When seeking to maximise the authenticity of assessment in programming courses it makes sense to provide students with practical programming problems to solve in an environment that is close to real software development practice, i.e., online, open book, and using their typical development environment. This creates an assessment environment that should afford students sufficient opportunities to evidence what they have learned, but also creates practical challenges in terms of academic integrity, flexibility in the automated grading process, and assumptions surrounding how the student may attempt to solve the problems both in terms of correct and incorrect solutions. In this experience report, we outline two independently observed cohorts of students sitting the same Java programming exam, with different weights, over three years. This is undertaken as a reflective exercise in order to derive a series of recommendations and retrospectively obvious pitfalls to act as guidance for educators considering online programming exams for large (i.e. n > 150) introductory programming courses. After discussing our assessment methodology, we provide 4 high-level observations and centre a set of recommendations around these to aid practitioners in their assessment design.

Abstract Deep Learning models based on convolutional neural networks are known to be uncalibrated, that is, they are either overconfident or underconfident in their predictions. Safety‐critical applications of neural networks, however, require models to be well‐calibrated, and there are various methods in the literature to increase model performance and calibration. Subnetwork ensembling is based on the over‐parametrization of modern neural networks by fitting several subnetworks into a single network to take advantage of ensembling them without additional computational costs. Data augmentation methods have also been shown to enhance model performance in terms of accuracy and calibration. However, ensembling and data augmentation seem orthogonal to each other, and the total effect of combining these two methods is not well‐known; the literature in fact is inconsistent. Through an extensive set of empirical experiments, we show that combining subnetwork ensemble methods with data augmentation methods does not degrade model calibration.

In this paper, we analyze 32,000 Java programming assessments submitted to the CodeRunner platform by introductory programming students. We identify common sequences of errors and link these patterns with observations of code that compiles but still contains logical errors. We establish reoccurring errors and common error pathways through a Markov chain analysis of attempts and use Association Rule Mining to link common error patterns with logical errors that occur later in the attempts. As a result, we provide a number of suggestions for instructors of introductory Java courses regarding how to potentially navigate these issues in their teaching practice and discuss possible pedagogical interventions on the basis of our observations.

Research on Fairness and Bias Mitigation in Machine Learning often uses a set of reference datasets for the design and evaluation of novel approaches or definitions. While these datasets are well structured and useful for the comparison of various approaches, they do not reflect that datasets commonly used in real-world applications can have missing values. When such missing values are encountered, the use of imputation strategies is commonplace. However, as imputation strategies potentially alter the distribution of data they can also affect the performance, and potentially the fairness, of the resulting predictions, a topic not yet well understood in the fairness literature. In this article, we investigate the impact of different imputation strategies on classical performance and fairness in classification settings. We find that the selected imputation strategy, along with other factors including the type of classification algorithm, can significantly affect performance and fairness outcomes. The results of our experiments indicate that the choice of imputation strategy is an important factor when considering fairness in Machine Learning. We also provide some insights and guidance for researchers to help navigate imputation approaches for fairness.

The goal of this paper is to ascertain with what accuracy the direction of Bitcoin price in USD can be predicted. The price data is sourced from the Bitcoin Price Index. The task is achieved with varying degrees of success through the implementation of a Bayesian optimised recurrent neural network (RNN) and a Long Short Term Memory (LSTM) network. The LSTM achieves the highest classification accuracy of 52% and a RMSE of 8%. The popular ARIMA model for time series forecasting is implemented as a comparison to the deep learning models. As expected, the non-linear deep learning methods outperform the ARIMA forecast which performs poorly. Finally, both deep learning models are benchmarked on both a GPU and a CPU with the training time on the GPU outperforming the CPU implementation by 67.7%.

Today, auto-scaling solutions are still largely reactive and are based on the load measured on existing nodes nearing a threshold or traffic forecast information provided in advance of a scheduled event. Despite these advancements, events which cause a flash flood of web traffic do not always benefit from this approach to auto-scaling because the latency of provisioning new nodes is not sufficient to prevent resource saturation. This paper proposes a novel approach to preemptive compute scaling based on the buzz of specific social network hashtags. We argue that improved scaling latency and service availability can be achieved for web services that receive intermittent as well as unexpected traffic load patterns. By combining social network monitoring, with auto-scaling frameworks, this approach can reduce the operational expense impact of over- and under-provisioning as well as the business costs of the latter. We demonstrate our approach using the AWS application suite, and demonstrate how our approach expedites the auto-scaling strategy for our use case: public transportation web sites.

