Machine learning (ML) has become highly relevant in applications across all industries, and specialists in the field are sought urgently. As it is a highly interdisciplinary field, requiring knowledge in computer science, statistics and the relevant application domain, experts are hard to find. Large corporations can sweep the job market by offering high salaries, which makes the situation for small and medium enterprises (SME) even worse, as they usually lack the capacities both for attracting specialists and for qualifying their own personnel. In order to meet the enormous demand in ML specialists, universities now teach ML in specifically designed degree programs as well as within established programs in science and engineering. While the teaching almost always uses practical examples, these are somewhat artificial or outdated, as real data from real companies is usually not available. The approach reported in this contribution aims to tackle the above challenges in an integrated course, combining three independent aspects: first, teaching key ML concepts to graduate students from a variety of existing degree programs; second, qualifying working professionals from SME for ML; and third, applying ML to real-world problems faced by those SME. The course was carried out in two trial periods within a government-funded project at a university of applied sciences in south-west Germany. The region is dominated by SME many of which are world leaders in their industries. Participants were students from different graduate programs as well as working professionals from several SME based in the region. The first phase of the course (one semester) consists of the fundamental concepts of ML, such as exploratory data analysis, regression, classification, clustering, and deep learning. In this phase, student participants and working professionals were taught in separate tracks. Students attended regular classes and lab sessions (but were also given access to e-learning materials), whereas the professionals learned exclusively in a flipped classroom scenario: they were given access to e-learning units (video lectures and accompanying quizzes) for preparation, while face-to-face sessions were dominated by lab experiments applying the concepts. Prior to the start of the second phase, participating companies were invited to submit real-world problems that they wanted to solve with the help of ML. The second phase consisted of practical ML projects, each tackling one of the problems and worked on by a mixed team of both students and professionals for the period of one semester. The teams were self-organized in the ways they preferred to work (e.g. remote vs. face-to-face collaboration), but also coached by one of the teaching staff. In several plenary meetings, the teams reported on their status as well as challenges and solutions. In both periods, the course was monitored and extensive surveys were carried out. We report on the findings as well as the lessons learned. For instance, while the program was very well-received, professional participants wished for more detailed coverage of theoretical concepts. A challenge faced by several teams during the second phase was a dropout of student members due to upcoming exams in other subjects.

Multidimensional aggregation is one of the most important computational building blocks and hence also a potential performance bottleneck in Online Analytic Processing (OLAP). In order to deliver fast query responses for interactive operations such as slicing, dicing, roll-up and drill-down, it is essential that aggregates along the relevant dimensions of a data cube can be calculated as efficiently as possible. General-purpose computing on graphics processing units (GPGPU) is a recent trend used in many computing domains with the potential for tremendous speedups through the massively data-parallel computation available on such devices. We present a GPU-based cube data structure and algorithms for fast multidimensional aggregation, implemented using Nvidia's CUDA framework. Our experimental tests show a substantial speedup over state-of-the-art sequential algorithms. Moreover, the performance gain is particularly high in cases exposing the weaknesses of traditional algorithms, i.e. when the number of base cells involved in an aggregation is large.

We propose extensions to the engagement taxonomy, based on a comparative review of several experiments carried out within that framework to assess the effectiveness of algorithm visualization.

We present a Java-based framework for the easy and on-the-fly creation of algorithm animations. Animations are created by sketching both the objects that should be animated and operations that should be performed on them. The IDEA (Interactive Domain rEcognition and Animation) framework combines the recognition of drawn structures, the interaction with these structures and their animation in order to achieve an effortless and natural creation of animations. The framework itself supports the creation of animations in arbitrary domains through a dynamic plug-in architecture, where so-called domain modules encapsulate all domain-specific semantics. Its application and potential is outlined by the help of three prototypical domain modules for linear lists, Petri nets, and the game Connect-four.

Compared to desktop PCs, mobile devices have several constraints such as restricted screen size and reduced interactivity. Hence, it is not clear if it is helpful to provide educational multimedia material for mobile media players. In this paper, we present a study of algorithm animations on mobile devices, i.e. an Apple iPod with video functionality. The results of our experiment indicate that despite limitations there is an actual learning effect and therefore mobile media players offer a potential to increase learning that deserves further investigation.

