Search

Multi-task learning (MTL) can boost the performance of individual tasks by mutual learning among multiple related tasks. However, when these tasks assume diverse complexities, their corresponding losses involved in the MTL objective inevitably compete with each other and ultimately make the learning biased towards simple tasks rather than complex ones. To address this imbalanced learning problem, we propose a novel MTL method that can equip multiple existing deep MTL model architectures with a sequential cooperative distillation (SCD) module. Specifically, we first introduce an efficient mechanism to measure the similarity between tasks, and group similar tasks into the same block to allow their cooperative learning from each other. Based on this, the grouped task blocks are sorted in a queue to determine the learning sequence of the tasks according to their complexities estimated with the defined performance indicator. Finally, a distillation between the individual task-specific models and the MTL model is performed block by block from complex to simple manner, achieving a balance between competition and cooperation among learning multiple tasks. Extensive experiments demonstrate that our method is significantly more competitive compared with state-of-the-art methods, ranking No.1 with average performances across multiple datasets by improving 12.95% and 3.72% compared with OMTL and MTLKD, respectively. [ABSTRACT FROM AUTHOR]

Cross-modal hashing that leverages hash functions to project high-dimensional data from different modalities into the compact common hamming space, has shown immeasurable potential in cross-modal retrieval. To ease labor costs, unsupervised cross-modal hashing methods are proposed. However, existing unsupervised methods still suffer from two factors in the optimization of hash functions: 1) similarity guidance, they barely give a clear definition of whether is similar or not between data points, leading to the residual of the redundant information; 2) optimization strategy, they ignore the fact that the similarity learning abilities of different hash functions are different, which makes the hash function of one modality weaker than the hash function of the other modality. To alleviate such limitations, this paper proposes an unsupervised cross-modal hashing method to train hash functions with discriminative similarity guidance and adaptively-enhanced optimization strategy, termed Deep Adaptively-Enhanced Hashing (DAEH). Specifically, to estimate the similarity relations with discriminability, Information Mixed Similarity Estimation (IMSE) is designed by integrating information from distance distributions and the similarity ratio. Moreover, Adaptive Teacher Guided Enhancement (ATGE) optimization strategy is also designed, which employs information theory to discover the weaker hash function and utilizes an extra teacher network to enhance it. Extensive experiments on three benchmark datasets demonstrate the superiority of the proposed DAEH against the state-of-the-arts. [ABSTRACT FROM AUTHOR]

Additional context information is vital for context-aware recommender systems. The whole paradigm of context-aware recommender systems is built upon the availability of contextual features. Apart from the significance of context, we highlight a key issue for existing context-aware recommendation paradigm that if the user environment did not provide contextual features such as time, location, or companion due to privacy constraints or if the data collection system is unable to record contextual attributes due to legal or technical concerns then the existing context-aware recommendation paradigm has no uniform mechanism to deal with this situation. In this research, we address these challenges and propose a novel item-context similarity (ICS) model capable of adaptively generating reliable collaborators for a subject user on a subject item. Additionally, ICS is fused into a weighting model called contextually reliable collaborators (CRC) that considers the current item context, the nonlinear relationship between candidate collaborators and the asymmetry between rating preferences of users to finally generate rating prediction. Experiments show that neighbors computed through ICS are more reliable than the classical similarity estimation methods and the ICS-based CRC model has outperformed state-of-the-art approaches. [ABSTRACT FROM AUTHOR]

Loop Closure Detection (LCD) is an essential component of visual Simultaneous Localization and Mapping (SLAM) systems. It enables the recognition of previously visited scenes to eliminate pose and map estimate drifts arising from long-term exploration. However, current appearance-based LCD methods face significant challenges, including high computational costs, viewpoint changes, and dynamic objects in scenes. This paper introduced an online appearance based LCD using Local Superpixel Grids Descriptor (LSGD) and Dynamic Nodes (DN), i.e., LSGDDN-LCD, to find similarities between scenes via handcrafted features extracted from the LSGD. Additionally, we proposed the adaptive mechanism to group similar scenes called Dynamic Nodes , which incrementally adjusted the database in an online manner, allowing for efficient and online retrieval of previously viewed images without need of the pre-training. Experimental results confirmed that the LSGDDN-LCD significantly improved LCD precision–recall and efficiency, and outperformed several state-of-the-art (SOTA) approaches on public and our own datasets, indicating its great potential as a generic LCD framework. Our implementation of the LSGDDN-LCD approach and the datasets were open-sourced on GitHub (https://github.com/BaoshengZhang0/LSGDDN-LCD.git). [ABSTRACT FROM AUTHOR]

