Undergraduate study of computer science is a common pathway into computing professions; however, attraction and retention of students from underrepresented groups is a long-standing problem. Higher education is in the middle of a well-documented "leaky pipeline," and the reasons for the dismal diversity statistics in computer science are wide-ranging and reach beyond the college classroom experience. Evidence has shown that Culturally Responsive Teaching (CRT) results in positive learning outcomes and feelings of belonging and inclusion, leading to stronger retention of students from underrepresented groups. This work details efforts across our department to incorporate three components of CRT into introductory and advanced courses: using diverse assets; encouraging identity connections; and structuring meaning-making. Our objective was to create and implement instructional materials that reflect a range of cultural perspectives, help students to express their unique identities in course activities, and craft opportunities for reflection on learning and connection to one's lived experience. We provide a repository of resources and discuss in more detail several examples of course materials targeted at these objectives. In addition to general lessons learned, we present survey results showing that students from underrepresented groups in courses using these materials indicated an increased sense of belonging. Their responses showed statistically significant improvement in their belief that computer science could better help them understand themselves, and in considering themselves a "computer science person."

Recent interest in alternative grading strategies has been increasing in the Computer Science Education community. The umbrella term ungrading has been used to refer to a variety of practices that de-emphasize numeric grades. In this paper we present the results of implementing an ungrading scheme that eliminates numeric grades, allows resubmission of assignments, and encourages student input into their final assigned letter grade. We administered surveys measuring student attitudes and motivation at the start and end of three different upper level Computer Science elective courses using the new grading scheme and found a significant increase in students' feelings of intrinsic goal orientation (valuing coursework for its own sake), self-efficacy (feeling able to be successful), and control of learning (taking responsibility for their own learning). We observed that, given the opportunity, most students chose to redo only a small number of assignments, and most students requested final grades within a half-letter of the instructor's estimate. Overall, compared with prior iterations of the courses that were graded traditionally, the final grade point average did not significantly increase, while students' reported level of effort did significantly increase. Comments on post-course surveys indicate that students liked the new grading scheme, and they reported improved learning and reduced anxiety.

Scrumage (SCRUM for AGile Education) is a recently proposed classroom management technique in which students are given autonomy to choose individually from a variety of pedagogies (e.g., traditional lectures, active learning, a flipped-based ap-proach, etc.). The result is multiple simultaneous pedagogical styles in a single course. In this paper we present the results of comparing six sections of an introductory programming course at the same university, three of which used Scrumage and three of which took a traditional approach. We adminis-tered surveys of both content acquisition and learning atti-tudes at the beginning and end of the course. While students in all sections improved in content learning, the students in the Scrumage classrooms outperformed those in the traditional sections. The improvement in content learning was also more uniformly distributed among students, not limited to the high achievers. Scrumage students showed generally improved atti-tudes about learning after the course, especially in the areas of Effort Regulation (perseverance in problem solving) and Con-trol of Learning (taking responsibility for learning success). We observed some correlation between this metalearning and improvements in content scores in the Scrumage sections, but not in the traditional sections. Finally, based on an analysis of student comments in the Scrumage sections, we show that as the course progressed, positive student comments about their abilities and confidence were more common, even as course material became more difficult. We believe that positive atti-tude changes we saw with Scrumage mean it has potential for widening the retention of students in Computer Science.

While the "flipped classroom" style has some educational benefits, there are also known benefits to other pedagogical approaches such as lectures, educational games, class discussions, and case studies. In addition to a wide variety of pedagogical approaches, there are a wide variety of computer science learning materials, including videos, interactive tutorials, e-textbooks and traditional textbooks. The choices of approach and materials present a series of trade-offs and may favor some groups of students over others. In this workshop, we present a methodology called Scrumage, (SCRUM for AGile Education) which allows each student in a course to adopt the pedagogical approach and materials that best fit each student's individual learning needs. Scrumage adapts concepts from the Scrum project management technique to manage student teams where the project is learning. Each team learns with the style they prefer, so that multiple pedagogical styles and materials are in use in the course simultaneously. Participants in this workshop will be introduced to the methodology, benefits, and tools of this approach and will work through guided steps to implementing it in the course of their choice.

