To address the challenge of deploying large AI models on resource-constrained edge devices, we propose SPICE, a systematic three-stage pipeline for model selection, optimization, and deployment. For a real-world animal ecology task requiring on-device image segmentation, we introduce our novel TaLK-Structure Pruning algorithm. TaLK intelligently allocates layer-wise pruning quotas based on first-order Taylor sensitivity and computational cost, preventing layer collapse while maximizing efficiency. We achieved significant reductions in model size, parameter count and computational overhead with pruning ratios ranging from 0.5 to 0.9. Knowledge distillation–assisted fine-tuning recovered the performance of the compressed models, achieving an mIoU of 0.54±0.07 and an accuracy of 63±8% (at a pruning ratio of 0.5) over our downstream task. Deployed on a Raspberry Pi, the optimized models meet strict operational constraints, demonstrating low latency (wall time < 30s), minimal energy consumption of ∼4±2W (for multiple inferences within a 60-second window) ensuring high inference confidence across all test cases. Our end-to-end pipeline provides a simple framework for effectively deploying complex vision models on niche edge devices. Alongside, our pipeline is fully reproducible, open-source, and available on GitHub at https://github.com/sdasosu/SPICE

This paper presents a case study on applying machine learning to the real-world problem of making data-driven player and team predictions for the National Football League (NFL). It is motivated by the legalization of sports betting and interest in sports analytics by franchises and fans. We present a reproducible methodology to produce accurate predictions of margins of victory, player rush yards, and player receiving yards that exceed the baseline established by previous researchers, along with publicly available code and datasets that can be found at https://code.osu.edu/cliffel.11/sportspredictions.

We propose TaxaDiffusion, a taxonomy-informed training framework for diffusion models to generate fine-grained animal images with high morphological and identity accuracy. Unlike standard approaches that treat each species as an independent category, TaxaDiffusion incorporates domain knowledge that many species exhibit strong visual similarities, with distinctions often residing in subtle variations of shape, pattern, and color. To exploit these relationships, TaxaDiffusion progressively trains conditioned diffusion models across different taxonomic levels -- starting from broad classifications such as Class and Order, refining through Family and Genus, and ultimately distinguishing at the Species level. This hierarchical learning strategy first captures coarse-grained morphological traits shared by species with common ancestors, facilitating knowledge transfer before refining fine-grained differences for species-level distinction. As a result, TaxaDiffusion enables accurate generation even with limited training samples per species. Extensive experiments on three fine-grained animal datasets demonstrate that outperforms existing approaches, achieving superior fidelity in fine-grained animal image generation. Project page: https://amink8.github.io/TaxaDiffusion/

We present a novel frequency-based Self-Supervised Learning (SSL) approach that significantly enhances its efficacy for pre-training. Prior work in this direction masks out pre-defined frequencies in the input image and employs a reconstruction loss to pre-train the model. While achieving promising results, such an implementation has two fundamental limitations as identified in our paper. First, using pre-defined frequencies overlooks the variability of image frequency responses. Second, pre-trained with frequency-filtered images, the resulting model needs relatively more data to adapt to naturally looking images during fine-tuning. To address these drawbacks, we propose FOurier transform compression with seLf-Knowledge distillation (FOLK), integrating two dedicated ideas. First, inspired by image compression, we adaptively select the masked-out frequencies based on image frequency responses, creating more suitable SSL tasks for pre-training. Second, we employ a two-branch framework empowered by knowledge distillation, enabling the model to take both the filtered and original images as input, largely reducing the burden of downstream tasks. Our experimental results demonstrate the effectiveness of FOLK in achieving competitive performance to many state-of-the-art SSL methods across various downstream tasks, including image classification, few-shot learning, and semantic segmentation.

This paper reports on the creation and composition of a new corpus of children's speech, the Ohio Child Speech Corpus, which is publicly available on the Talkbank-CHILDES website. The audio corpus contains speech samples from 303 children ranging in age from 4 – 9 years old, all of whom participated in a seven-task elicitation protocol conducted in a science museum lab. In addition, an interactive social robot controlled by the researchers joined the sessions for approximately 60% of the children, and the corpus itself was collected in the peri‑pandemic period. Two analyses are reported that highlighted these last two features. One set of analyses found that the children spoke significantly more in the presence of the robot relative to its absence, but no effects of speech complexity (as measured by MLU) were found for the robot's presence. Another set of analyses compared children tested immediately post-pandemic to children tested a year later on two school-readiness tasks, an Alphabet task and a Reading Passages task. This analysis showed no negative impact on these tasks for our highly-educated sample of children just coming off of the pandemic relative to those tested later. These analyses demonstrate just two possible types of questions that this corpus could be used to investigate.

Graph-level representations (and clustering/classification based on these representations) are required in a variety of applications. Examples include identifying malicious network traffic, prediction of protein properties, and many others. Often, data has to stay in isolated local systems due to a variety of considerations like privacy concerns, lack of trust between the parties, regulations, or simply because the data is too large to be shared sufficiently quickly. This points to the need for federated learning for graph-level representations, a topic that has not been explored much, especially in an unsupervised setting.Addressing this problem, this paper presents a new framework we refer to as Federated Contrastive Learning of Graph-level Representations (FCLG). Our approach builds on contrastive learning. However, what is unique is that we apply contrastive learning at two levels. The first application is for local unsupervised learning of graph representations. The second level is to address the challenge associated with data distribution variation (i.e. the "Non-IID issue") when combining local models. Through extensive experiments on the downstream task of graph-level clustering, we demonstrate FCLG outperforms baselines with significant margins.

