must be transmitted over an open or closed network (Internet or Intranet) within a reasonable amount of time to make them accessible in an interactive fashion to the researchers all over the world. Our approach deals with extracting, compressing and transmitting these data sets using the Haar wavelets, over a low- to medium-bandwidth network. These compressed data sets are then transformed and reconstructed into a 3-D volume on the client side using texture mapping in Java3D. These data sets are handled using the Scalable Visualization Toolkits provided by the NPACI (National Partnership for Advanced Computational Infrastructure). Sub-volumes of the data sets are extracted to provide a detailed view of a particular region of interest (ROI). This application is being ported to C++ platform to obtain higher rendering speed and better performance but lacks platform independency.

The feature models used by existing Thermal InfraRed (TIR) tracking methods are usually learned from RGB images due to the lack of a large-scale TIR image training dataset. However, these feature models are less effective in representing TIR objects and they are difficult to effectively distinguish distractors because they do not contain fine-grained discriminative information. To this end, we propose a dual-level feature model containing the TIR-specific discriminative feature and fine-grained correlation feature for robust TIR object tracking. Specifically, to distinguish inter-class TIR objects, we first design an auxiliary multi-classification network to learn the TIR-specific discriminative feature. Then, to recognize intra-class TIR objects, we propose a fine-grained aware module to learn the fine-grained correlation feature. These two kinds of features complement each other and represent TIR objects in the levels of inter-class and intra-class respectively. These two feature models are constructed using a multi-task matching framework and are jointly optimized on the TIR object tracking task. In addition, we develop a large-scale TIR image dataset to train the network for learning TIR-specific feature patterns. To the best of our knowledge, this is the largest TIR tracking training dataset with the richest object class and scenario. To verify the effectiveness of the proposed dual-level feature model, we propose an offline TIR tracker (MMNet) and an online TIR tracker (ECO-MM) based on the feature model and evaluate them on three TIR tracking benchmarks. Extensive experimental results on these benchmarks demonstrate that the proposed algorithms perform favorably against the state-of-the-art methods.

Many object re-identification (Re-ID) methods that depend on large-scale training datasets have been proposed in recent years. However, the performance of these methods degrades dramatically when insufficient training data are available. To address this challenging problem, we propose a few-shot object re-identification (FSOR) method that enhances the generalization and discrimination abilities of object Re-ID models trained on small datasets. This method applies two novel techniques: reparameterization for feature vectors and dual-distance metric learning. The reparameterization mechanism transforms the primary feature vector of each input image into a Gaussian distribution to enhance the robustness of the FSOR method when performing object Re-ID tasks. The dual-distance metric learning technique, called H&C learning, considers both the hard mining distance and the center-point distance between each query sample and each support set of different object identities. H&C learning extracts the characteristics of the entire training dataset more precisely than other approaches and thus improves the discriminative abilities of object Re-ID models. Extensive experiments on both person and vehicle Re-ID datasets, such as Market-1501, DukeMTMC-ReID, CUHK03, and VeRi-776, show that the FSOR method has improved performance and outperforms state-of-the-art methods when the amount of labeled training data is small.

Application of the modified barrier method in large-scale Quadratic Programming problems

We propose a block-based sampling method (I-sampling) which randomly selects the base data blocks from the block pool of a large-scale dataset rather than directly chooses records from the original dataset. I-sampling firstly partitions the given large-scale dataset into the non-overlapping primary data blocks. Secondly, the records in each primary data block are randomly shuffled and the corresponding shuffling data blocks are formed. Thirdly, I-sampling generates a block pool which is a set of base data blocks. The theoretical analysis proves that the base data block has the approximately equal probability distribution with the original dataset. The training dataset is finally produced by randomly selecting base data blocks from the block pool. Different from the traditional record-based sampling method, I-sampling has the good extendibility to deal with the large-scale datasets on distributed system. The simulated experiments demonstrate the feasibility of I-sampling and reveal that the training datasets which are provided by I-sampling are equivalent to the simple random sampling (one of representative record-based sampling methods) for the given machine learning algorithm.

An outbreak of the coronavirus (Covid-19) has spread around the world, causing widespread health concerns. The pandemic's reach has spread to nearly every development sector. There have been numerous steps attempted to prevent the in areas where Covid-19 has spread, wearing a face mask is mandatory. The Covid-19 transmission could not have been controlled without it. To top it all off, seeks to detect finegrained wearing states of the face mask: make sure your face is covered with the right mask Without a disguise. There are two very difficult jobs in our job: first, because of an online market for augmented data sets as well as large-scale training datasets. A face mask identification app for mobile devices is presented in this paper. In its entirety, Cloud-based machine learning technology known as Google Cloud ML API automates the process using TensorFlow files to train the model The ML's effectiveness is demonstrated in this research model. The progress of cloud technology is also examined in this paper. Over traditional coding methods, machine learning is becoming increasingly popular.