In this paper, we propose a method for estimating 3D human pose from a single RGB image. Compared to methods that either provide point estimates for coordinate regression or unimodal predictions of joint locations, our approach predicts joint locations using multimodal distributions. In addition, we apply a data-driven approach to learn the conditional dependencies of the relative positions of joints. Our end-to-end approach takes as input images with either 2D or 3D labels and performs on par or better than the state-of-the-art on the Human3.6M and MPII datasets.

One often-overlooked area for assistive technology is help for those with emotional needs. Since these individuals may not emote in a typical way, most techniques for affective computing will not work for this population. Further, the applications that detect emotion are generally concerned with helping the user with some task, not simply helping them with their emotional difficulties. In this work, we present React 2 Me, a system that uses ambient technology to detect multimodal behavioral cues that may indicate emotional distress and adjust the environment to help the individual regulate their emotions.

We address the problem of automated discretization for continuous labels in the context of head pose estimation from overhead cameras. Due to the lack of visual detail, precise head pose estimates are not always possible. A common approach is to discretize the space of head pose angles, turning a real-valued prediction task into a coarser (ordered) classification variant. Often, however, the ranges are arbitrarily defined (e.g., dividing up parameter space evenly). Our work incorporates label discretization into the feature learning process and improves the accuracy of coarse head pan angle prediction from overhead cameras on a benchmark dataset.

We present a data-driven method for semi-supervised multioutput regression on image manifolds, which simultaneously considers the manifold structure of the input data and complex output labels. Compared to related methods, our method achieves superior prediction accuracy on a variety of data sets, with as few as 5% of the input examples labeled. Also, with a few labeled examples and no domain-specific tuning, our method performs on par with specialized algorithms for tasks such as face landmark detection.

Much communication between developers of free, libre, and open source software (FLOSS) projects happens on email mailing lists. Geographically and temporally dispersed development teams use email as an asynchronous, centralized, persistently stored institutional memory for sharing code samples, discussing bugs, and making decisions. Email is especially important to large, mature projects, such as the Linux kernel, which has thousands of developers and a multi-layered leadership structure. In this paper, we collect and analyze data to understand the communication patterns in such a community. How do the leaders of the Linux Kernel project write in email? What are the salient features of their writing, and can we discern one leader from another? We find that there are clear written markers for two leaders who have been particularly important to recent discussions of leadership style on the Linux Kernel Mailing List (LKML): Linux Torvalds and Greg Kroah-Hartman. Furthermore, we show that it is straightforward to use a machine learning strategy to automatically differentiate these two leaders based on their writing. Our findings will help researchers understand how this community works, and why there is occasional controversy regarding differences in communication styles on the LKML.

Gamesourcing has emerged as an approach for rapidly acquiring labeled data for learning-based, computer vision recognition algorithms. In this article, we present an approach for using RGB-D sensors to acquire annotated training data for human pose estimation from 2D images. Unlike other gamesourcing approaches, our method does not require a specific game, but runs alongside any gesture-based game using RGB-D sensors. The automatically generated datasets resulting from this approach contain joint estimates within a few pixel units of manually labeled data, and a gamesourced dataset created using a relatively small number of players, games, and locations performs as well as large-scale, manually annotated datasets when used as training data with recent learning-based human pose estimation methods for 2D images.

For human action recognition methods, there is often a trade-off between classification accuracy and computational efficiency. Methods that include 3D information from multiple cameras are often computationally expensive and not suitable for real-time application. 2D, frame-based methods are generally more efficient, but suffer from lower recognition accuracies. In this paper, we present a hybrid keypose-based method that operates in a multi-camera environment, but uses only a single camera at a time. We learn, for each keypose, the relative utility of a particular viewpoint compared with switching to a different available camera in the network for future classification. On a benchmark multi-camera action recognition dataset, our method outperforms approaches that incorporate all available cameras.