Social media and the resulting tidal wave of the available data have changed the ways and methods researchers analyze communities at scale. But the full potential for social scientists (and others) is not yet achieved. Despite the popularity of social media analysis in the past decade, few researchers invest in cross-platform analyses. This is a major oversight as a majority of online social media users have multiple social media accounts. Missing are the models and tools necessary to undertake analysis at scale across multiple platforms. Especially promising in support of cross-platform analysis is the mixed-method approach (e.g., qualitative and quantitative methods) in order to better understand how users and society interact online. This special issue “Following User Pathways” addresses methodological, analytical, conceptual, and technological challenges and opportunities of cross-platform analysis in social media ecosystems.

Summary We are in the midst of a scientific data explosion in which the rate of data growth is rapidly increasing. While large‐scale research projects have developed sophisticated data distribution networks to share their data with researchers globally, there is no such support for the many millions of research projects generating data of interest to much smaller audiences (as exemplified by the long tail scientist). In data‐oriented research, every aspect of the research process is influenced by data access. However, sharing and accessing data efficiently as well as lowering access barriers are difficult. In the absence of dedicated large‐scale storage, many have noted that there is an enormous storage capacity available via connected peers, none more so than the storage resources of many research groups. With widespread usage of the content delivery network model for disseminating web content, we believe a similar model can be applied to distributing, sharing, and accessing long tail research data in an e‐Science context. We describe the vision and architecture of a social content delivery network – a model that leverages the social networks of researchers to automatically share and replicate data on peers' resources based upon shared interests and trust. Using this model, we describe a simulator and investigate how aspects such as user activity, geographic distribution, trust, and replica selection algorithms affect data access and storage performance. From these results, we show that socially informed replication strategies are comparable with more general strategies in terms of availability and outperform them in terms of spatial efficiency. Copyright © 2016 John Wiley & Sons, Ltd.

Research data is experiencing a seemingly endless increase in both volume and production rate. At the same time, efficiently transferring, storing, and analyzing large scale research data have become major research foci. In this paper, we expand on our approach to sharing data for e-Science: a Social Content Delivery Network (S-CDN). A S-CDN leverages the social networks of researchers to automatically share data and place replicas on peers' resources based upon the premises of trust and interest in shared data. We denote a consumer of shared data as a data follower, similar to the notion of Twitter followers, except we add the element of bilateral authorization to capture a notion of trust. We describe a prototypical implementation for a S-CDN that captures an efficient asynchronous transfer mechanism for data management and replication. In addition, we study via simulation the interplay of user behavior with different replication strategies that capture social as well as more general premises for data sharing. Our results illustrate the opportunities and pitfalls of various replication and data access management strategies. Specifically, we show that socially-informed replication strategies are competitive with more general strategies in terms of availability, and outperform them in terms of spatial efficiency.

Social network platforms have rapidly changed the way that people communicate and interact. They have enabled the establishment of, and participation in, digital communities as well as the representation, documentation and exploration of social relationships. We believe that as `apps' become more sophisticated, it will become easier for users to share their own services, resources and data via social networks. To substantiate this, we present a social compute cloud where the provisioning of cloud infrastructure occurs through “friend” relationships. In a social compute cloud, resource owners offer virtualized containers on their personal computer(s) or smart device(s) to their social network. However, as users may have complex preference structures concerning with whom they do or do not wish to share their resources, we investigate, via simulation, how resources can be effectively allocated within a social community offering resources on a best effort basis. In the assessment of social resource allocation, we consider welfare, allocation fairness, and algorithmic runtime. The key findings of this work illustrate how social networks can be leveraged in the construction of cloud computing infrastructures and how resources can be allocated in the presence of user sharing preferences.

