Hasso-Plattner-Institut25 Jahre HPI
Hasso-Plattner-Institut25 Jahre HPI

Online Learning for Self-Healing and Self-Optimization (Sommersemester 2020)

Dozent: Prof. Dr. Holger Giese (Systemanalyse und Modellierung) , Christian Medeiros Adriano (Systemanalyse und Modellierung) , Sona Ghahremani (Systemanalyse und Modellierung)

Allgemeine Information

  • Semesterwochenstunden: 4
  • ECTS: 6
  • Benotet: Ja
  • Einschreibefrist: 06.04.2020 - 22.04.2020
  • Lehrform: Projekt / Seminar
  • Belegungsart: Wahlpflichtmodul
  • Lehrsprache: Englisch

Studiengänge, Modulgruppen & Module

IT-Systems Engineering MA
  • IT-Systems Engineering
    • HPI-ITSE-K Konstruktion
  • IT-Systems Engineering
    • HPI-ITSE-M Maintenance
  • SAMT: Software Architecture & Modeling Technology
    • HPI-SAMT-K Konzepte und Methoden
  • SAMT: Software Architecture & Modeling Technology
    • HPI-SAMT-T Techniken und Werkzeuge
  • SAMT: Software Architecture & Modeling Technology
    • HPI-SAMT-S Spezialisierung
  • OSIS: Operating Systems & Information Systems Technology
    • HPI-OSIS-K Konzepte und Methoden
  • OSIS: Operating Systems & Information Systems Technology
    • HPI-OSIS-T Techniken und Werkzeuge
  • OSIS: Operating Systems & Information Systems Technology
    • HPI-OSIS-S Spezialisierung
Data Engineering MA


For the adaptation of architecture-based self-adaptive systems, offline learning (training) can be used to obtain prediction functions for the impact of decisions on the system utility (utility-changes) [1] rather than analytically constructing these functions. This is often challenging or even not feasible, because of non-linearities, complex dynamic architectures, or simply black-box models which make designing these functions often not trivial. Therefore, an offline learned (trained) prediction cannot reflect that the desirability and benefits of specific decisions may change over time, because the prediction functions will keep guiding the system as if the situation has not evolved. Instead, it would be highly desirable to be able to learn such prediction functions for the impact of decisions on the system utility online such that the evolution of the system and its context is considered. Then, the long-term behavior concerning the self-adaptation would adjust to the environment changes and the corresponding system adaptive decisions rather than stick to the past when the training data had been collected.


In this project seminar, we want to explore with you, based on an existing example system in the form of an e-commerce shop and a related experimentation environment, how such a transition from an offline learning (training) scenario to an online scenario can be achieved. We will study different techniques for online learning and design a prototypical example to compare their pros-and-cons.


The example system is mRUBiS [2], a simulated online market place modeled after eBay that is equipped with self-healing and self-optimizing capabilities. This means that the system has built in capabilities to repair itself as a response to runtime failures (e.g., component crashes) and optimize itself as a response to the changes in the environment (e.g., increase in user demand) [1,3].

Among the various machine learning models [4], will investigate a few on-line techniques that have been successfully applied to self-adaptive systems. Examples of these techniques are reinforcement learning [5], concept drift detection [6], and time series prediction [7].


[1] Ghahremani, S., Adriano, C. M., & Giese, H. (2018, September). Training prediction models for rule-based self-adaptive systems. In 2018 IEEE International Conference on Autonomic Computing (ICAC) (pp. 187-192). IEEE.

[2] Thomas Vogel. 2018. MRUBiS: an exemplar for model-based architectural self-healing and self-optimization. In Proceedings of the 13th International Conference on Software Engineering for Adaptive and Self-Managing Systems (SEAMS ’18). Association for Computing Machinery, New York, NY, USA, 101–107. DOI:https://doi.org/10.1145/3194133.3194161

[3] Sona Ghahremani, Holger Giese, and Thomas Vogel. 2020. Improving Scalability and Reward of Utility-Driven Self-Healing for Large Dynamic Architectures. ACM Trans. Auton. Adapt. Syst. 14, 3, Article 12 (February 2020), 41 pages. DOI:https://doi.org/10.1145/3380965

[4] Russell, Stuart, and Peter Norvig. "Artificial intelligence: a modern approach." (2002).

[5]  D. Kim and S. Park, “Reinforcement learning-based dynamic adaptation planning method for architecture-based self-managed software,” in 2009 ICSE Workshop on Software Engineering for Adaptive and Self-Managing Systems, May 2009, pp. 76–85.

[6] Dinithi Nallaperuma, Rashmika Nawaratne, Tharindu Bandaragoda, Achini Adikari, Su Nguyen, Thimal Kempitiya, Daswin De Silva, Damminda Alahakoon and Dakshan Pothuhera. Online incremental machine learning platform for big data-driven smart traffic management. In IEEE Transactions on Intelligent Transportation Systems, Vol. 20(12):4679--4690, IEEE, 2019

[7] M. Zuefle et al., "Autonomic Forecasting Method Selection: Examination and Ways Ahead," 2019 IEEE International Conference on Autonomic Computing (ICAC), Umea, Sweden, 2019, pp. 167-176.

Lern- und Lehrformen

The course is a project seminar, which has an introductory phase comprised by initial short lectures. After that, the students will work in groups on jointly identified experiments applying specific solutions to given data sets and finally prepare a presentation and write a report about their findings concerning the experiments.

There will be an introductory phase to present basic concepts for the theme including the necessary foundations for online machine learning.


Announcement Regarding COVID-19
Because of the COVID-19 outbreak, we will organize all meetings as online meeting initially and will switch back to regular meetings later when the circumstances permit so and all the participants agree with it.


We will grade the group's experiments (50%), reports (40%), and presentations (10%). Participation in the project seminar during meetings and other groups' presentations in the form of questions and feedback will also be required.


 After the introductory phase with few initial short lectures, we will identify the group topics and then there will be regular individual feedback rounds of the groups with their supervisors. In addition, there will also be regular meetings during the semester for the whole project seminar to discuss the progress of all groups and open questions in general.


The first lecture will be held on 29.04.20 , at 15:15 - 16:45 via a Zoom meeting.

The interested students in addition to registering for the project seminar, need to contact us here to obtain the Zoom meeting credentials for attending the lecture.