InES Seminar Spring Semester 2021

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.

Starting Papers:

• A. Artale (2010): Formal Methods, Lecture III: Linear Temporal Logic (Lecture Slides). http://web.iitd.ac.in/~sumeet/slide3.pdf
• J. Tumova, D. Dimarogonas (2016). Multi-Agent Planning under Local LTL Specifications and Event-Based Synchronization. Automatica 70:239–248.
• R. Alur, T. Henzinger, O. Kupferman (2002). Alternating-time temporal logic. J. ACM 49, 5 (September 2002), pp.672–713.

Introduction: In one of InES’ current research projects, a fleet of drones and transport vehicles is used to deliver goods from warehouses to customers. The corresponding optimization problem can be solved with two distinct models: The drones can make their own decisions and cooperate in the overall task, or everything is planned and controlled by a central governing authority.

We want to examine the second approach and find out how to design and set up a central controller which is able to manage such a system, given some constraints and an overall delivery objective. This can include machine learning techniques, but also plain optimization algorithms or heuristics.

Goal and Objective: The goal of this seminar is to investigate approaches and algorithms which are well-suited to solve the above-mentioned real-world problem and its specific challenges, and to recommend an approach for implementation in the project.

The expected result is a research paper listing and comparing relevant existing approaches with their respective benefits and drawbacks in the given context.

Starting Papers:

• Troudi et al. (2018). Sizing of the Drone Delivery Fleet Considering Energy Autonomy. Sustainability 2018.
• Bemposta et al. (2020). Remote Management Architecture of UAV Fleets for Maintenance, Surveillance, and Security Tasks in Solar Power Plants. Energies 2020.
• Hult et al. (2020). Optimisation-based coordination of connected automated vehicles at intersections. Vehicle System Dynamics 2020.

Introduction: In one of InES’ current research projects, an autonomous fleet of drones is used to deliver goods from warehouses to customers. The drones can make use of transport vehicles to be carried the better part of the way, which saves energy and is much faster than flying the entire distance. The drones need to make intelligent decisions and act autonomously, such that the system does not depend on a central controlling instance. As a consequence, the entire system is modeled as a Governed Multi-Agent System, and the drone behavior is an essential part of its functionality.

Goal and Objective: The goal of this seminar is to find out which ML techniques are suitable for enabling the drones to act effectively and efficiently in the scenario defined above. You will be given precise information about the (self-interested) goals of the drone as well as the constraints imposed by the use case and the general environment. Within these boundaries, your task will be to identify and compare ML approaches which can (and subsequently will) be implemented in the controlling unit of a real-world drone.

The expected result is a research paper listing and comparing relevant existing approaches with their respective benefits and drawbacks in the given context.

Starting Papers:

• Hoen et al. (2006). An Overview of Cooperative and Competitive Multiagent Learning. Learning and Adaption in Multi-Agent Systems, First International Workshop (LAMAS 2005).
• Cui et al. (2019). The Application of Multi-Agent Reinforcement Learning in UAV Networks. IEEE International Conference on Communications Workshops (ICC Workshops), Shanghai, China, 2019.
• Albrecht and Stone (2017). Multiagent Learning Foundations and Recent Trends. www.cs.utexas.edu/~larg/ijcai17_tutorial/multiagent_learning.pdf.

Introduction: Data-driven approaches for reducing manual effort and increasing automation have gained popularity for various application areas over the past years. Current research started to adopt these approaches for supporting different tasks in software engineering e.g. automatic method and commit message generation or semantic code retrieval approaches, among many others. Recently, researchers started to propose data-driven approaches to support users during GUI prototyping and therefore reduce required time, effort and skills to create well-designed GUI prototypes. The goal of this seminar work is to provide a clear overview and discussion of data-driven approaches that provide assistance to users for GUI prototyping in various ways.

Goal and Objective: Overview and discussion of different data-driven state-of-the-art approaches that provide assistance to users for GUI prototyping in various ways.