Abstract Estimating the similarity between two images or image patches is at the heart of many computer vision problems including content-based image retrieval, image registration, and scene recognition. However, commonly used distance-based similarity estimation is not always reliable due to the limitations in both image understanding techniques and distance metrics. In this paper, we present a scheme to improve the similarity estimation under image search scenario. To this end, we explore the discriminative capability underlying global distance distribution obtained by querying an auxiliary image dataset in an unsupervised manner. According to the results of motivational experiments, we discover that global distance distributions have the desired capability in distinguishing inter-class images which can be applied to enhance the original distance metric. Following this finding, we propose a novel approach to incorporate the global distance distribution into the original distance metric to improve the reliability of the similarity estimation. One key novelty of this approach is to model the global distance distribution as Rayleigh distribution and then represent the difference between two distributions by the relative entropy. In this way, the difference between two global distance distributions can be calculated in an extremely efficient way. We also demonstrate that Rayleigh distribution leads to consistent performance compared to the real distribution. Extensive experiments on three public datasets with various image representations and distance metrics show that the enhanced similarity estimation remarkably outperforms the original one. Furthermore, the proposed approach shows the desired scalability for handling large-scale image search scenarios. [ABSTRACT FROM AUTHOR]

Data mining from web pages becomes more frequently adapted in business areas. However on the one hand while analyzing the current situation, we observe that solutions for mining structured data from web pages exists. On the other hand we see that a scientific dataset for unstructured data that would allow create and test new data selection methods does not exist. This limits the development and research of unstructured web data therefore we propose a method for HTML code block similarity estimation. The method combines both data and structure comparison and allows quantitative similarity presentation of two HTML code blocks.

Recently, tremendous growth of web services to share the program, data and resources requires the optimal recommendation strategy. The major issues observed in existing recommendation strategies are scalability, sparsity and the cold start. The employment of matrix factorization (MF) models addressed all the issues effectively. But, they increase the scalability of the system. This paper proposes the new framework that contains web service grouping, distance estimation, service utilization level estimation and the item-to-item comparison (Pearson Correlation Coefficient (PCC)) to improve the recommendation performance. The grouping of users according to the Haversine distance formulation to reduce the complexity in the relevant web service recommendation against the complex queries. The locations and the fields in the services utilization in proposed work provide the effective recommendation performance. The comparative analysis between the proposed novel recommendation framework with the existing techniques assures the effectiveness of proposed approach in web service recommendation. [ABSTRACT FROM AUTHOR]

The application of automated correlation optimized warping (ACOW) to the correction of retention time shift in the chromatographic fingerprints of Radix Puerariae thomsonii (RPT) was investigated. Twenty-seven samples were extracted from 9 batches of RPT products. The fingerprints of the 27 samples were established by the HPLC method. Because there is a retention time shift in the established fingerprints, the quality of these samples cannot be correctly evaluated by using similarity estimation and principal component analysis (PCA). Thus, the ACOW method was used to align these fingerprints. In the ACOW procedure, the warping parameters, which have a significant influence on the alignment result, were optimized by an automated algorithm. After correcting the retention time shift, the quality of these RPT samples was correctly evaluated by similarity estimation and PCA. It is demonstrated that ACOW is a practical method for aligning the chromatographic fingerprints of RPT. The combination of ACOW, similarity estimation, and PCA is shown to be a promising method for evaluating the quality of Traditional Chinese Medicine.

User preferences play a significant role in market analysis. In the database literature, there has been extensive work on query primitives, such as the well known top-$k$ query that can be used for the ranking of products based on the preferences customers have expressed. Still, the fundamental operation that evaluates the similarity between products is typically done ignoring these preferences. Instead products are depicted in a feature space based on their attributes and similarity is computed via traditional distance metrics on that space. In this work, we utilize the rankings of the products based on the opinions of their customers in order to map the products in a user-centric space where similarity calculations are performed. We identify important properties of this mapping that result in upper and lower similarity bounds, which in turn permit us to utilize conventional multidimensional indexes on the original product space in order to perform these user-centric similarity computations. We show how interesting similarity calculations that are motivated by the commonly used range and nearest neighbor queries can be performed efficiently, while pruning significant parts of the data set based on the bounds we derive on the user-centric similarity of products. [ABSTRACT FROM AUTHOR]

