Recent advancement in deep-neural network performance led to the development of new state-of-the-art approaches in numerous areas. However, the black-box nature of neural networks often prohibits their use in areas where model explainability and model transparency are crucial. Over the years, researchers proposed many algorithms to aid neural network understanding and provide additional information to the human expert. One of the most popular methods being Layer-Wise Relevance Propagation (LRP). This method assigns local relevance based on the pixel-wise decomposition of nonlinear classifiers. With the rise of attribution method research, there has emerged a pressing need to assess and evaluate their performance. Numerous metrics have been proposed, each assessing an individual property of attribution methods such as faithfulness, robustness, or localization. Unfortunately, no single metric is deemed optimal for every case, and researchers often use several metrics to test the quality of the attribution maps. In this work, we address the shortcomings of the current LRP formulations and introduce a novel method for determining the relevance of input neurons through layer-wise relevance propagation. Furthermore, we apply this approach to the recently developed Vision Transformer architecture and evaluate its performance against existing methods on two image classification datasets, namely ImageNet and PascalVOC. Our results clearly demonstrate the advantage of our proposed method. Furthermore, we discuss the insufficiencies of current evaluation metrics for attribution-based explainability and propose a new evaluation metric that combines the notions of faithfulness, robustness, and contrastiveness. We utilize this new metric to evaluate the performance of various attribution-based methods. Our code is available at: https://github.com/davor10105/relative-absolute-magnitude-propagation

As cloud computing becomes the prevailing aspect of software engineering, paradigms such as Service-Based Systems (SBSs) or Software as a Service (SaaS) are coming into focus. They are based on cloud services responding to numerous client requests. Selecting the actual service instance for request can be an issue, if requirements for multiple Quality of Service (QoS) attributes need to be satisfied for many users simultaneously. The problem becomes more complex if we take into account the compositeness of users applications, consisting of many tasks, where QoS properties are calculated over the whole composition. The existing approaches for this problem lack either efficiency or generality. In this paper, we propose a fast heuristic method for multi-criteria service selection, designed for multi-user composite workflows with the goal of satisfying all, or as many as possible, of the given QoS requirements. The proposed method reduces the problem to several independent transportation problems, using a global-aware utility cost based on expected compositional QoS, and iterative solution improvements. Apart from being more general than the existing approaches, the proposed method turns out to be more efficient than the alternatives (up to 5x faster), as shown by extensive experiments covering both special and more general cases.

As SOA gains more traction through various implementations, building reliable service compositions remains one of the principal research concerns. Widely researched reliability assurance methods, often rely on applying redundancy or complex optimization strategies that can make them less applicable when it comes to designing service compositions on a larger scale. To address this issue, we propose a design time reliability improvement method that enables selective service composition improvements by focusing on the most reliability-critical workflow components, named weak points. With the aim of detecting most significant weak points, we introduce a method based on a suite of recommendation algorithms that leverage a belief network reliability model. The method is made scalable by using heuristic algorithms that achieve better computational performance at the cost of recommendation accuracy. Although less accurate heuristic algorithms on average require more improvement steps, they can achieve better overall performance in cases when the additional step-wise overhead of applying improvements is low. We confirm the soundness of the proposed solution by performing experiments on data sets of randomly generated service compositions.

While constructing QoS-aware composite work-flows based on service oriented systems, it is necessary to assess nonfunctional properties of potential service selection candidates. In this paper, we present CLUS, a model for reliability prediction of atomic web services that estimates the reliability for an ongoing service invocation based on the data assembled from previous invocations. With the aim to improve the accuracy of the current state-of-the-art prediction models, we incorporate user-service-, and environment-specific parameters of the invocation context. To reduce the scalability issues present in the state-of-the-art approaches, we aggregate the past invocation data using K-means clustering algorithm. In order to evaluate different quality aspects of our model, we conducted experiments on services deployed in different regions of the Amazon cloud. The evaluation results confirm that our model produces more scalable and accurate predictions when compared to the current state-of-the-art approaches.

The modern information systems on the Internet are often implemented as composite services built from multiple atomic services. These atomic services have their interfaces publicly available while their inner structure is unknown. The quality of the composite service is dependent on both the availability of each atomic service and their appropriate orchestration. In this paper, we present LUCS, a formal model for predicting the availability of atomic web services that enhances the current state-of-the-art models used in service recommendation systems. LUCS estimates the service availability for an ongoing request by considering its similarity to prior requests according to the following dimensions: the user's and service's geographic location, the service load, and the service's computational requirements. In order to evaluate our model, we conducted experiments on services deployed in different regions of the Amazon cloud. For each service, we varied the geographic origin of its incoming requests as well as the request frequency. The evaluation results suggest that our model significantly improves availability prediction when all of the LUCS input parameters are available, reducing the prediction error by 71 percent compared to the current state-of-the-art.

Contemporary web applications are often designed as composite services built by coordinating atomic services with the aim of providing the appropriate functionality. Although functional properties of each atomic service assure correct functionality of the entire application, nonfunctional properties such as availability, reliability, or security might significantly influence the user-perceived quality of the application. In this paper, we present CLUS, a model for reliability prediction of atomic web services that improves state-of-the-art approaches used in modern recommendation systems. CLUS predicts the reliability for the ongoing service invocation using the data collected from previous invocations. We improve the accuracy of the current state-of-the-art prediction models by considering user-, service- and environment-specific parameters of the invocation context. To address the computational performance related to scalability issues, we aggregate the available previous invocation data using K-means clustering algorithm. We evaluated our model by conducting experiments on services deployed in different regions of the Amazon cloud. The evaluation results suggest that our model improves both performance and accuracy of the prediction when compared to the current state-of-the-art models.