In this paper we present a pilot program at San Francisco State University, Promoting INclusivity in Computing (PINC), that is designed to achieve two goals simultaneously: (i) improving diversity in computing, and (ii) increasing computing literacy in data-intensive fields. To achieve these goals, the PINC program enrolls undergraduate students from non Computer Science (non-CS) fields, such as, Biology, that have become increasingly data-driven, and that traditionally attract diverse student population. PINC incorporates several well-established pedagogical practices, such as, cohort-based program structure, near-peer mentoring, and project-driven learning, to attract, retain, and successfully graduate a highly diverse and interdisciplinary student body. On successful completion of the program, students are awarded a minor in Computing Applications. Since its inception 18 months ago, 60 students have participated in this program. Of these 73% are women, and 51% are underrepresented minorities (URM). 74% of the participating students had nominal or no exposure to computer programming before PINC. Findings from student surveys show that majority of the PINC students now feel less intimidated about computer programming, and vividly see its utility and necessity. For several students, participation in the PINC program has already opened up career pathways (industry and academic summer internships) that were not available to them before.

The overall goal of our Software Engineering Teamwork Assessment and Prediction (SETAP) project is to develop effective machine-learning-based methods for assessment and early prediction of student learning effectiveness in software engineering teamwork. Specifically, we use the Random Forest (RF) machine learning (ML) method to predict the effectiveness of software engineering teamwork learning based on data collected during student team project development. These data include over 100 objective and quantitative Team Activity Measures (TAM) obtained from monitoring and measuring activities of student teams during the creation of their final class project in our joint software engineering classes which ran concurrently at San Francisco State University (SFSU), Fulda University (Fulda) and Florida Atlantic University (FAU). In this paper we provide the first RF analysis results done at SFSU on our full data set covering four years of our joint SE classes. These data include 74 student teams with over 380 students, totaling over 30000 discrete data points. These data are grouped into 11 time intervals, each measuring important phases of project development during the class (e.g. early requirement gathering and design, development, testing and delivery). We briefly elaborate on the methods of data collection and describe the data itself. We then show prediction results of the RF analysis applied to this full data set. Results show that we are able to detect student teams who are bound to fail or need attention in early class time with good (about 70%) accuracy. Moreover, the variable importance analysis shows that the features (TAM measures) with high predictive power make intuitive sense, such as late delivery/late responses, time used to help each other, and surprisingly statistics on commit messages to the code repository, etc. In summary, we believe we demonstrate the viability of using ML on objective and quantitative team activity measures to predict student learning of software engineering teamwork, and point to easy-to-measure factors that can be used to guide educators and software engineering managers to implement early intervention for teams bound to fail. Details about the project and the complete ML training database are downloadable from the project web site.

EMG pattern recognition has been studied for control of prostheses and rehabilitation systems for decades. Existing research platforms for developing EMG pattern recognition algorithms are typically based on MATLAB and the collection of EMG signals is often done by expensive, non-portable data acquisition systems. The requirement of these resources usually limits the use of these platforms in the lab environments and prohibits their widespread to other fields and applications. To address this limitation, this paper presents a low-cost, easy to use, and flexible platform called MyoHMI for developing real-time human machine interfaces for myoelectric controlled applications. MyoHMI facilitates the interface with a commercial EMG-based armband Myo, which costs less than $200 and can be easily worn by the user without the need of special preparation. MyoHMI also provides a highly modular and customizable C/C++ based software engine which seamlessly integrates a variety of interfacing and signal processing modules, from data acquisition through signal processing and pattern recognition, to real-time evaluation and control. The experimental results on able-bodied human subjects for controlling two evaluation platforms in real time verified the merit of the MyoHMI platform and demonstrated the feasibility of a low-cost solution for the development of myoelectric controlled applications.

