This work presents a prototype of a learning tool that was developed to remedy current challenges and issues in the field of block-based programming for ML and AI education. Based on recent research strengths and weaknesses of current solutions were identified. Most of tools don't follow defined guidelines, don't meet certain requirements, or they neglect important points such as simple design or manageable black-box consideration. Based on these and other criteria we developed a block-based programming learning tool prototype that consists of three different learning games about supervised and reinforcement learning. Unsupervised learning might be included in further steps. The tool provides an introduction to algorithms such as decision trees, image classification, and agent based systems. These were identified as common algorithms to be taught in the area of ML and AI. Finally, we talk about current and expected results of this work. First study shows that the prototype we developed made a noticeable impact for programming newbies. Furthermore, younger people seem to enjoy using the block-based tool more than the older ones. The main goal of this work is to create a block-based programming environment that represents complex ML and AI concepts in a simplified way and makes them tangible for the students.

Problem solving plays a central role in computer science classes, whereby the problems to be analyzed are often already available in encoded form. The initial process where the learners have to parse information into symbols using adequate representations, has hardly been considered systematically so far. However, encoding development is essential for the understanding of sign processes and for general education. This paper therefore presents the development and evaluation of a teaching unit called “Cup Song Encoding” designed to model information encoding using a percussion song as an example. It simultaneously imparts coding theory concepts and fosters Computational Thinking of students. The material of the teaching unit is available online. The teaching unit was successfully tested with over 200 students in different age groups in K-12 during which the students' solutions were recorded and examined. The outcomes of the unit demonstrate the potential for encoding modeling to be a valuable addition to K-12 computer science education. The results also show how Computational Thinking methods can be integrated in computer science classes.

This work identifies the capabilities of a block-based programming approach for learning machine learning concepts. It focuses on the following overarching research question: “How can block-based programming tools be used to facilitate the understanding and application of machine learning concepts in K-12 education?”. To answer this question, guidelines for conducting a systematic literature review are followed, resulting in the study of 17 different learning tools. These tools are examined for their technical nature, content coverage, design features, intelligibility, evaluations, and deployability. The findings suggest that the vast majority of tools focus on a high-level representation of classification models that children can create in an extended version of the Scratch programming environment. By this, however, only one facet of machine learning is addressed, and deeper insights into the underlying functions are not provided. In addition, technical, linguistic, and conceptual barriers to the design of tools and the wider curricula become apparent.

Automated Assessment Systems (AAS) are increasingly used in large computer science lectures to evaluate student solutions to programming assignments. The AAS normally carries out static and dynamic analysis of the program code. In addition, simple forms of learning analytics can often be generated quite easily. However, structural analyses and comparison of solutions for larger sets of student programs are, in many cases, complicated and time-consuming. In this article, we introduce a methodology with which thousands of programs can be analyzed in less than a second, for example, to search for the use of certain control structures or the application of recursion. For this purpose, we have developed a software that creates a structural representation for each programming solution in the form of a TGraph, which is inserted into a graph database using Neo4j. On this database, we can search for structural features by queries in the language Cypher. We have tested this methodology extensively for Java programs, measured its performance, and validated the results. Our software can also be applied to programs in other programming languages, such as Scratch. Additionally, we plan to make our software available to the community.

To overcome the shortage of computer specialists, there is an increased need for correspondent study and training offers, in particular for learning programming. The automated assessment of solutions to programming tasks could relieve teachers of time-consuming corrections and provide individual feedback even in online courses without any personal teacher. The e-assessment system JACK has been successfully applied for more than 12 years up to now, e.g., in a CS1 lecture. However, there are only few solid research results on competencies and competence models for object-oriented programming (OOP), which could be used as a foundation for high-quality feedback.In a joint research project of research groups at two universities, we aim to empirically define competencies for OOP using a mixed-methods approach. In a first step, we performed a qualitative content analysis of source code (sample solutions and students’ solutions) and as a result identified a set of suitable competency components that forms the core of further investigations. Semi-structured interviews with learners will be used to identify difficulties and misconceptions of the learners and to adapt the set of competency components. Based on that we will use Item Response Theory (IRT) to develop an automatically evaluable test instrument for the implementation of abstract data types. We will further develop empirically founded and competency-based feedback that can be used in e-assessment systems and MOOCs.

We report on the first steps of KETTI, a project that aims towards the development of a competence model for undergraduate teaching assistants (UTAs) in computer science. Using qualitative methods, we obtained a classification of existing designs for teaching that employ UTAs; some of the observed factors directly influence the methodological decision space of UTAs. We developed and implemented a UTA training scheme designed to foster student-oriented teaching in recitation sessions along with an instrument to gauge the effects of this instruction on a variety of psychometric scales. We report results from a small-scale pilot study at three institutions showing positive effects on teaching-related beliefs and self-efficacy.

The prerequisite knowledge regarding Computer Science (CS) varies strongly among freshmen at university and it seems advisable to compensate for these differences before the first lecture starts. Massive open online courses (MOOCs) might represent a possible solution. We therefore designed and developed a MOOC (called “LOOP: Learning Object-Oriented Programming”) which provides a gentle introduction to computational thinking and object-oriented concepts before the programming part. In addition to the common quizzes, we developed various we-based interactive exercises to enable the learners to experiment and interact directly with the presented concepts. Furthermore, we implemented programming exercises with constructive feedback for the learners using a web-based integrated development environment and additionally an automatic grading system. The target group of the course are prospective students of science or engineering that are due to attend CS lessons in their first terms. The course was conducted as a prototype with a limited number of participants. In a concluding survey, the participants submitted textual feedback on the course; some of them proposed specific improvements for the employed interactive exercises. Yet, the overall feedback was encouragingly positive. In this paper, we describe the design and the development of the course, as well as our initial results.