Relationship extraction is an important task in natural language processing and knowledge mapping. Traditional entity relationship extraction methods have achieved high accuracy in practical applications. However, when faced with the task of entity relationship extraction that is not easy to obtain large-scale supervision training data sets, traditional methods cannot get good results. In this paper, a few-shot relation extraction method based on multi-level feature metric learning is proposed. This method takes the prototype network as the baseline network to generate a class prototype. Firstly, a multi-level feature extraction module is proposed. This module combines the multi-level features of the text with the multi-level attention mechanism, which can fully extract the features of the text. Secondly, a loss function based on label value and negative sample distance is proposed. This algorithm introduces an evaluation mechanism of negative sample distance on the basis of the prototype network, so that the model can adaptively allocate parameters and improve the clustering ability of small samples. Experiments are conducted on FewRel1.0 which is a small sample relational data set. Experiment results show that compared with other models, our model can improve classification accuracy.

Large-scale face datasets usually exhibit a massive number of classes, a long-tailed distribution, and severe label noise, which undoubtedly aggravate the difficulty of training. In this paper, we propose a training strategy that treats the head data and the tail data in an unequal way, accompanying with noise-robust loss functions, to take full advantage of their respective characteristics. Specifically, the unequal-training framework provides two training data streams: the first stream applies the head data to learn discriminative face representation supervised by Noise Resistance loss; the second stream applies the tail data to learn auxiliary information by gradually mining the stable discriminative information from confusing tail classes. Consequently, both training streams offer complementary information to deep feature learning. Extensive experiments have demonstrated the effectiveness of the new unequal-training framework and loss functions. Better yet, our method could save a significant amount of GPU memory. With our method, we achieve the best result on MegaFace Challenge 2 (MF2) given a large-scale noisy training data set.

A semi-supervised deep learning approach for cropped image detection

The appropriateness of using statistical data to inform the design of any given service development or initiative often depends upon judgements regarding scale. Large-scale data sets, perhaps national in scope, whilst potentially important in informing the design, implementation and roll-out of experimental initiatives, will often remain unused for small-scale interventions involving only small numbers of service users or participants. This can be the result of legitimate concerns regarding representativity and rigour. Whilst large data sets may allow for strong claims to made regarding trends and tendencies with the population, a small scale intervention involving perhaps only tens of participants will not offer the opportunity to devise developmental rationales that are supported by such statistical rigour. This creates a situation in which important pilot studies and ‘suck-it-and-see’ interventions are conducted in a basic empirical fashion that does not draw upon the informed reflection that large scale data makes possible. This article will argue that such a situation is not necessary. Through graduated and selective engagement with large-scale data sets it is possible to make good use of what they offer to inform the design and up-scaling of small pilot interventions. The discussion will focus on the case of a small intervention involving just nine parents and their babies conducted in Liverpool schools. More specifically it will consider the ways in which one national data-set, the UK Millennium Cohort Study (MCS), can be used to develop this project. It will do this by way of profiling the MCS in order to both select out those aspects that could or should not be used and those aspects that could be used beneficially to inform the development of the initiative. In the process it will devise a set of criteria that could usefully be adapted for use by others in relation to service situations in which similar issues of scalar mis-match between available data-sets and service design exist.

For large-scale data sets, the traditional AP clustering algorithm processes all data samples through the idea of iteration and three metrics calculation. This kind of processing will lead to poor clustering effect and high time consumption. Based on this problem, an algorithm which is called KCAP based on Affinity Propagation clustering and Canopy clustering algorithm was proposed. Firstly, KNN distance was used to calculate the local density for each data. Secondly, we used the Canopy algorithm to roughly cluster the data sets, and the center points were formed into a new data set \(Y\). Then, we applied affinity propagation clustering algorithm to cluster the data set \(Y\). Finally, we assigned the data to the nearest center. Compared with the traditional AP algorithm, K-means algorithm and FCM algorithm, the experimental results on synthetic data sets and UCI data sets show that algorithm KCAP not only improves the clustering accuracy, but also have higher efficiency.KeywordsK-Nearest Neighbors (KNN)Initial centerCanopy algorithmAffinity Propagation (AP)Rough clustering

Evaluation of the performance of clustering algorithms in kernel-induced feature space