Increases in the size of research data and the move towards citizen science, in which everyday users contribute data and analyses, have resulted in a research data deluge. Researchers must now carefully determine how to store, transfer and analyze "Big Data" in collaborative environments. This task is even more complicated when considering budget and locality constraints on data storage and access. In this paper we investigate the potential to construct a Social Content Delivery Network (S-CDN) based upon the social networks that exist between researchers. The S-CDN model builds upon the incentives of collaborative researchers within a given scientific community to address their data challenges collaboratively and in proven trusted settings. In this paper we present a prototype implementation of a S-CDN and investigate the performance of the data transfer mechanisms (using Glob us Online) and the potential cost advantages of this approach.

Social collaboration scenarios, such as sharing resources between friends, are becoming increasingly prevalent in recent years. An example of this new paradigm is Social Cloud Computing, which aims at leveraging existing digital relationships within social networks for the exchange of resources among users and user communities. Due to their complexity, such platforms and systems have to be carefully designed and engineered to suit their purpose. In this paper, we propose a general-purpose simulation tool to help in the design and analysis of Social Collaboration Platforms, and discuss potential use cases and the architecture of the simulator. To show the usefulness of the simulator, we present a simple use case in which we study the effects of an incentive scheme on the system and its user community.

There is an overriding interest in measuring the well-being of communities and institutions: healthy (flourishing) individuals and groups perform “better” than those that are not. Capturing the facets of well-being is, however, not straightforward: it contains personal information with sometimes uncomfortable self-realizations associated to it. Yet, the benefit of such data is the ability to observe and react to imbalances of a community, i.e. it can facilitate community management. Due to its personal nature, the observation of well-being needs to leverage carefully considered constructs. To have a comprehensive look at the concept of individual well-being, we propose a gamified frame of reference within a social network platform to lower traditional entrance barriers for data collection and encourage continued usage. In our setting, participants can record aspects of their well-being as a part of their “normal” social network activities, as well as view trends of themselves and their community. To evaluate the feasibility of our approach, we present the results of an initial study conducted via Facebook.

SUMMARY Many research institutions and Universities own computational capacity that is not effectively utilized, thereby providing an opportunity for such institutions to use such capacity to offer Cloud services (to both internal and external users). However, the unreliability and unpredictability of these resources mean that their use in the context of a Service Level Agreement (SLA) is high risk, leading to a reduction in reputation as well as economic penalties in case of SLA violation. We propose a methodology that addresses the issues of unreliability and unpredictability such that Cloud software services could be hosted upon volunteered resources. To enable the harnessing of these resources we rely on autonomic fault management techniques that allow such systems to independently adapt the resources they use based upon their perception of individual resource reliability. Using our approach we were able to scale out the backend infrastructure of the Cloud service elastically (min 30thinspaces per worker), opportunistically and autonomically. We address two key questions in this article: can a campus volunteer infrastructure be used in Cloud provisioning? What measures are necessary in order to ensure reliability at the resource level? Copyright © 2011 John Wiley & Sons, Ltd.

Online relationships in social networks are often based on real world relationships and can therefore be used to infer a level of trust between users. We propose leveraging these relationships to form a dynamic "Social Cloud,” thereby enabling users to share heterogeneous resources within the context of a social network. In addition, the inherent socially corrective mechanisms (incentives, disincentives) can be used to enable a cloud-based framework for long term sharing with lower privacy concerns and security overheads than are present in traditional cloud environments. Due to the unique nature of the Social Cloud, a social market place is proposed as a means of regulating sharing. The social market is novel, as it uses both social and economic protocols to facilitate trading. This paper defines Social Cloud computing, outlining various aspects of Social Clouds, and demonstrates the approach using a social storage cloud implementation in Facebook.

Combining the strengths of the Cloud Computing and Social Network paradigms, the vision of Social Clouds aims to provide a resource sharing mechanism where participants dynamically share and trade resources on the premise of the relationships encoded in a social network. By building upon existing relationships in social networks and the inherent trust that accompanies these relationships, a Social Cloud is able to address one of the most cited obstacles in the adoption of current cloud solutions, the missing trust between Cloud service providers and users. However, as with other computing approaches relying on user participation, incentivisation of potential and existing participants is crucial for the success and sustainability of a Social Cloud. Therefore, an incentive engineering approach is needed and will be discussed in this paper considering all phases of user participation in a Social Cloud in order to provide proper incentives for active user participation and desired user behavior.