Starting Papers:

• Lee, C., Kim, S., Han, D., Yang, H., Park, Y. W., Kwon, B. C., & Ko, S. (2020, April). GUIComp: A GUI Design Assistant with Real-Time, Multi-Faceted Feedback. In Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems (pp. 1–13).
• Kolthoff, K., Bartelt, C., & Ponzetto, S. P. (2020, September). GUI2WiRe: Rapid wireframing with a mined and large-scale GUI repository using natural language requirements. In 2020 35th IEEE/ACM International Conference on Automated Software Engineering (ASE) (pp. 1297-1301). IEEE.

Introduction: Deep Learning is continuously showing its enormous success in a variety of complex tasks related to different fields, e.g., computer vision and natural language processing. However, the performance of Deep Learning still highly depends on the input data. This leads to limitations in certain application domains with low data availability or quality. Integrating prior knowledge into the specification and learning of a neural network offers potential to overcome these limitations.

Goal and Objective: In this seminar you will familiarize yourself with approaches to combine Deep Learning with prior knowledge based on scientific literature and your own exploration. You should provide an overview and classification of current approaches as well as describing their advantages and disadvantages.

Starting Papers:

• H. D. Gupta and V. S. Sheng, “A Roadmap to Domain Knowledge Integration in Machine Learning,” in 2020 IEEE International Conference on Knowledge Graph (ICKG), Aug. 2020, pp. 145–151, doi: 10.1109/ICBK50248.2020.00030.
• C. Yin, R. Zhao, B. Qian, X. Lv, and P. Zhang, “Domain Knowledge Guided Deep Learning with Electronic Health Records,” in 2019 IEEE International Conference on Data Mining (ICDM), Nov. 2019, pp. 738–747, doi: 10.1109/ICDM.2019.00084.
• N. Muralidhar, M. R. Islam, M. Marwah, A. Karpatne, and N. Ramakrishnan, “Incorporating Prior Domain Knowledge into Deep Neural Networks,” in 2018 IEEE International Conference on Big Data (Big Data), Dec. 2018, pp. 36–45, doi: 10.1109/BigData.2018.8621955.
• R. Rai and C. K. Sahu, “Driven by Data or Derived Through Physics? A Review of Hybrid Physics Guided Machine Learning Techniques With Cyber-Physical System (CPS) Focus,” IEEE Access, vol. 8, pp. 71050–71073, 2020, doi: 10.1109/ACCESS.2020.2987324.

Introduction: Over the past decade the task of recognizing human activities from observed (sensor-) data was extensively studied. Basically the approaches that solve this task can be divided into two groups: Knowledge-based approaches and data-driven approaches. As the names suggest, knowledge-based approaches rely on handcrafted domain knowledge to solve the task, whereas data-driven approaches try to learn to solve the task from historical training data. In addition, there is also a third group of approaches that try to combine knowledge-based and data-driven techniques to hybrid approaches.

Goal and Objective: The aim of this seminar is to provide an overview and comparison of existing hybrid approaches to activity recognition. Furthermore, the benefits and weaknesses of the different approaches should be critically discussed.

Starting Papers:

• Riboni, Daniele, and Claudio Bettini. „COSAR: hybrid reasoning for context-aware activity recognition.“ Personal and Ubiquitous Computing 15.3 (2011): 271–289.
• Riboni, Daniele, et al. „SmartFABER: Recognizing fine-grained abnormal behaviors for early detection of mild cognitive impairment.“ Artificial intelligence in medicine 67 (2016): 57–74.
• Helaoui, Rim, Daniele Riboni, and Heiner Stuckenschmidt. „A probabilistic ontological framework for the recognition of multilevel human activities.“ Proceedings of the 2013 ACM international joint conference on Pervasive and ubiquitous computing. 2013.
• Civitarese, Gabriele, et al. „NECTAR: Knowledge-based collaborative active learning for activity recognition.“ 2018 IEEE International Conference on Pervasive Computing and Communications (PerCom). IEEE, 2018.