TOPIC 1: Using Temporal Logic for AI-governed Multi-Agent Systems
Introduction: Multi-Agent Systems consist of multiple autonomous entities with individual goals, acting on a shared environment in an intertwined manner. To steer their collective behavior whilst maintaining autonomy, it is necessary for a governing instance to reason about agent strategies and to predict their future actions, which might not just be a straight-forward extrapolation of the current observations but also exhibit a temporal evolution of goals and strategies.
Goal and Objective: The goal of this seminar is to find out how temporal logic—in particular Linear Temporal Logic, but also related or extended logics—can be applied to create an AI-based governance capable of approximating an agent’s (potentially dynamic) strategic behavior from observed actions and subsequently choosing suitable counter-actions leading to a given system goal.
The expected result is a research paper showing relevant existing approaches and their respective benefits and drawbacks, as well as own ideas on how this problem can be defined, formalized and solved.
TOPIC 2: Dialogue Systems for Software Engineering: A survey
Goal and Objective: Overview and discussion of different state-of-the-art approaches that apply dilogue systems in the field Software Engineering and comparison between them
TOPIC 3: A Performance Test for a novel IoT-centric Reuse Approach
Introduction: As people are diverse, so is the IoT world. The IoT diversity can be seen in their interfaces and payloads providing similar functionalities. For example, a light switch that controls a lamp (e.g. Phillips Hue) must know which payload (e.g. JSON) to send to which function (e.g. HTTP interface). However, if multiple device manufacturers (i.e. light switch and lamp) do not collaborate, glue code to integrate both devices is required.
Goal and Objective: In this seminar, you will evaluate a novel architecture-centric reuse approach for generating glue code automatically. Therefore, you will create and execute a performance data set for the knowledge-driven architecture composition method. At the end of your seminar, you have gain hands-on experience in performing a structured evaluation for your future studies (e.g. bachelor/master thesis). Furthermore, you gain first insights in IoT challenges from a software engineering perspective.
TOPIC 4: Comparison of Similarity Measures in their applicability to Vector Sequences in a network
Introduction: To achieve meaningful clustering and classification appropriate similarity measures are required. The key challenge is to choose the right similarity measure. Vector Sequence Networks are a new prototype approach to give a condensed view on the workflow of complex software systems.
Goal and Objective: Your task in this seminar will be to examine and evaluate the applicability of current similarity measures to vector sequence networks. The overarching goal is to find appropriate clustering procedures for similar sequences inside the network.
TOPIC 5: Static Code Analysis (SCA)
Introduction: Static code analysis (SCA) is a common technique for checking programs for flaws security vulnerabilities. In contrast to other approaches (e.g., dynamic analysis), the static analysis is feasible in an early stage of the development lifecycle.
Goal and Objective: This seminar aims to examine SCA based on categories in which different approaches of SCA (tools) can get clustered. The starting papers offer a taxonomy  of criterion to start from and benchmarks on how to evaluate static analysis tools . The seminarist should:
1. Introduce SCA in general
2. Compare the different (SCA) approaches found in the literature
3. Give details on their advantages and disadvantages
TOPIC 6: Explainable Artificial Intelligence (XAI)
Introduction: Currently, the explainability of AI techniques (XAI) is a widely discussed topic, both in AI research and in society. Missing or insufficient explanatory capabilities of AI-based systems are a serious limitation for the use of these techniques in many applications domains for various reasons.
Goal and Objective: In this seminar you will familiarize yourself with XAI approaches for AI techniques (like artificial neural networks) based on scientific literature as well as your own exploration.
TOPIC 7: Model-Driven Development Approaches for Microservices Architectures
Introduction: There is a recent trend in the adoption of microservice architectures. The design and implementation of such architectures, however, is a complex task because of the high amount of involved autonomous stakeholders. This is especially true, if microservices are reused in many different use cases, as many functional and non-functional requirements have to be considered in a consistent way across the services, to achieve a use case specific API.
Model-driven development approaches may be beneficial in that regard, as they could reduce the involved complexity in such development processes.
Goal and Objective: For this seminar paper you should provide an overview and classification of the kinds of model-driven development approaches which are suited for the distributed development processes of microservice architectures which consider all the (non-)functional requirements to compose an API for different use cases. Additionally you should provide the advantages and disadvantage of the types of the found approaches.
TOPIC 8: Random Forests in Dynamic Feature Spaces
Introduction: Random Forests have become a go to method for a lot of machine learning problems. Originating from Decision Trees, they were originally intended for offline learning settings. However, this seminar topic focuses on Random Forests for Online Learning. Random Forests provide good results consistently by combining several smaller Decision Trees for prediction. However, not only their performance but also their robustness make them interesting for an online learning problem.
Goal and Objective: In this topic, you will explore the state of the art literature for Random Forests and evaluate their fit for an online learning setting with regard to a dynamic feature spaces. In such dynamic feature spaces features are missing occasionally, vanish or new features emerge.
TOPIC 9: Goal- and Plan Recognition: State of the Art
Introduction: Recently, activity recognition became very popular, especially in the fields of ubiquitous and pervasive computing. The major application domains for these approaches are in health- and elderly care. However, these approaches only allow a supporting system to react to recognized incidents (for example a recognized fall). To enable such systems to also act proactively, it has to be able to reason about current goals and plans of an observed agent.
Goal recognition is defined as the problem to recognize the current goal(s) of an agent given a sequence of observations of this agent. Respectively, plan recognition is defined as the problem of recognizing the entire plan of an agent given a sequence of observations of the agent.
Goal and Objective: Over the past decades several different approaches to goal- and plan recognition evolved. Thereby, some rely on symbolic methods (like classical AI planning), whereas others make use of statistical methods (e.g., Bayesian Networks). The objective of this seminar is to provide an overview of the current state-of-the-art in the field of goal- and plan recognition and also highlight shortcomings of current approaches.
|Institute for Enterprise Systems||L15, firstname.lastname@example.org||Office Management|
|Schloss||68131 Mannheim||+49 621 181-3560|