Edge deployments perform complex deep learning inference and analysis in the wild in highly resource constrained environment. They are positioned everywhere from our largest cities to the bottom of our oceans, and often necessitate significant financial resources and labor to create and deploy. These properties make correctness of edge deployments simultaneously extremely important and difficult to verify a priori. In the past decade, a series of IoT and cloud testbeds have emerged to facilitate this testing. They provide users with access to resources and, less often, sensors that can be used to emulate workloads before deployment. While developers can use these resources to verify the correctness of their configurations, often users would like to “right-size” their deployments - that is, to find a minimal resource configuration that guarantees correctness - to decrease cost and prevent over-provisioning. The current suite of cloud and IoT testbeds does not provide this capability. We present Righteous, an automatic deployment right-sizing tool for edge deployments. Righteous treats configuration as a hyperparameter optimization problem, testing hyperparameter combinations to find a near-optimal configuration as quickly as possible. Righteous uses a new optimization algorithm, informed Pareto Simulated Annealing (iPSA) to find near-optimal configurations faster than other leading approaches. We use Righteous in conjunction with the PROWESS testbed to optimize a drone swarm deployment workload. Our results demonstrate that Righteous configurations use up to 3.5X less resources than those identified by leading hyperparameter tuning and resource allocation techniques, and does so up to 76.3X faster.

Standard modern machine-learning-based imaging methods have faced challenges in medical applications due to the high cost of dataset construction and, thereby, the limited labeled training data available. Additionally, upon deployment, these methods are usually used to process a large volume of data on a daily basis, imposing a high maintenance cost on medical facilities. In this paper, we introduce a new neural network architecture, termed LoGoNet, with a tailored self-supervised learning (SSL) method to mitigate such challenges. LoGoNet integrates a novel feature extractor within a U-shaped architecture, leveraging Large Kernel Attention (LKA) and a dual encoding strategy to capture both long-range and short-range feature dependencies adeptly. This is in contrast to existing methods that rely on increasing network capacity to enhance feature extraction. This combination of novel techniques in our model is especially beneficial in medical image segmentation, given the difficulty of learning intricate and often irregular body organ shapes, such as the spleen. Complementary, we propose a novel SSL method tailored for 3D images to compensate for the lack of large labeled datasets. The method combines masking and contrastive learning techniques within a multi-task learning framework and is compatible with both Vision Transformer (ViT) and CNN-based models. We demonstrate the efficacy of our methods in numerous tasks across two standard datasets (i.e., BTCV and MSD). Benchmark comparisons with eight state-of-the-art models highlight LoGoNet's superior performance in both inference time and accuracy.

LU factorization for sparse matrices is an important computing step for many engineering and scientific problems such as circuit simulation. There have been many efforts toward parallelizing and scaling this algorithm, which include the recent efforts targeting the GPUs. However, it is still challenging to deploy a complete sparse LU factorization workflow on a GPU due to high memory requirements and data dependencies. In this paper, we propose the first complete GPU solution for sparse LU factorization. To achieve this goal, we propose an out-of-core implementation of the symbolic execution phase, thus removing the bottleneck due to large intermediate data structures. Next, we propose a dynamic parallelism implementation of Kahn's algorithm for topological sort on the GPUs. Finally, for the numeric factorization phase, we increase the parallelism degree by removing the memory limits for large matrices as compared to the existing implementation approaches. Experimental results show that compared with an implementation modified from GLU 3.0, our out-of-core version achieves speedups of 1.13--32.65X. Further, our out-of-core implementation achieves a speedup of 1.2--2.2 over an optimized unified memory implementation on the GPU. Finally, we show that the optimizations we introduce for numeric factorization turn out to be effective.

Despite the popularity of in-situ analytics in scientific computing, there is only limited work to date on in-situ analytics for simulations running on GPUs. Notably, two unaddressed challenges are 1) performing memory-efficient in-situ analysis on accelerators and 2)automatically choosing the processing resources and suitable data representation for a given query and platform. This paper addresses both problems. First, GAP makes several new contributions toward making bitmap indices suitable, effective, and efficient as a compressed data summary structure for the GPUs - this includes introducing a layout structure, a method for generating multi-attribute bitmaps, and novel techniques for bitmap-based processing of major operators that comprise complex data analytics. Second, this paper presents a performance modeling methodology, aiming to predict the placement (i.e., CPU or GPU) and the data representation choice (summarization or original) that yield the best performance on a given configuration. Our extensive evaluation of complex in-situ queries and real-world simulations shows that with our methods, analytics on GPU using bitmaps almost always outperforms other options, and the GAP performance model predicts the optimal placement and data representation for most scenarios.

Interactions between (social) entities can be frequently represented by an attributed graph, and node clustering in such graphs has received much attention lately. Multiple efforts have successfully applied Graph Convolutional Networks (GCN), though with some limits on accuracy as GCNs have been shown to suffer from over-smoothing issues. Though other methods (particularly those based on Laplacian Smoothing) have reported better accuracy, a fundamental limitation of all the work is a lack of scalability. This paper addresses this open problem by relating the Laplacian smoothing to the Generalized PageRank, and applying a random-walk based algorithm as a scalable graph filter. This forms the basis for our scalable deep clustering algorithm, RwSL. Using 6 real-world datasets and 6 clustering metrics, we show that RwSL achieved improved results over several recent baselines. Most notably, by demonstrating execution of RwSL on a graph with 1.8 billion edges using only a single GPU. We show that RwSL can continue to scale, unlike other existing deep clustering frameworks.

In this paper, we present a nuanced exploration of deep-learning techniques (DL) for extracting structural infor-mation from document images generated from the digitization of business processes. The driving example presented is the extraction of columns and rows of tables using a simple stacked CNN architecture and a combination of ensemble techniques. In addition, the component models of the ensemble are diversified by training on datasets created by applying a “semantics-preserving” transformation on the base dataset. This “semantics-preserving” transformation also aims to alleviate hard recognition in certain noisy images commonly encountered in practice. Our experiments demonstrate how DL techniques can be applied and innovatively combined to measurably improve the accuracy of structure extraction.

