Set-based Similarity Research Articles

Measuring Similarity for Data-Aware Business Processes

Business process similarity measures are of vital importance for process repository management applications, such as process query, process recommendation, and process clustering. Most existing approaches measure process similarity by relying on control-flow structures only. This article investigates the role of data in process similarity measure. To incorporate data-flow information into business process control flow, it proposes a data-aware workflow net (DWF-net) by extending the classical workflow net with data reading and writing semantics. Then, we introduce three types of similarity measures, i.e., data item set-based similarity, data operation set-based similarity, and data-aware behavior-based similarity, to quantify the similarity of data-aware business processes from different perspectives. Next, a methodology is introduced to help process analysts apply these three measures in a systematical way. Finally, we evaluate the effectiveness and applicability of the proposed similarity measures by a group of comparative experiments.

Ranked List Loss for Deep Metric Learning.

The objective of deep metric learning (DML) is to learn embeddings that can capture semantic similarity and dissimilarity information among data points. Existing pairwise or tripletwise loss functions used in DML are known to suffer from slow convergence due to a large proportion of trivial pairs or triplets as the model improves. To improve this, ranking-motivated structured losses are proposed recently to incorporate multiple examples and exploit the structured information among them. They converge faster and achieve state-of-the-art performance. In this work, we unveil two limitations of existing ranking-motivated structured losses and propose a novel ranked list loss to solve both of them. First, given a query, only a fraction of data points is incorporated to build the similarity structure. Consequently, some useful examples are ignored and the structure is less informative. To address this, we propose to build a set-based similarity structure by exploiting all instances in the gallery. The learning setting can be interpreted as few-shot retrieval: given a mini-batch, every example is iteratively used as a query, and the rest ones compose the gallery to search, i.e., the support set in few-shot setting. The rest examples are split into a positive set and a negative set. For every mini-batch, the learning objective of ranked list loss is to make the query closer to the positive set than to the negative set by a margin. Second, previous methods aim to pull positive pairs as close as possible in the embedding space. As a result, the intraclass data distribution tends to be extremely compressed. In contrast, we propose to learn a hypersphere for each class in order to preserve useful similarity structure inside it, which functions as regularisation. Extensive experiments demonstrate the superiority of our proposal by comparing with the state-of-the-art methods on the fine-grained image retrieval task. Our source code is available online: https://github.com/XinshaoAmosWang/Ranked-List-Loss-for-DML.

RETRACTED ARTICLE: A novel hybrid approach based on principal component analysis and tolerance rough similarity for face identification

Face identification plays one of the most important roles in biometrics to recognize a person. However, face identification is very difficult because of variations in size, orientations, different illuminations, and face expressions. In this paper, a hybrid approach is proposed based on principal component analysis (PCA) and tolerance rough similarity (TRS) for face identification. This paper comprises of three steps. First, PCA has been used to extract the feature vector from face images (eigenvectors). Second, the tolerance rough set-based similarity is applied for face matching and finally, the test image is compared with lower and upper approximation of similarity values that were found using TRS. The proposed hybrid approach gives a better recognition rate compared to other standard techniques like Euclidean distance and cosine similarity. The proposed work is evaluated on three face databases namely OUR databases and ORL databases and Yale databases. The experimental result of the proposed PCA-TRS approach is compared with other standard classification techniques like support vector machine (SVM), multilayer perceptron (MLP), back propagation network (BPN) and simple decision tree (CART) to conclude that proposed approach is better for face identification because of high accuracy and minimum error rate.

Set-based similarity learning in subspace for agricultural remote sensing classification

Similarity between spectral lines is key in the field of agricultural sensing classification, however, the measured spectral lines mostly mislead the classification because of unexpected disturbance in application. To enhance the accuracy of classification, similarity learning is introduced into agricultural remote sensing classification. Within the framework of similarity learning, the training set is generated by pairing the labeled spectral lines which means the size of training set for learning similarity is heavily increasing. Noticed this problem, a novel spectrum-set similarity learning algorithm is reported for balancing the gain in classification and the computational burden of learning similarity. Different from traditional point-based similarity, the spectrum-set similarity measures the similarity between two spectral sets which contain some spectral lines. Following the idea, set-based training set is generated by clustering the spectral lines in the point-based training set. Experimental results have shown the effectiveness and efficiency of learning spectrum-set similarity measure for agriculture sensing classification.

Intelligent Evidence-Based Management for Data Collection and Decision-Making Using Algorithmic Randomness and Active Learning

We describe here a comprehensive framework for intelligent information management (IIM) of data collection and decision-making actions for reliable and robust event processing and recognition. This is driven by algorithmic information theory (AIT), in general, and algorithmic randomness and Kolmogorov complexity (KC), in particular. The processing and recognition tasks addressed include data discrimination and multilayer open set data categorization, change detection, data aggregation, clustering and data segmentation, data selection and link analysis, data cleaning and data revision, and prediction and identification of critical states. The unifying theme throughout the paper is that of “compression entails comprehension”, which is realized using the interrelated concepts of randomness vs. regularity and Kolmogorov complexity. The constructive and all encompassing active learning (AL) methodology, which mediates and supports the above theme, is context-driven and takes advantage of statistical learning, in general, and semi-supervised learning and transduction, in particular. Active learning employs explore and exploit actions characteristic of closed-loop control for evidence accumulation in order to revise its prediction models and to reduce uncertainty. The set-based similarity scores, driven by algorithmic randomness and Kolmogorov complexity, employ strangeness / typicality and p-values. We propose the application of the IIM framework to critical states prediction for complex physical systems; in particular, the prediction of cyclone genesis and intensification.