Effective teaching of teamwork skills in local and globally distributed Software Engineering (SE) teams is recognized as an important part of the education of current and future software engineers. Effective methods for assessment and early prediction of learning effectiveness in SE teamwork are not only a critical part of teaching but also of value in industrial training and project management. This paper presents a novel analytical approach to the assessment and, most importantly, the prediction of learning outcomes in SE teamwork based on data from our joint software engineering class concurrently taught at San Francisco State University (SFSU), Florida Atlantic University (FAU) and Fulda University, Germany (Fulda). Our approach focuses on assessment and prediction of SE teamwork in terms of ability of student teams to apply best SE processes and develop SE products. It differs from existing work in the following aspects: a) it develops and uses only objective and quantitative measures of team activity from multiple sources, such as statistics of student time use, software engineering tool use, and instructor observations; b) it leverages powerful machine learning (ML) techniques applied to team activity measurements to identify quantitative and objective factors which can assess and predict learning of software engineering teamwork skills at the team level. In this paper we provide the following contributions: a) we present in detail for the first time the full team activity measurement data set we developed, consisting of over 40 objective and quantitative measures extracted from student teams working on class projects; b) we present a ML framework which applies the Random Forest (RF) algorithm to the team activity measurements and team outcomes, focusing on predicting teams that are likely to fail; c) we describe in detail our now fully implemented and operational data processing pipeline, consisting of data collection methods from multiple sources, ML training database creation, and ML analysis subsystems; and finally d) we present very preliminary results of ML analysis results based on the data from our joint software engineering classes in Fall 2012, and Spring 2013, with the data from 17 student teams. While our ML training database is currently small, it continuously grows. Our preliminary results, verified with two independent accuracy measures, show that RF is able to predict SE Process and SE Product team performance in intuitively explainable manner.

Machine learning-based prediction of protein functions plays a key role in bioinformatics and pharmaceutical research, facilitating swift discovery of new drugs in high-throughput settings. This paper presents an adaptation of Random Forest to the structure-based protein function prediction. Our system represents protein's 3D physicochemical structural information in microenvironment descriptors whose spatial resolution is much finer than other sequence-based protein descriptors. We prepare our datasets for seven active sites from five protein function classes by using multiple public data banks and train Random Forest classifiers to identify these seven function models in proteins. This paper presents two experiment studies: 1) a 5-fold stratified cross-validation for comparing Random Forest with Naive Bayes and Support Vector Machine and 2) systematic comparison of Random Forest's two variable importance measures. Promising results of these studies demonstrate a potential for Random Forest to improve the accuracy of the current protein function assays.

One of the challenges in effective software engineering (SE) education is the lack of objective assessment methods of how well student teams learn the critically needed teamwork practices, defined as the ability: (i) to learn and effectively apply SE processes in a teamwork setting, and (ii) to work as a team to develop satisfactory software (SW) products. In addition, there are no effective methods for predicting learning effectiveness in order to enable early intervention in the classroom. Most of the current approaches to assess achievement of SE teamwork skills rely solely on qualitative and subjective data taken as surveys at the end of the class and analyzed only with very rudimentary data analysis. In this paper we present a novel approach to address the assessment and prediction of student learning of teamwork effectiveness in software engineering education based on: a) extracting only objective and quantitative student team activity data during their team class project; b) pairing these data with related independent observations and grading of student team effectiveness in SE process and SE product components in order to create “training database” and c) applying a machine learning (ML) approach, namely random forest classification (RF), to the above training database in order to create ML models, ranked factors and rules that can both explain (e.g. assess) as well as provide prediction of the student teamwork effectiveness. These student team activity data are being collected in joint and already established (since 2006) SE classes at San Francisco State University (SFSU), Florida Atlantic University (FAU) and Fulda University, Germany (Fulda), from approximately 80 students each year, working in about 15 teams, both local and global (with students from multiple schools).

In this paper, we study the performance of the Greedy Perimeter Stateless Routing (GPSR) [1] and the Greedy Routing for Maximum Lifetime (GRMax)[2] which is our proposal, in wireless sensor networks in random topology and grid topology wireless sensor network. We decide to compare the performance of GPSR and GRMax in different topology of networks using the same simulation environment. Evaluation and comparison of GPSR and GRMax is done through OPNET. Results show that GRMax performs better than GPSR with respect to packets delivery fraction, number of dropped packets and network lifetime. With the introduction of VIP nodes, GRMax avoid early network partition in networks with randomly distributed nodes and in networks with grid distribution nodes. Moreover, with the criteria used to choose the next hop node in GRMax, battery power resources are used efficiently. GRMax performs better than GPSR because with its algorithm, from the source to the destination GRMax uses a path close to the straight line.