Road construction projects maintain transportation infrastructures, and range from short- to long-term. Deciding what the next construction project is and when it is to be scheduled is traditionally done through inspection by humans using special equipment, which is costly and difficult to scale. An alternative is the use of computational approaches that integrate and analyze multiple types of past and present spatiotemporal data to predict location and time of future road constructions. This paper reports on such an approach, one that uses a deep-neural-network-based model to predict future constructions, based on a heterogeneous dataset consisting of construction, weather, map and road-network data. We also report on how we addressed the lack of adequate publicly available data - by building a large scale dataset named "US-Constructions", that includes 6.2 million road constructions augmented by a variety of spatiotemporal attributes and road-network features, collected in the contiguous United States (US) between 2016 and 2021. Extensive experiments on several major cities in the US show the applicability of our approach to accurately predict future constructions.

All Pairs Shortest Path (APSP) is one of the graph problems where the output size is significantly larger than the input size. This paper examines the issues in scaling GPU implementations for this problem beyond the memory limits. Because the existing (in-core) methods offer a complex trade-off between the overall computation complexity and the available parallelism, choosing the best out-of-core version for a given matrix is challenging. We develop three efficient out-of-core implementations, which are based on the blocked Floyd-Warshall algorithm, Johnson's algorithm, and the boundary algorithm, respectively. Next, we develop a methodology to select the best implementation for a given graph. Experimental results show that compared with an efficient multi-core APSP implementation, the out-of-core version achieves speedups of 8.22 to 12.40 for graphs with a small separator, and speedups of 2.23 to 2.79 for other sparse graphs, and our models can select the best implementation in most cases.

Automated generation of commercial tweets has become a useful and important tool in the use of social media for marketing and advertising. In this context, paraphrase generation has emerged as an important problem. This type of paraphrase generation has the unique requirement of requiring certain elements to be kept in the result, such as the product name or the promotion details. To address this need, we propose a Constraint-Embedded Language Modeling (CELM) framework, in which hard constraints are embedded in the text content and learned through a language model. This embedding helps the model learn not only paraphrase generation but also constraints in the content of the paraphrase specific to commercial tweets. In addition, we apply knowledge learned from a general domain to the generation task of commercial tweets. Our model is shown to outperform general paraphrase generation models as well as the state-of-the-art CopyNet model, in terms of paraphrase similarity, diversity, and the ability to conform to hard constraints.

Learning a vector representation of locations that reflect human mobility patterns is useful for various tasks, including location recommendation, city planning, urban analysis, and even understanding the neighborhood effects on individuals' health and well-being. Existing approaches that model and learn such representations either do not scale or require significant resources to scale. They often need the entire data to be loaded in memory along with the intermediate data representation (typically a co-location graph) and are usually not feasible to execute on low-resource embedding systems such as edge devices. The research question we seek to address in this article is, can one develop efficient federated learning models for location representation learning such that the training and the subsequent updates of the model can occur on edge devices? We present a simple yet novel model called LocationTrails for learning efficient location embeddings to address this question. We show that our proposed model can be trained under the federated learning paradigm and can, therefore, ensure that the model can be trained in a distributed fashion without centralizing locations visited by all users, thereby mitigating some risks to privacy. We evaluate the performance of LocationTrails on five real-world human mobility datasets drawn from two use cases (four of them from driving trajectory data obtained from a national insurance agency; and one of them from a unique study of adolescent mobility patterns in an urban setting). We compare our proposed LocationTrails model against the strong baselines from the network representation learning field. We show the efficacy of LocationTrails in terms of better embedding quality generation, memory consumption, and execution time. To the best of our knowledge, the federated LocationTrails model is the first model that can generate efficient location embeddings without requiring the complete data to be loaded on a central server.

Multiplication of two sparse matrices (SpGEMM) is a popular kernel behind many numerical solvers, and also features in implementing many common graph algorithms. Though many recent research efforts have focused on implementing SpGEMM efficiently on a single GPU, none of the existing work has considered the case where the memory requirements exceed the size of GPU memory. Similarly, the use of the aggregate computing power of CPU and GPU has also not been addressed for those large matrices. In this paper, we present a framework for scaling SpGEMM computations for matrices that do not fit into GPU memory. We address how the computation and data can be partitioned across kernel executions on GPUs. An important emphasis in our work is overlapping data movement and computation. We achieve this by addressing many challenges, such as avoiding dynamic memory allocations, and re-scheduling data transfers with the computation of chunks. We extend our framework to make efficient use of both GPU and CPU, by developing an efficient work distribution strategy. Our evaluation on 9 large matrices shows that our out-of-core GPU implementation achieves 1.98-3.03X speedups over a state-of-the-art multi-core CPU implementation, our hybrid implementation further achieves speedups up to 3.74x, and that our design choices are directly contributing towards achieving this performance.

Heterogeneous, dense computing architectures consisting of several accelerators, such as GPUs, attached to general-purpose CPUs are now integral High-Performance Computing (HPC) systems. However, these architectures pose severe memory and I/O constraints to computations involving in-situ analytics. This paper introduces MoHA, a framework for in-situ analytics that is designed to efficiently use the limited resources available on heterogeneous platforms. MoHA achieves this efficiency through the extensive use of bitmaps as a compressed or summary representation of simulation outputs. Our specific contributions in this paper include the design of bitmap generation and storage methods suitable for GPUs, the design and efficient implementation of a set of key operators for MoHA, and demonstrations of how several real queries on real datasets can be implemented using these operators. We demonstrate that MoHA reduces I/O transfer as well as overall processing time when compared to a baseline that does not use compressed representations.