The popularity of the online peer-to-peer (P2P) lending platform in the financial supply chain (FSC) has grown tremendously in the past few years. However, it is pretty challenging to develop an efficient financial supply chain for Micro, Small, and Medium-Sized Enterprises (MSMEs) by providing credits and services. In this research, an FSC model analyzes by emphasizing the retailer and supplier relationship. The retailer predicts the market demand to maximize profit, and accordingly, orders to the suppliers and suppliers utilize the production capabilities in limited monetary value. In such a case, the supplier also needs to apply for loans to meet the production commitment. With an intelligent lending platform, applicants can get loans quickly and more conveniently from the lenders. Still, identifying defaulter borrowers is a difficult task on the peer-to-peer lending platform. The main focus of the lenders is to maximize profit and minimize risk by giving the loan to non-defaulter suppliers. The present study describes the importance of the financial parameters (λ) of the retailers and suppliers (MSMEs) and their impact on the related supply chain. We provide the proof of dependency of the financial parameter on the cost of debt and the ratio of the debt-to-equity by developing two propositions. We propose an innovative k -Random Boosting Classifiers (k -RBC) algorithm for identifying potential good and bad borrowers to capture this case. The time complexity of the k -RBC algorithm is O (n 2 l o g (n)). The results obtained from the study show a significant improvement in comparison to the outputs from existing approaches on the same datasets. Our algorithm gives 90% accuracy to identify potential good borrowers, whereas existing algorithms achieve up to 87% accuracy. Furthermore, the importance of the borrowers' features and its impact are analyzed of the lending platform in FSC. • Defines the importance of financial parameters and its impact on supply chain. • Provides the proof of dependency of the financial parameter on the cost of debt. • Prove dependency of the financial parameter on the ratio of the debt-to-equity. • Proposes the k-Random Boosting Classification model to identify defaulters. • Identifies crucial features to enhance the probability of getting loans approval. [ABSTRACT FROM AUTHOR]

Object tracking methods based on the principal component analysis (PCA) are effective against object change caused by illumination variation and motion blur. However, when the object is occluded, the tracking result of the PCA-based methods will drift away from the target. In this paper, we propose a new robust object tracking method based on the PCA and local sparse representation (LSR). First, candidates are reconstructed through the PCA subspace model in global manner. To handle occlusion, a patch-based similarity estimation strategy is proposed for the PCA subspace model. In the patch-based strategy, the PCA representation error map is divided into patches to estimate the similarity between target and candidate considering the occlusion. Second, the LSR is introduced to detect the occluded patches of the object and estimate the similarity through the residual error in the sparse coding. Finally, the two similarities of each candidate from the PCA subspace model and LSR model are fused to predict the tracking result. The experimental results demonstrate that the proposed tracking method favorably performs against several state-of-the-art methods on challenging image sequences. [ABSTRACT FROM PUBLISHER]

Estimating the fitness value of individuals in an evolutionary algorithm in order to reduce the computational expense of actually calculating the fitness has been a classical pursuit of practitioners. One area which could benefit from progress in this endeavour is bot evolution, i.e. the evolution of non-playing characters in computer games. Because assigning a fitness value to a bot (or rather, the decision tree that controls its behaviour) requires playing the game, the process is very costly. In this work, we introduce two major contributions to speed up this process in the computer game Unreal Tournament 2004™. Firstly, a method for fitness value approximation in genetic programming which is based on the idea that individuals that behave in a similar fashion will have a similar fitness. Thus, similarity of individuals is taken at the performance level, in contrast to commonly employed approaches which are either based on similarity of genotypes or, less frequently, phenotypes. The approximation performs a weighted average of the fitness values of a number of individuals, attaching a confidence level which is based on similarity estimation. The latter is the second contribution of this work, namely a method for estimating the similarity between individuals. This involves carrying out a number of tests consisting of playing a 'static' version of the game (with fixed inputs) with the individuals whose similarity is under evaluation and comparing the results. Because the tests involve a limited version of the game, the computational expense of the similarity estimation plus that of the fitness approximation is much lower than that of directly calculating the fitness. The success of the fitness approximation by similarity estimation method for bot evolution in UT2K4 allows us to expect similar results in environments that share the same characteristics. [ABSTRACT FROM AUTHOR]