In this paper, we present GRMax for WSN (Greedy Routing for Maximum Lifetime in Wireless Sensor Networks) whose goal is to manage the restricted energy so that the Wireless Sensor Network (WSN) stay connected for the maximum time possible. We evaluate the performance of GRMax for WSN by replacing the greedy routing of Greedy Perimeter Stateless Routing (GPSR) protocol with GRMax. Evaluation and comparison of GPSR and GRMax for WSN is done through NS2. Results show that GRMax for WSN performs better than GPSR with respect to packet delivery fraction, number of dropped packet and network lifetime. With the introduction of VIP nodes we avoid early network partition whatever random the traffic is. And with our new metric, we make sure that battery power resources are used efficiently since we suppose that every node can fix his transmission range depending on the distance to the selected next hop. As metric we use the cosine of the deviation angle. We can use our resources efficiently because with our metric, from the source to the destination we use a path close to the straight line.

This paper proposes a novel approach to improve the segmentation quality of a 3D random walks algorithm using classifiability criteria. We produce a range of potential threshold values by extending the decision function of a random walks algorithm using a likelihood ratio test. Optimal threshold values are quantitatively isolated using two data-driven methods: maximum total accuracy and Bayesian cross validation criteria. The proposed methods are evaluated using a dataset of 28 dental lesions in 3D cone-beam CT scans. Both methods produce viable thresholds, the first corresponding to a conservative segmentation and the second a relaxed segmentation. We qualitatively compare the results to determine the best method.

This paper presents robust click-point linking: a novel localized registration framework that allows users to interactively prescribe where the accuracy has to be high. By emphasizing locality and interactivity, our solution is faithful to how the registration results are used in practice. Given a user-specified point, the click-point linking provides a single point-wise correspondence between a data pair. In order to link visually dissimilar local regions, a correspondence is sought by using only geometrical context without comparing the local appearances. Our solution is formulated as a maximum likelihood estimation (MLE) without estimating a domain transformation explicitly. A spatial likelihood of Gaussian mixture form is designed to capture geometrical configurations between the point-of-interest and a hierarchy of global-to-local 3D landmarks that are detected using machine learning and entropy based feature detectors. A closed-form formula is derived to specify each Gaussian component by exploiting geometric in-variances under specific group of domain transformation via RANSAC-like random sampling. A mean shift algorithm is applied to robustly and efficiently solve the local MLE problem, replacing the standard consensus step of the RANSAC. Two transformation groups of pure translation and scaling/translation are considered in this paper. We test feasibility of the proposed approach with 16 pairs of whole-body CT data, demonstrating the effectiveness.

To handle more phone and personal computer users in such a natural disaster as terribly strong earthquake, a traffic control has been previously proposed by limiting the individual call holding time. This traffic control mechanism leads to our T-limited service. By T-limited service, we mean that the service time is limited to a threshold T. The call whose service time reaches T is assumed to be lost. For evaluating the traffic control performance, we present multi-server loss systems with T-limited service. Without any retrial queues, we analyze a Poisson input and general service time loss system to derive the steady-state distribution of the number of calls in the system. With a retrial queue, assuming further that the call sojourn time at the retrial queue is exponentially distributed and that the T-limited service time is also exponentially distributed, we propose an approximation for the steady-state distribution of the number of calls in the system. Our approximation accuracy is validated by a simulation result.

We propose EMD-L1: a fast and exact algorithm for computing the Earth Mover's Distance (EMD) between a pair of histograms. The efficiency of the new algorithm enables its application to problems that were previously prohibitive due to high time complexities. The proposed EMD-L1 significantly simplifies the original linear programming formulation of EMD. Exploiting the L1 metric structure, the number of unknown variables in EMD-L1 is reduced to O(N) from O(N2) of the original EMD for a histogram with N bins. In addition, the number of constraints is reduced by half and the objective function of the linear program is simplified. Formally, without any approximation, we prove that the EMD-L1 formulation is equivalent to the original EMD with a L1 ground distance. To perform the EMD-L1 computation, we propose an efficient tree-based algorithm, Tree-EMD. Tree-EMD exploits the fact that a basic feasible solution of the simplex algorithm-based solver forms a spanning tree when we interpret EMD-L1 as a network flow optimization problem. We empirically show that this new algorithm has an average time complexity of O(N2), which significantly improves the best reported supercubic complexity of the original EMD. The accuracy of the proposed methods is evaluated by experiments for two computation-intensive problems: shape recognition and interest point matching using multidimensional histogram-based local features. For shape recognition, EMD-L1 is applied to compare shape contexts on the widely tested MPEG7 shape data set, as well as an articulated shape data set. For interest point matching, SIFT, shape context and spin image are tested on both synthetic and real image pairs with large geometrical deformation, illumination change, and heavy intensity noise. The results demonstrate that our EMD-L1-based solutions outperform previously reported state-of-the-art features and distance measures in solving the two tasks.