This paper presents techniques to detect offline activities of a person when she is tweeting in order to create a dynamic profile of the user, for uses such as better targeting of advertisements. To this end, we propose a hybrid LSTM model for rich contextual learning, along with studies on the effects of applying and combining multiple LSTM based methods with different contextual features. The hybrid model outperforms a set of baselines as well as state-of-the-art methods.

Reducing traffic accidents is an important public safety challenge, therefore, accident analysis and prediction has been a topic of much research over the past few decades. Using small-scale datasets with limited coverage, being dependent on extensive set of data, and being not applicable for real-time purposes are the important shortcomings of the existing studies. To address these challenges, we propose a new solution for real-time traffic accident prediction using easy-to-obtain, but sparse data. Our solution relies on a deep-neural-network model (which we have named DAP, for Deep Accident Prediction); which utilizes a variety of data attributes such as traffic events, weather data, points-of-interest, and time. DAP incorporates multiple components including a recurrent (for time-sensitive data), a fully connected (for time-insensitive data), and a trainable embedding component (to capture spatial heterogeneity). To fill the data gap, we have - through a comprehensive process of data collection, integration, and augmentation - created a large-scale publicly available database of accident information named US-Accidents. By employing the US-Accidents dataset and through an extensive set of experiments across several large cities, we have evaluated our proposal against several baselines. Our analysis and results show significant improvements to predict rare accident events. Further, we have shown the impact of traffic information, time, and points-of-interest data for real-time accident prediction.

Dashboard camera installations are becoming increasingly common due to various Advanced Driver Assistance Systems (ADAS) based services provided by them. Though deployed primarily for crash recordings, calibrating these cameras can allow them to measure real-world distances, which can enable a broad spectrum of ADAS applications such as lane-detection, safe driving distance estimation, collision prediction, and collision prevention. Today, dashboard camera calibration is a tedious manual process that requires a trained professional who needs to use a known pattern (e.g., chessboard-like) at a calibrated distance. In this paper, we propose SmartDashCam, a system for automatic and live calibration of dashboard cameras which always ensures highly accurate calibration values. Smart-DashCam leverages collecting images of a large number of vehicles appearing in front of the camera and using their coarse geometric shapes to derive the calibration parameters. In sharp contrast to the manual process we are proposing the use of a large amount of data and machine learning techniques to arrive at calibration accuracies that are comparable to the manual process. SmartDashCam implemented using commodity dashboard cameras estimates real-world distances with mean errors of 5.7% which closely rivals the 4.1% mean error obtained from traditional manual calibration using known patterns.

Pattern discovery in geo-spatiotemporal data (such as traffic and weather data) is about finding patterns of collocation, co-occurrence, cascading, or cause and effect between geospatial entities. Using simplistic definitions of spatiotemporal neighborhood (a common characteristic of the existing general-purpose frameworks) is not semantically representative of geo-spatiotemporal data. We therefore introduce a new geo-spatiotemporal pattern discovery framework which defines a semantically correct definition of neighborhood; and then provides two capabilities, one to explore propagation patterns and the other to explore influential patterns. Propagation patterns reveal common cascading forms of geospatial entities in a region. Influential patterns demonstrate the impact of temporally long-term geospatial entities on their neighborhood. We apply this framework on a large dataset of traffic and weather data at countrywide scale, collected for the contiguous United States over two years. Our important findings include the identification of 90 common propagation patterns of traffic and weather entities (e.g., rain --> accident --> congestion), which results in identification of four categories of states within the US; and interesting influential patterns with respect to the "location", "duration", and "type" of long-term entities (e.g., a major construction --> more traffic incidents). These patterns and the categorization of the states provide useful insights on the driving habits and infrastructure characteristics of different regions in the US, and could be of significant value for applications such as urban planning and personalized insurance.

Because of the increasing availability of spatiotemporal data, a variety of data-analytic applications have become possible. Characterizing driving context, where context may be thought of as a combination of location and time, is a new challenging application. An example of such a characterization is finding the correlation between driving behavior and traffic conditions. This contextual information enables analysts to validate observation-based hypotheses about the driving of an individual. In this paper, we present DriveContext, a novel framework to find the characteristics of a context, by extracting significant driving patterns (e.g., a slow-down), and then identifying the set of potential causes behind patterns (e.g., traffic congestion). Our experimental results confirm the feasibility of the framework in identifying meaningful driving patterns, with improvements in comparison with the state-of-the-art. We also demonstrate how the framework derives interesting characteristics for different contexts, through real-world examples.

Cloud-based services today depend on many layers of virtual technology and application services. Incidents and problems that arise in such complex operational environments are logged as a ticket, worked on by experts and finally resolved. To assist these experts, any machine recommendation method must meet the following critical business requirements: 1) the ticket must be resolved, meeting specific time constraints or Service Level Targets (SLTs), and 2) any predictive assistance must be trustworthy. Existing research uses probabilistic models to recommend transfers between experts based on limited features intrinsic to the ticket content, and does not demonstrate how to meet SLTs. To address this lack of research and ensure SLT-compliance for an incoming ticket given its recommended sequence of experts, there needs to be an accurate time-to-resolve (TTR) estimation. This research aims to identify important features for modeling time-to-resolve estimation given the routing recommendation sequences. This work particularly makes the following contributions: 1) constructs a framework for assessing TTR estimations and their SLT-compliance, 2) applies the assessment to a baseline estimation model to identify the need for better TTR modeling, and 3) uses language modeling to study the impact of anomalous content on the estimation error, and 4) introduces a set of dynamic features, and a methodology to rigorously model the TTR estimation.