Currently, the Cloud landscape is a fragmented, static and shapeless market that hinders the paradigm's ability to fulfil its promise of ubiquitous computing on tap and as a commodity. In this paper, we present our vision of an autonomic self-aware Cloud market platform, and argue that autonomic market platforms for Clouds can step up to the challenge of today's status quo. As our first steps towards achieving this vision, we present a market monitoring methodology, which includes a series of realistic market goals, sets of extractable metrics from a market platform and how to map (i.e. combine and transform) metrics to access goal performance such that autonomic adaption of the market could be undertaken. We have extended a known market simulator for distributed infrastructures (GridSim) with relevant sensors. To demonstrate the usefulness of our approach, we simulate a sudden cease in demand for goods in our market platform.

With the increasingly ubiquitous nature of Social networks and Cloud computing, users are starting to explore new ways to interact with, and exploit these developing paradigms. Social networks are used to reflect real world relationships that allow users to share information and form connections between one another, essentially creating dynamic Virtual Organizations. We propose leveraging the pre-established trust formed through friend relationships within a Social network to form a dynamic “Social Cloud”, enabling friends to share resources within the context of a Social network. We believe that combining trust relationships with suitable incentive mechanisms (through financial payments or bartering) could provide much more sustainable resource sharing mechanisms. This paper outlines our vision of, and experiences with, creating a Social Storage Cloud, looking specifically at possible market mechanisms that could be used to create a dynamic Cloud infrastructure in a Social network environment.

Abstract A system implemented in MATLAB is described, which may be deployed over a Campus Grid utilizing the Condor job management system. Our approach can re‐distribute jobs as node availability changes. The architecture of the system, its components, and their deployment across the Cardiff University Campus Grid (consisting of 2500 machines) are presented. Challenges in image processing applications that can be deployed over such infrastructure are presented, along with performance results that demonstrate the use of our system alongside a standard Condor deployment, demonstrating a significant increase in throughput using our approach ‡ . Copyright © 2008 John Wiley & Sons, Ltd.

Diffusion tensor imaging (DTI) provides insight into the white matter of the human brain, which is affected by schizophrenia. By comparing a patient group to a control group, the DTI-images are on average expected to be different for white matter regions. Principal component analysis (PCA) and linear discriminant analysis (LDA) are used to classify the groups. In this work, the number of principal components is optimised for obtaining the minimal classification error. A robust estimate of this error is computed in a cross-validation framework, using different compositions of the data into a training and a testing set Previously, sequential runs were performed in MATLAB, resulting in long execution times. In this paper we describe an experiment where this application was run on a grid with minimal modifications and user effort. We have adopted a service-based approach that autonomously launches image analysis services onto a campus-wide Condor pool comprising of volunteer resources. This allows high throughput analysis of our data in a dynamic resource pool. The challenge in adopting such an approach comes from the nature of the resources, which change randomly with time and thus require fault tolerance. Through this approach we have reduced the computation time of each dataset from 90 minutes to less than 10. A minimal classification error of 22% was obtained, using 15 principal components.

Interactive image processing is an important requirement in many industrial applications, such as the inspection of industrial parts within a manufacturing environment, or the processing of images from surveillance cameras. Being able to achieve this quickly and accurately is often essential for the success of such industrial applications. A service-based approach that autonomously launches Image Analysis Services (accessible through a Central Service Manager) onto spare network resources through a Condor system is presented. This allows high throughput analysis of these images in a dynamic resource pool. The Central Service Manager reacts to new tasks submitted to the Image Analysis Services and is able to add new service instances to manage these tasks dynamically. Each service instance here corresponds to a computational resource that is able to execute image processing algorithms. New service instances may be requested by the Central Service Manager from the Condor system, based on the number of tasks that need to be processed. This enables entire image repositories to be acted upon interactively and in parallel, as opposed to the analysis of single images individually. The approach is demonstrated through a campus-wide test bed utilising a Condor system with 90 machines.