This paper proposes a new variational bound optimization framework for incorporating spatial prior information to the mean shift-based data-driven mode analysis, offering flexible control of the mean shift convergence. Two forms of Gaussian spatial priors are considered. Attractive prior pulls the convergence toward a desired location. Repulsive prior pushes away from such a location. Using a generic variational optimization formulation via construction of quadratic lower and upper bounds, we show that the priorconstrained mean shift step can be interpreted as an information fusion of the data and prior terms in the sense of the best linear unbiased estimator. This approach is used to propose a mode parsing algorithm using the inhibitionof-return principle. The proposed algorithm is used for a semi-automatic 3D segmentation of lung nodules in CT data for evaluating its effectiveness. Our experiments demonstrate that the proposed solution can successfully segment challenging wall-attached cases. 1

In this paper we propose diffusion distance, a new dissimilarity measure between histogram-based descriptors. We define the difference between two histograms to be a temperature field. We then study the relationship between histogram similarity and a diffusion process, showing how diffusion handles deformation as well as quantization effects. As a result, the diffusion distance is derived as the sum of dissimilarities over scales. Being a cross-bin histogram distance, the diffusion distance is robust to deformation, lighting change and noise in histogram-based local descriptors. In addition, it enjoys linear computational complexity which significantly improves previously proposed cross-bin distances with quadratic complexity or higher. We tested the proposed approach on both shape recognition and interest point matching tasks using several multi-dimensional histogram-based descriptors including shape context, SIFT, and spin images. In all experiments, the diffusion distance performs excellently in both accuracy and efficiency in comparison with other state-of-the-art distance measures. In particular, it performs as accurately as the Earth Mover’s Distance with much greater efficiency.

This paper describes a novel classification method for computer aided detection (CAD) that identifies structures of interest from medical images. CAD problems are challenging largely due to the following three characteristics. Typical CAD training data sets are large and extremely unbalanced between positive and negative classes. When searching for descriptive features, researchers often deploy a large set of experimental features, which consequently introduces irrelevant and redundant features. Finally, a CAD system has to satisfy stringent real-time requirements.This work is distinguished by three key contributions. The first is a cascade classification approach which is able to tackle all the above difficulties in a unified framework by employing an asymmetric cascade of sparse classifiers each trained to achieve high detection sensitivity and satisfactory false positive rates. The second is the incorporation of feature computational costs in a linear program formulation that allows the feature selection process to take into account different evaluation costs of various features. The third is a boosting algorithm derived from column generation optimization to effectively solve the proposed cascade linear programs.We apply the proposed approach to the problem of detecting lung nodules from helical multi-slice CT images. Our approach demonstrates superior performance in comparison against support vector machines, linear discriminant analysis and cascade AdaBoost. Especially, the resulting detection system is significantly sped up with our approach.

We propose a novel semi-automatic figure-ground segmentation solution for blob-like objects in multi-dimensional images. The blob-like structure constitutes various objects of interest that are hard to segment in many application domains, such as tumor lesions in 3D medical data. The proposed solution is motivated towards computer-aided diagnosis medical applications, justifying our semi-automatic and figure-ground approach. The efficient segmentation is realized by combining the robust anisotropic Gaussian model fitting and the likelihood ratio test (LRT)-based non-parametric segmentation in joint space-intensity domain. The robustly fitted Gaussian is exploited to estimate the foreground and background likelihoods for both spatial and intensity variables. We demonstrate that the LRT with the bootstrapped likelihoods is assured to be the optimal Bayesian classification while automatically determining the LRT threshold. A 3D implementation of the proposed algorithm is applied to the lung nodule segmentation in CT data and validated with 1310 cases. Our efficient solution segments a target nodule in less than 3 seconds in average.