Within the enterprise the timely resolution of incidents that occur within complex Information Technology (IT) systems is essential for the business, yet it remains challenging to achieve. To provide incident resolution, existing research applies probabilistic models locally to reduce the transfers (links) between expert groups (nodes) in the network. This approach is inadequate for incident management that must meet IT Service Levels (SLs). We show this using an analysis of enterprise 'operational big data' and the existence of collective problem solving in which expert skills are often complementary and are applied in sequences that are meaningful. We call such a network - 'Collective Expert Network' (or CEN). We propose a probabilistic model which uses the content-base of transfer sequences to generate assistive recommendations that improves the performance of CEN by: (1) resolving incidents to meet customer time constraints and satisfaction (and not just minimize number of transfers), (2) conforming to previous transfer sequences that have already achieved their SLs, and additionally (3) address trust in order to ensure adoption of recommendations. We present a two-level classification framework that learns regular patterns first and then recommends SL-achieving sequences on a subset of tickets, and for the remaining directly recommends knowledge improvement. The experimental validation shows 34% accuracy improvement over other existing research and locally applied generative models. In addition we show 10% reduction in the volume of SL breaching incidents, and 7% reduction in MTTR of all tickets.

Big Data is characterized by the five V's - of Volume, Velocity, Variety, Veracity and Value. Research on Big Data, that is, the practice of gaining insights from it, challenges the intellectual, process, and computational limits of an enterprise. Leveraging the correct and appropriate toolset requires careful consideration of a large software ecosystem. Powerful algorithms exist, but the exploratory and often ad-hoc nature of analytic demands and a distinct lack of established processes and methodologies make it difficult for Big Data teams to set expectations or even create valid project plans. The exponential growth of data generated exceeds the capacity of humans to process it, and compels us to develop automated computing methods that require significant and expensive computing power in order to scale effectively. In this paper, we characterize data-driven practice and research and explore how we might design effective methods for systematizing such practice and research [19, 22]. Brief case studies are presented in order to ground our conclusions and insights.

Collaborative learning is a key component of software engineering (SE) courses in most undergraduate computing curricula. Thus these courses include fairly intensive team projects, the intent being to ensure that not only do students develop an understanding of key software engineering concepts and practices, but also develop the skills needed to work effectively in large design and development teams. But there is a definite risk in collaborative learning in that there is a potential that individual learning gets lost in the focus on the team's success in completing the project (s). While the team's success is indeed the primary goal of an industrial SE team, ensuring individual learning is obviously an essential goal of SE courses. We have developed a novel approach that exploits the affordances of mobile and web technologies to help ensure that individual students in teams in SE courses develop a thorough understanding of the relevant concepts and practices while working on team projects, indeed, that the team contributes in an essential manner to the learning of each member of the team. We describe the learning theory underlying our approach, provide some details concerning the prototype implementation of a tool based on the approach, and describe how we are using it in an SE course in our program.

Conflict and cooperation would seem to be ideas that are diametrically opposed to each other. But, in fact, classic work by Piaget on how children and adults learn shows that when learners engage with peers in critical discussion of ideas concerning which they have different understandings, that contributes very effectively to learners developing deep understanding of the concepts involved. At the same time, getting students in undergraduate computing (or other technical/engineering) courses to engage with other students in thoughtful discussion of important concepts is very challenging. It can be especially difficult to get women students and students from other underrepresented groups to participate effectively in such discussions. In our work, we exploit the affordances of mobile and web technologies to address these challenges. Our approach not only helps address these challenges, it has a number of other important advantages over face-to-face discussions. We present the theoretical underpinnings of the approach, some details of our prototype implementation, preliminary results from the use of the prototype in a junior/senior level class on Software Engineering, and the design for the next version of our tool. We also discuss the possibilities and usefulness of applying this approach in a range of computing courses from traditional classrooms to MOOCs.

Subsampling workloads compute statistics from a set of observed samples using a random subset of sample data (i.e., a subsample). Data-parallel platforms group these samples into tasks, each task subsamples its data in parallel. In this paper, we study subsampling workloads that benefit from tiny tasks-i.e., tasks comprising few samples. Tiny tasks reduce processor cache misses caused by random subsampling, which speeds up per-task running time. However, they can also cause significant scheduling overheads that negate the time reduction from reduced cache misses. For example, vanilla Hadoop takes longer to start tiny tasks than to run them. We compared the task scheduling overheads of vanilla Hadoop, lightweight Hadoop setups, and BashReduce. BashReduce, the best platform, outperformed the worst by 3.6X but scheduling overhead was still 12% of a task's running time. We improved BashReduce's scheduler by allowing it to size tasks according to kneepoints on the miss rate curve. We tested these changes on high-throughput genotype data and on data obtained from Netflix. Our improved BashReduce outperformed vanilla Hadoop by almost 3X and completed short, interactive jobs almost as efficiently as long jobs. These results held at scale and across diverse, heterogeneous hardware.

Multi-domain network monitoring systems based on active measurements are being widely deployed in high-performance computing and other communities that support large-scale data transfers. Security mechanisms such as policy-driven access to related federated Network Performance Monitoring (NPM) services are important to protect measurement resources and data. In this paper, we present a novel, secure middleware framework viz., “OnTimeSecure” that enables `user-to-service' and `service-to-service' authentication, and enforces federated authorization entitlement policies for timely orchestration of NPM services. OnTimeSecure is built using RESTful APIs and features a hierarchical policy-engine that interfaces with a meta-scheduler for prioritization of measurement requests when there is contention of users concurrently attempting to utilize measurement resources. We validate OnTimeSecure in a federated multi-domain NPM infrastructure by performing threat modeling and security risk assessments based on overall attack likelihood and impact factors.