Quantitative genetics (QG) analyses variation in traits of humans, other animals, or plants in ways that take account of the genealogical relatedness of the individuals whose traits are observed. 'Classical' QG, where the analysis of variation does not involve data on measurable genetic or environmental entities or factors, is reformulated in this article using models that are free of hypothetical, idealized versions of such factors, while still allowing for defined degrees of relatedness among kinds of individuals or 'varieties.' The gene- free formulation encompasses situations encountered in human QG as well as in agricultural QG. This formulation is used to describe three standard assumptions involved in classical QG and provide plausible alternatives. Several concerns about the partitioning of trait variation into components and its interpretation, most of which have a long history of debate, are discussed in light of the gene-free formulation and alternative assumptions. That discussion is at a theoretical level, not dependent on empirical data in any particular situation. Additional lines of work to put the gene-free formulation and alternative assumptions into practice and to assess their empirical consequences are noted, but lie beyond the scope of this article. The three standard QG assumptions examined are: (1) partitioning of trait variation into components requires models of hypothetical, idealized genes with simple Mendelian inheritance and direct contributions to the trait; (2) all other things being equal, similarity in traits for relatives is proportional to the fraction shared by the relatives of all the genes that vary in the population (e.g., fraternal or dizygotic twins share half of the variable genes that identical or monozygotic twins share); (3) in analyses of human data, genotype-environment interaction variance (in the classical QG sense) can be discounted. The concerns about the partitioning of trait variation discussed include: the distinction between traits and underlying measurable factors; the possible heterogeneity in factors underlying the development of a trait; the kinds of data needed to estimate key empirical parameters; and interpretations based on contributions of hypothetical genes; as well as, in human studies, the labeling of residual variance as a non-shared environmental effect; and the importance of estimating interaction variance. [ABSTRACT FROM AUTHOR]

To quickly find documents with high similarity in existing documentation sets, fingerprint group merging retrieval algorithm is proposed to address both sides of the problem: a given similarity threshold could not be too low and fewer fingerprints could lead to low accuracy. It can be proved that the efficiency of similarity retrieval is improved by fingerprint group merging retrieval algorithm with lower similarity threshold. Experiments with the lower similarity threshold r=0.7 and high fingerprint bits k=400 demonstrate that the CPU time-consuming cost decreases from 1 921 s to 273 s. Theoretical analysis and experimental results verify the effectiveness of this method. [ABSTRACT FROM AUTHOR]

Most of the written materials are consisted of Multimedia (MM) information because beside text usually contain image information. The present information retrieval and filtering systems use only text parts of the documents or in best case images represented by keywords or image captions. Why do not use both, text and image features of the documents and in the retrieval or filtering process utilize more completely the document information content? Can such approach increase the effectiveness of retrieval and filtering processes? There is a very little difference between retrieval and filtering at an abstract level. In this paper, we will discuss some possible similarities and differences between them on the application level taking into account the experiments in retrieval and filtering of multimedia mineral information. [ABSTRACT FROM AUTHOR]

Single-cell RNA sequencing technologies have revolutionized biomedical research by providing an effective means to profile gene expressions in individual cells. One of the first fundamental steps to perform the in-depth analysis of single-cell sequencing data is cell type classification and identification. Computational methods such as clustering algorithms have been utilized and gaining in popularity because they can save considerable resources and time for experimental validations. Although selecting the optimal features (i.e., genes) is an essential process to obtain accurate and reliable single-cell clustering results, the computational complexity and dropout events that can introduce zero-inflated noise make this process very challenging. In this paper, we propose an effective single-cell clustering algorithm based on the ensemble feature selection and similarity measurements. We initially identify the set of potential features, then measure the cell-to-cell similarity based on the subset of the potentials through multiple feature sampling approaches. We construct the ensemble network based on cell-to-cell similarity. Finally, we apply a network-based clustering algorithm to obtain single-cell clusters. We evaluate the performance of our proposed algorithm through multiple assessments in real-world single-cell RNA sequencing datasets with known cell types. The results show that our proposed algorithm can identify accurate and consistent single-cell clustering. Moreover, the proposed algorithm takes relative expression as input, so it can easily be adopted by existing analysis pipelines. The source code has been made publicly available at https://github.com/jeonglab/scCLUE. [ABSTRACT FROM AUTHOR]