This paper proposes a robust statistical estimation and verification framework for characterizing the ellipsoidal (anisotropic) geometrical structure of pulmonary nodules in the Multislice X-ray computed tomography (CT) images. Given a marker indicating a rough location of a target, the proposed solution estimates the target's center location, ellipsoidal boundary approximation, volume, maximum/average diameters, and isotropy by robustly and efficiently fitting an anisotropic Gaussian intensity model. We propose a novel multiscale joint segmentation and model fitting solution which extends the robust mean shift-based analysis to the linear scale-space theory. The design is motivated for enhancing the robustness against margin-truncation induced by neighboring structures, data with large deviations from the chosen model, and marker location variability. A chi-square-based statistical verification and analytical volumetric measurement solutions are also proposed to complement this estimation framework. Experiments with synthetic one-dimensional and two-dimensional data clearly demonstrate the advantage of our solution in comparison with the gamma-normalized Laplacian approach (Linderberg, 1998) and the standard sample estimation approach (Matei, 2001). A quasi-real-time three-dimensional nodule characterization system is developed using this framework and validated with two clinical data sets of thin-section chest CT images. Our experiments with 1310 nodules resulted in (1) robustness against intraoperator and interoperator variability due to varying marker locations, (2) 81% correct estimation rate, (3) 3% false acceptance and 5% false rejection rates, and (4) correct characterization of clinically significant nonsolid ground-glass opacity nodules. This system processes each 33-voxel volume-of-interest by an average of 2 s with a 2.4-GHz Intel CPU. Our solution is generic and can be applied for the analysis of blob-like structures in various other applications.

A framework for learning an accurate and general parametric facial model from 2D images is proposed and its application for analyzing and synthesizing facial images with pose variation is demonstrated. Our parametric piecewise linear subspace method covers a wide range of pose variation in a continuous manner through a weighted linear combination of local linear models distributed in a pose parameter space. The linear design helps to avoid typical nonlinear pitfalls such as overfitting and time-consuming learning. Experimental results show sub-degree and sub-pixel accuracy within /spl plusmn/55 degree full 3D rotation and good generalization capability over unknown head poses when learned and tested for specific persons.

A method of manifold representation for human faces with pose variations is proposed. Our model consists of mappings between 3D head angles and facial images separately represented in shape and texture, via sub-space models spanned by principal components (PC). Explicit mappings to and from 3D head angles are used as processes of pose estimation and transformation, respectively. Generalization capability to unknown head poses enables our model to continuously cover pose parameter space, providing high approximation accuracy. The feasibility of this model is evaluated in a number of experiments. We also propose a novel pose-invariant face recognition system using our model as the entry format for a gallery of known persons. Experimental results with 3D facial models recorded by a Cyberware scanner show that our model provides a superior recognition performance against pose variations, and that the texture synthesis process is carried out correctly.

A framework for integrating the processes of object recognition and knowledge acquisition is proposed and applied to solve a task of automatic video indexing based on personal appearance events in a video stream. Spatiotemporal segmentation using multiple cues and example based adaptation of a known person gallery are combined in a prototype system which demonstrated successful results in our preliminary experiments.

This paper investigates the performance of distributed dynamic channel assignment (DDCA) strategies in sector cell layout systems using directional antennas. Simulations are done under the same propagation conditions and with the same system parameters. They show that DDCA strategies in sector cell layout systems gain 2.3/spl sim/8.2 times higher capacity than do the fixed channel assignment (FCA) strategy in conventional cellular systems using omnidirectional antennas and 1.5/spl sim/2.5 times higher than do DDCA strategies in conventional systems.

Traffic congestion often occurs on highways and results in a heavy call traffic load that impairs the performance of dynamic channel assignment (DCA) strategies. The authors therefore propose a channel framework in which all the available channels are divided into two subsets according to the directions in which the mobile stations move. The decision of the division ratio depends on the offered load of each direction. The authors investigate, through simulations, DCA performances with channel framework for one-sided and both-sided traffic congestion and compare these performances with that of DCA without the framework. They show that the channel framework increases call blocking but considerably reduces the occurrence of forced call termination and of channel changing during a call.

New algorithms for frequency channel assignment in a small cellular mobile radio system are proposed. The algorithms are for a channel assignment method which control the transmitter power level based on the distance between the cell site and the mobile unit. These algorithms are such that any channel can be used by any cell site and mobile unit, as long as a required threshold level of carrier to interference ratio is maintained. As the cochannel reuse distance decreases, the proposed algorithms have given an increase in capacity in comparison to a fixed channel assignment.>