In this paper, we present the issues to be addressed and the practical solutions to these issues in a mobile application framework for reading and displaying musical scores enhanced to assist the visually impaired in reading and perform the pieces. This framework, currently operating on MusicXML input files, provides the structures and methods for developers to adapt for other music encoding file formats. It also provides the flexible user-settable colors and enlargement parameters to meet the needs of users with various visual impairments. The development challenges fall into three categories: Variable visual impairment driven requirements, Musical notation complexity, and screen real-estate limitations of a 10-inch tablet. The framework's practical solutions to each of these challenges are presented and contrasted with traditional solutions and competing solutions.

Companies seeking to make their web sites usable for viewing by mobile devices currently need to create a parallel set of web pages designed and built specifically for browsing by devices with screen size and computing constraints. Most companies, however, do not have IT resources to spare for this essentially duplicate manual effort. Hence, effective techniques that can ease the burden of developing mobile web pages can be of great value. In this paper, we present the architecture, design and implementation of a 'web site mobilizer' for semi-automatically converting existing web pages to pages suitable for mobile viewing. The mobilizer extracts the component hierarchy of the page using web crawler techniques, analyzes the structure of the web page, and then converts it to the mobile version using a combination of techniques that include link analysis, ranking algorithms based on component content, and interactive removal of irrelevant content. This mobilizer is in limited production use.

An analysis of hiring patterns showed emerging trends: the complexity of information technology (IT) is shifting from development to post-deployment and integration needed for services. Given the complexity of deployed service systems, generated big data, and the national dialogue on educating engineers, we asked ourselves related questions. Do our graduate students have evaluation skills needed to work at the most advanced level of Bloom's taxonomy? Can they learn to frame and solve the problems within complex industry environments while applying the current research? How do we structure a graduate curriculum and an environment that provides experiences in innovation within the constraints of the academic calendar? Here we present an interdisciplinary curriculum comprised of three components: a service interaction blueprint for framing the industry problem, agile principles focusing on aspects of the solution, and Christensen's theory-building to frame the next iteration of research. The environment for industry problems was created through an National Science Funded Industry & University Cooperative Research Center. The feedback from a pilot graduate-level class is positive and provides insights for further research. We show through feedback discussions that it is possible to have translational activity at the industry-university enterprise boundary resourced in by advanced experiential learning.

The National Science Foundation Industry & University Cooperative Research Center (I/UCRC) program is intended to foster productive collaboration between industry organizations and academia. The focus of the I/UCRC research site herein is on the application of technology within the complex extended enterprise. The center's goal is to conduct research that is of interest to both the industry sponsor and the university partner, with the provision that the industry organization must provide major support to the center. In this paper, we describe the Agile Translation Process (ATP) for complex innovations that was developed at the center. The process meets the constraints of the academic calendar, the knowledge needs and the typical length of stay for a master's student, and the availability constraints of the students. At the same time, the process is designed to provide value to the industry sponsor. Specifically, it describes how the process meets the needs of technology consumers in industry seeking to derive tactical value through the funding of the center. In addition, we demonstrate how to derive research results for technology providers through subsequent activities. We also provide metrics from the center for a period of five years, which show, in particular, the benefit of using the ATP method over the last three years. These metrics provide insights on how to reconcile tactical industry needs with the long-term research and funding goals of academia, while understanding the innovations needed within complex contexts. This case study also provides insights on concurrently meeting the needs of all stakeholders - including industry clients, translational faculty members, adjunct faculty from partner companies, graduate students, and the center's affiliated research faculty - within the constraints of the academic calendar. By using an agile translation process and a set of expanded performance metrics, the center effectively applies research to bring innovation to its industry partners.

Several published papers describe various frameworks to implement an Open World Dialog system. This research conducts a critical review of one system using commodity hardware and software, independent of the vendor and authors. The results delineate the parts of the system that are implemented via the SDK and identify the components that require development. Furthermore, we estimate the difficulty of implementing each component of the dialog system using the SDKs.

Project-oriented environments are key to supporting the co-operative work essential to collaborative research activities. However, personnel and resources in project-oriented environments are typically diverse and heterogeneous as they come from both internal as well as external domains. Providing a robust data security system in such an environment becomes critical. The ideal access control architecture should manage access to resources not only based on roles but also based on the specific nature of each resource and its involvement within the project. Traditional role-based access control (RBAC) does not consider the context which often modifies the responsibility given to resources. We propose using an enhanced role-based access control (RBAC) mechanism to address this problem. Specifically, we discuss the implementation of RBAC using ontological methods borrowed from semantic web technology. We used an ontology-based approach for specification and implementation of the RBAC in a collaborative system used within a research group to manage proteomics data, where the access control policy depends on how the project team hierarchy is structured. We describe the design and implementation of this system in this paper. We also provide a preliminary evaluation of the implementation. We find there are several advantages to using ontological methods to implement RBAC. The most significant of these is standardization, which is essential for portability. Also key is modifiability as the actual roles are defined by the ontology itself. Since data access is provided through URI handling moving to a federated system is made easier. This becomes very important in collaborative environments as the data in question is invariably distributed.

This paper discusses our ongoing experiences with teaching software engineering through an inverted classroom. This course format moves traditional lectures out of in-class hours and into the student's personal study time with prerecorded lectures. We support the inverted classroom with complementary techniques, such as structured discussions, weekly quizzes to ensure students watch the lectures before discussion, an innovative Lego-based workshop, a term project, and guest lectures by industry professionals. The inverted classroom allows the students to have an effective educational experience that encompasses both traditional lectures and an active learning environment. To evaluate the efficacy of this format, we use surveys and interviews of both instructors and students. We examine the time commitment of teaching with this method, from both the instructors' perspective and the students'. We also discuss the time commitment for instructor preparation, and quantitative measures of how the inverted classroom helps smooth the variance in the quality of each instructor's teaching. We also analyze the effectiveness of this technique and our methods for mitigating unintended consequences, such as students having an inexact understanding of the material. Through this evaluation, we distill the effects on student learning and instructor teaching.

