There has been a considerable shift in the way how software is built and delivered today. Most deployed software systems in modern times are created by (autonomous) distributed teams in heterogeneous environments making use of many artifacts, such as externally developed libraries, drawn from a variety of disparate sources. Stakeholders such as developers, managers, and clients across the software delivery value chain are interested in gaining insights such as how and why an artifact came to where it is, what other artifacts are related to it, and who else is using this. Software provenance encompasses the origins of artifacts, their evolution, and usage and is critical for comprehending, managing, decision-making, and analyzing software quality, processes, people, issues etc. In this paper, we propose an extensible framework based on standard provenance model specifications and blockchain technology for capturing, storing, exploring, and analyzing software provenance data. Our framework (i) enhances trustworthiness of provenance data (ii) uncovers non-trivial insights through inferences and reasoning, and (iii) enables interactive visualization of provenance insights. We demonstrate the utility of the proposed framework using open source project data.

In globally distributed software delivery, autonomous teams (crowd workers, vendors etc.) work together to build complex software. One of the key challenges in such an environment is to ensure integrity of software as it crosses the teams' boundaries. For example, during globally distributed software development, vulnerable open source components should not get introduced, or code should not be inadvertently changed. To track such essential characteristics of software, we propose a notion of a composite identity of software. ShIFt - the Software Identity Framework can construct sub-identities based on various elements of a software such as the code itself, third party components, run-time configurations etc. These sub-identities are then combined to generate a composite identity of a software. The key contributions of this paper are (i) an approach to create composite software identity and detect integrity issues between two instances of software, (ii) identification of the cause that led to integrity discrepancies, and (iii) prescription of remediation measures to maintain the integrity of software in the global delivery environment. We further use a Blockchain system to store and assess software identity, and consequently maintain software integrity.

Modern software delivery happens in a geographically distributed environment and resembles like a supply chain - consists of various participants, involves various phases, needs adherence to multiple regulations and needs to maintain artifacts' integrity throughout the delivery phases. This shift in software development brings along with it several challenges ranging from communication of information/knowledge, coordination and control of teams, activities adhering to goals and policies and artifacts adhering to quality, visibility, and management. With the dispersion of centralized control over software delivery to autonomous delivery organizations, the variety of processes and tools used turns transparency into opacity as autonomous teams use different software processes, tools, and metrics, leading to issues like ineffective compliance monitoring, friction prone coordination, and lack of provenance, and thereby trust. In this paper, we present a delivery governance framework based on distributed ledger technology that uses a notion of `software telemetry' to record data from disparate delivery partners and enables compliance monitoring and adherence, provenance and traceability, transparency, and thereby trust.

We have recently witnessed tremendous success of Machine Learning (ML) in practical applications. Computer vision, speech recognition and language translation have all seen a near human level performance. We expect, in the near future, most business applications will have some form of ML. However, testing such applications is extremely challenging and would be very expensive if we follow today's methodologies. In this work, we present an articulation of the challenges in testing ML based applications. We then present our solution approach, based on the concept of Metamorphic Testing, which aims to identify implementation bugs in ML based image classifiers. We have developed metamorphic relations for an application based on Support Vector Machine and a Deep Learning based application. Empirical validation showed that our approach was able to catch 71% of the implementation bugs in the ML applications.

Although most business processes change over time, contemporary process mining techniques tend to analyze these processes as if they are in a steady state. Processes may change suddenly or gradually. The drift may be periodic (e.g., because of seasonal influences) or one-of-a-kind (e.g., the effects of new legislation). For the process management, it is crucial to discover and understand such concept drifts in processes. This paper presents a generic framework and specific techniques to detect when a process changes and to localize the parts of the process that have changed. Different features are proposed to characterize relationships among activities. These features are used to discover differences between successive populations. The approach has been implemented as a plug-in of the ProM process mining framework and has been evaluated using both simulated event data exhibiting controlled concept drifts and real-life event data from a Dutch municipality.

More and more information about processes is recorded in the form of so-called 'event logs'. High-tech systems such as X-ray machines and high-end copiers provide their manufacturers and services organizations with detailed event data. Larger organizations record relevant business events for process improvement, auditing, and fraud detection. Traces in such event logs can be classified as desirable or undesirable (e.g., faulty or fraudulent behavior). In this paper, we present a comprehensive framework for discovering signatures that can be used to explain or predict the class of seen or unseen traces. These signatures are characteristic patterns that can be used to discriminate between desirable and undesirable behavior. As shown, these patterns can, for example, be used to predict remotely whether a particular component in an X-ray machine is broken or not. Moreover, the signatures also help to improve systems and organizational processes. Our framework for signature discovery is fully implemented in ProM and supports class labeling, feature extraction and selection, pattern discovery, pattern evaluation and cross-validation, reporting, and visualization. A real-life case study is used to demonstrate the applicability and scalability of the approach. Keywords: Discriminatory Patterns; Event Log; Process Mining; Signature Patterns

Process Mining refers to the extraction of process models from event logs. Real-life processes tend to be less structured and more flexible. Traditional process mining algorithms have problems dealing with such unstructured processes and generate spaghetti-like process models that are hard to comprehend. An approach to overcome this is to cluster process instances (a process instance is manifested as a trace and an event log corresponds to a multi-set of traces) such that each of the resulting clusters correspond to a coherent set of process instances that can be adequately represented by a process model. In this paper, we propose a context aware approach to trace clustering based on generic edit distance. It is well known that the generic edit distance framework is highly sensitive to the costs of edit operations. We define an automated approach to derive the costs of edit operations. The method proposed in this paper outperforms contemporary approaches to trace clustering in process mining. We evaluate the goodness of the formed clusters using established fitness and comprehensibility metrics defined in the context of process mining. The proposed approach is able to generate clusters such that the process models mined from the clustered traces show a high degree of fitness and comprehensibility when compared to contemporary approaches.

Microsoft Windows uses the notion of registry to store all configuration information. The registry entries have associations and dependencies. For example, the paths to executables may be relative to some home directories. The registry being designed with faster access as one of the objectives does not explicitly capture these relations. In this paper, we explore a representation that captures the dependencies more explicitly using shared and unifying variables. This representation, called mRegistry exploits the tree-structured hierarchical nature of the registry, is concept-based and obtained in multiple stages. mRegistry captures intra-block, inter-block and ancestor-children dependencies (all leaf entries of a parent key in a registry put together as an entity constitute a block thereby making the block as the only child of the parent). In addition, it learns the generalized concepts of dependencies in the form of rules. We show that mRegistry has several applications: fault diagnosis, prediction, comparison, compression etc.

Hybrid models combining the analytical (rule-based) and connectionist (artificial neural network (ANN)) paradigms are called knowledge based neural networks (KBNN). The knowledge based artificial neural network (KBANN) is one such model that makes use of the domain theory represented as propositional rules and training examples. In this article, we analyze the performance of KBANN and suggest some ideas of improving its capabilities. A study on the effect of inductive bias on KBANN, use of adaptive learning algorithms instead of the standard backpropagation to improve the training times and the use of regularization methods for improving generalization performance is presented. It is shown that for better performance, the initial weight assignment to links obtained by domain theory varies with the domain and the performance of KBANN is improved by using regularization methods like the weight decay along with Rprop training algorithm instead of the standard backpropagation.