Collaborative learning with recorded lectures can be supported by allowing users to anchor notes in the documents and exchange them with others. While the traditional modality for annotation and discussion is text, there are a number of reasons in favor of supporting other modalities as well. We present an extension of a lecture-on-demand annotation and discussion system that allows learners to use spoken notes. Our main focus is on the development of a suitable user interface that facilitates the retrieval of speech data employing signal-processing algorithms while at the same time being simple and easy to use

Studies on the pedagogical value of algorithm visualizations have suggested that the level of learner engagement plays a crucial role for the learning outcome. We report on an evaluation carried out within a common research framework in order to compare the effects of three different levels of student interaction with visualizations, ranging from simply viewing animations to constructing visualizations from smaller algorithmic building blocks. The hypothesis that an increasing level of engagement leads to significantly better learning was not substantiated by the test results, which might be due to the influence of the accompanying lectures on the results. Suggestions for the design of future evaluations are given to better control such effects.

In this demo, we present new interfaces for interactive navigation in continuous, time-based multimedia files, such as video recordings. In contrast to common techniques for multimedia skimming, our approaches enable users to interactively navigate audio files along the timeline and support time-synchronized browsing in both audio as well as visual data streams.

Literature on speech skimming reports on techniques for playing speech backwards in a way that is still intelligible to the user. However, so far there is no empirical evidence for reasonable parameter settings of the respective algorithms and few examinations have been conducted to verify the usefulness of this feature for actual tasks. We present a user study testing different ways of backward skimming in relation to topic classification. Our evaluation shows a high classification performance and suggests implications for the design of the user interface.

Students' understanding of algorithms and data structures can be assessed with the help of construction tasks where students have to build their own visualizations. It has been claimed that active construction of visualizations results in better learning outcomes than passive viewing or merely changing the input of a visualized algorithm. This paper presents a system for the generation, execution, and evaluation of construction tasks. Its key feature is the flexibility in all three stages, ranging from fully automated to fully manual generation and evaluation as well as several different types of automatic feedback during the execution phase. Besides its use in daily teaching, the system can serve as a test bed for evaluations regarding the effectiveness of visualizations in the learning process.

Computer-based recording of live presentations has become a widespread method of producing learning materials in both higher education and companies. The success of such contents depends on their acceptance by the end-users working with the materials. From their point of view, the support of certain features and, thus, the choice of formats used for network delivery and replay are crucial issues. We present a list of key factors from the end-user's perspective and report on a user study carried out to evaluate a number of formats used for delivering recorded presentations to learners.

In This work we introduce new approaches for visual information seeking in video documents. We argue that a discrete visualization of continuous media streams via static keyframes is not sufficient to fulfill all information needs and therefore additional, more interactive techniques are required. We extend the concept of elastic interfaces, which has been used successfully with discrete documents, by applying it to continuous data streams. In addition, we propose a different integration of the elastic browsing functionality called "elastic panning". A first usability test is presented in order to prove its feasibility and to highlight its main advantages, namely the ability to access any position within a file independent of its length, and the smoother visual feedback during navigation compared to a standard slider interface.

Today's interfaces for time-scaled audio replay have limitations especially regarding highly interactive tasks such as skimming and searching, which require quick temporary speed changes. Motivated by this shortcoming, we introduce a new interaction technique for speech skimming based on the so called rubber-band metaphor. We propose an "elastic" audio slider which is especially useful for temporary manipulation of replay speed and which integrates seamlessly into standard interface designs. The feasibility of this concept is proven by an initial user study.

This paper describes a new way of creating multimedia documents combining recorded lectures and students' personal notes. Students use multimedia recordings of live presentations for revision and exam preparation. Based on the "Authoring on the Fly" concept we have developed a new system for multi-layer recording, where prepared materials, annotations made by the lecturer, and students' personal notes are captured during a live lecture. We describe the shift from the stand-alone lecture recording system to shared session recording and discuss the main issues which need to be tackled in the recording and the replay scenarios. We also point out the differences to other existing concepts. Finally we give a short vision of how the system could be used as a collaborative tele-recording tool in the future.