Object-oriented software design and programming is an essential part of a computer science curriculum. We have observed that novice software developers, such as fresh college graduates who have been taught object-oriented design, are able to apply good design principles in theory. However, this rarely extends into their professional practice, when they are asked to design software intended to run inside a software framework. In fact, we observe that even advanced software developers abandon good design practices when developing software while using a framework, and focus on simply “making it work.” This paper presents and discusses a methodology developed for designing software in the context of frameworks to overcome these issues. We show how design patterns can serve as the bridge between the paradigms imposed by the framework and the ideal, unconstrained design of the system. We also suggest an evaluation method for observing the results of using this methodology when used by the students.

Capstone courses are expected to prepare students for the “real world” by putting them into a microcosm of the real world. In these courses, students are given a problem of some complexity, and are expected to exercise and develop problem-solving skills as they address the problem. Within our Computer Science and Engineering program we have, over the past eight years, successfully scaled up the Capstone courses. Doing so has required innovative thinking about the roles of the students, faculty, and project sponsors. In this paper, we discuss issues with scaling up the components that have made this program successful. These include housing the courses in an NSF IUCRC that enable the cultivation of highly-committed industry partners, the creation of strong pre-requisite courses, careful development of faculty resources through the selective hiring and mentoring of clinical faculty, a commitment of the faculty to give up close management and control, strong partnerships with other organizations within the university to provide students greater access to resources, an emphasis on cross-team knowledge sharing and learning, and the development of unique assessment and evaluation tools so as to be able to monitor, measure and fairly assess a wide-spectrum of projects.

Case studies have been successfully integrated into a wide variety of educational contexts and disciplines. Today, case studies are increasingly accepted as valuable teaching tools in science and engineering curriculums to complement the underlying theory of the field. Well-articulated cases can reinforce abstract concepts, demonstrate the nature of real client interactions, and showcase the relevance of soft skills to students that lack significant practical experience. However, assembling and delivering quality case studies to students requires a great deal of practical disciplinary knowledge, and a careful alignment of the case content and delivery style with curricular objectives, course learning outcomes, and the overarching institutional format. In this paper, we summarize our experience with an approach for constructing case study teaching materials that are integrative and deep in content, but also carefully aligned to the core principles and format of a senior-level software engineering course. Our approach ensures that the cases are complex enough to retain their realism and intrinsic appeal, while mirroring the format and objectives of the course such that the cases reinforce key points in a familiar and consistent fashion to the students.

From a student's perspective, the standard computer science curriculum can effectively develop fundamental software design principles and techniques, but may struggle to fully prepare students for professional practice. Real-world projects require many skills that are challenging to foster in the classroom, including the ability to implement large applications, interact professionally with others, and independently learn new concepts. Undergraduate programs have attempted to develop these abilities through capstone classes and by encouraging participation in co-ops and internships. At Ohio State University, nearly a dozen students have attempted to foster these abilities by doing long-term, real-world, large-scale, commercial-grade software development projects. The first such project recently released an iPhone-based, stadium-centric infotainment application to end-users in time for the 2010 football season. This paper, whose first author is an undergraduate computer science student, captures, from a student's perspective, the educational benefits of ongoing and real-world projects over the more traditional approaches. Following an examination of the educational impacts of these projects relative to the impacts of co-ops, internships and capstone classes, results suggest that long-term, real world projects are a valuable and synergistic component of an undergraduate education in computer science.

Standard methodologies, which have been developed for large software development teams, and Agile practices, developed for small teams, make up the software engineering practices taught in the Computer Science classroom. However, we have found that there is a significant prevalence of “micro” teams doing business-critical software development in the field. Thus, software development best practices for micro teams must be incorporated into the software curriculum. Towards this end, we created a multiple-case case study (comprising five micro team projects) showing how micro teams handle the software development process. Through each of these projects, we seek to showcase what practices from existing software development methodologies are undertaken by the developers of the projects, to achieve similar ends as developers in larger teams. Specifically, the case study highlights how existing software development methodologies need to be modified, adapted or extended for micro teams. The case study and micro team guidelines were presented to students in a software engineering class within the Computer Science department at a large R1 university. The teaching was assessed using a mix of surveys and structured interviews. Initial evaluations showed promise. Students were positively inclined to accept the lessons, and showed good recall of the concepts taught in tests.

In order to better prepare students for professional practice, we have created a software engineering curriculum that provides an end-to-end perspective that begins with the business context of software, and goes all the way to the ongoing management of software services after deployment. This paper examines how the theoretical aspects of this broad-based curriculum may be effectively delivered through a single course within a traditional computer science program. This curriculum is under a diverse set of constraints and requirements, such as the need for pedagogical consistency, faculty development, consideration of the learning style of computer science students, and a need for an effective continuous improvement process. Our approach uses “engineering-oriented” analysis frameworks such as Porter's Five Forces model for the business aspects, and attribute-driven design for software architectures, an “inverted” classroom mode of teaching where lectures are delivered on line with interactions and exercises that promote active learning reserved for the classroom, case studies developed from real projects to serve as concrete examples, open discussion boards and weekly short quizzes for concept refinement and retention, and a paper-based project where students apply the concepts learned. Faculty development and replication outside the current site are also discussed.

Capstone courses in many disciplines often fall into a single paradigm: they allow students to practice the skills they should have gathered through their progress in the department curriculum in a real-world or near-real venue. However, these courses often fail the real-world test by one important factor: they are not interdisciplinary projects, which is not indicative of industry experiences. We are attempting to create an interdisciplinary environment for capstone courses, involving both design and computer science students, to more adequately prepare students for industry work. This work-in-progress paper describes our experiences and plans for bettering the interdisciplinary capstone experience. The experiences show that there is a fundamental miscommunication between students of different disciplines that hinders their ability to collaborate. By analyzing qualitative questionnaires from thirty-three computer science students, we have affirmed the existence of this rift in inter-departmental understanding. This realization has formed our basis for creating educational modules to ease the collaboration between computer science and design students.

Industry-practiced agile methods must become an integral part of a software engineering curriculum. It is essential that graduates of such programs seeking careers in industry understand and have positive attitudes toward agile principles. With this knowledge they can participate in agile teams and apply these methods with minimal additional training. However, learning these methods takes experience and practice, both of which are difficult to achieve in a direct manner within the constraints of an academic program. This paper presents a novel, immersive boot camp approach to learning agile software engineering concepts with LEGO®bricks as the medium. Students construct a physical product while inductively learning the basic principles of agile methods. The LEGO®-based approach allows for multiple iterations in an active learning environment. In each iteration, students inductively learn agile concepts through their experiences and mistakes. Subsequent iterations then ground these concepts, visibly leading to an effective process. We assessed this approach using a combination of quantitative and qualitative methods. Our assessment shows that the students demonstrated positive attitudes toward the boot-camp approach compared to lecture-based instruction. However, the agile boot camp did not have an effect on the students' recall on class tests when compared to their recall of concepts taught in lecture-based instruction.

Software often lives in a complex software eco-system with complex interactions and dependencies between different modules or components. In Windows, this problem is exacerbated both by the overall system complexity and its closed source nature. Even when source is available, there are still interactions with modules which are only in binary form.

Reasoning about the impact of change is critical throughout the Information Technology (IT) architecture lifecycle management processes and this is especially challenging because installed architectures are complex, evolve constantly, and most changes have some global impact. We present an enterprise-interaction ontology for integrated query, analysis, and monitoring that supports features to allow architects and engineers pin-point the impact of change to the installed architecture before implementation. The ontology represents select associations between the enterprise’s business processes, services and infrastructure so that significant consequences of a change are propagated to affected areas based on underlying rules. Thus, interdependencies and relationships that are not obvious are identified and the impact is quantified. This allows the architect to know the complete scope of modifications required in order to accomplish a change in a manner consistent with best practices (like ITIL version 3). We illustrate - 1) the rules and taxonomy relationships that give us the ability to propagate changes and determine the impact, and 2) how actual questions and decision-making during the architecture management processes can be better supported using a more precise and factual understanding. Not only does the interaction methodology help analyze the potential impact of adding a new component, a change due to an incident, or the deletion of an existing component from the architecture, it also supports business-IT alignment processes like chargeback, capacity management and disaster recovery.

We propose a systems analysis framework based on goal modelling and transactions for improved decision-making about security solution architectures - with a specific focus on layered security of physical spaces and assets. The framework assists in defining more complete security strategies as well as analyzing tradeoffs between security and other factors such as cost and privacy. Using the conceptual transaction or requirements-execution-delivery (or RED) Transaction model as the basis, we provide a dynamic virtual structure and methodology for security analysis. The benefit is that the implemented security can be optimized based on the value to the various stakeholders and to minimize the benefit to the attacker.

In this position paper, we present certain observed characteristics of global software development for the enterprise, as well as trends in enterprise information technology needs for a global enterprise. We then identify IT workforce needs and the consequent curriculum elements desired

The Kansei testbed at the Ohio State University is designed to facilitate research on networked sensing applications at scale. Kansei embodies a unique combination of characteristics as a result of its design focus on sensing and scaling: (i) Heterogeneous hardware infrastructure with dedicated node resources for local computation, storage, data exfiltration and back-channel communication, to support complex experimentation, (ii) Time accurate hybrid simulation engine for simulating substantially larger arrays using testbed hardware resources, (iii) High fidelity sensor data generation and real-time data and event injection, (iv) Software components and associated job control language to support complex multi-tier experiments utilizing real hardware resources and data generation and simulation engines. In this paper, we present the elements of Kansei testbed architecture, including its hardware and software platforms as well as its hybrid simulation and sensor data generation engines.

Research in sensor networks has begun to address the use of mobility to improve the reachability of the network, but a number of network principles and application patterns remain to be explored in this context. We propose here a network architecture that uses energy constrained devices for enabling new campus wide applications. Specifically, our demonstration illustrates a new network stack and application framework for a class of locality specific applications. The locality specific nature favors exploiting the limited, slow and regional mobility pattern present in large campuses, as opposed to exclusively exploiting the Internet or the cellular network.

Project ExScal (for extreme scale) fielded a 1000+ node wireless sensor network and a 200+ node peer-to-peer ad hoc network of 802.11 devices in a 13km by 300m remote area in Florida, USA during December 2004. In comparison with previous deployments, the ExScal application is relatively complex and its networks are the largest ones of either type fielded to date. In this paper, we overview the key requirements of ExScal, the corresponding design of the hardware/software platform and application, and some results of our experiments.

The programming of efficient parallel software typically requires extensive experimentation with program prototypes. To facilitate such experimentation, any programming system that supports rapid prototyping of parallel programs should provide high-level language primitives with which programs can be explicitly, statically, or dynamically tuned with respect to performance and reliability. Such language primitives should be able to refer conveniently to the information about the executing program and the parallel hardware required for tuning. Such information may include monitoring data about the current or previous program or even hints regarding appropriate tuning decisions. Language primitives and an associated programming system for program tuning are presented. The primitives and system have been implemented, and have been tested with several parallel applications on a network of Unix workstations.>