Novel models for self-adaptation and evolution in collective self-adaptive systems

• Thesis title in Czech: Novel models for self-adaptation and evolution in collective self-adaptive systems
• Thesis title in English: Novel models for self-adaptation and evolution in collective self-adaptive systems
• Academic year of topic announcement: 2021/2022
• Thesis type: dissertation
• Thesis language: angličtina
• Department: Department of Distributed and Dependable Systems (32-KDSS)
• Supervisor: prof. RNDr. Tomáš Bureš, Ph.D.
• Author: hidden - assigned and confirmed by the Study Dept.
• Date of registration: 22.11.2021
• Date of assignment: 22.11.2021
• Confirmed by Study dept. on: 22.11.2021

Guidelines

Modern complex systems, such as IoT (Internet of Things), CPS (Cyber-Physic Systems), swarm robots and edge/fog-cloud systems, often rely on autonomous components that operate together, share data, and collaborate in real-time. Such systems often come under the name “collective adaptive systems”, to underline the fact that they must continuously adapt and reconfigure at runtime to cope with the dynamically evolving environment and to deal with uncertainty caused by the environment, human in the loop and the overall complexity and decentralization. To manage the complexity of collective adaptive systems, novel architecture modelling approaches are required.

Traditional software architecture models are invariably static and fail to manage the dynamicity and the inherent uncertainty (both epistemic and aleatoric) which is connected with the environment, but also the system itself. The uncertainty that the modern collective adaptive systems deal with grows with the increasing requirements on smartness and ubiquity of these systems. The uncertainty reaches the point when the existing traditional methods and models for architecting these systems do not scale any more. The traditional approaches thus fail to provide suitable abstractions which would on one have allow modeling the system with all its dynamicity and on the other hand would ensure that the system is analyzable and correct behavior can be enforced from the model level.

Recently, there have been also approaches which employ machine-learning to tackle uncertainty and to control self-adaptation [1, 2, 3, 4, 5, 6]. the self-adaptation in modern collective adaptive systems. However, these approaches are still lacking on the modeling side and overall model-driven methodology.

The goal of this thesis is to investigate and propose novel models and architecture modeling methods that would address architecting of modern collective adaptive systems.

The focus should be on combining the architectural concepts of collective adaptive systems with machine learning. This means employing and elaborating abstractions that address the dynamicity (such as those proposed in [7,8]) and abstractions aiming at controlling the uncertainty (e.g. [9, 10, 11]) and aligning and extending them with abstractions reflecting approaches that target system optimization and evolution via machine learning (e.g. [2, 5, 6]).

References

1] T. R. D. Saputri and S.-W. Lee, “The Application of Machine Learning in Self-Adaptive Systems: A Systematic Literature Review,” IEEE Access, vol. 8, pp. 205948–205967, 2020, doi: 10.1109/ACCESS.2020.3036037.
[2] M. D. Sanctis, H. Muccini, and K. Vaidhyanathan, “Data-driven Adaptation in Microservice-based IoT Architectures,” in 2020 IEEE International Conference on Software Architecture Companion (ICSA-C), Mar. 2020, pp. 59–62. doi: 10.1109/ICSA-C50368.2020.00019.
[3] J. Cámara, H. Muccini, and K. Vaidhyanathan, “Quantitative Verification-Aided Machine Learning: A Tandem Approach for Architecting Self-Adaptive IoT Systems,” in 2020 IEEE International Conference on Software Architecture (ICSA), Mar. 2020, pp. 11–22. doi: 10.1109/ICSA47634.2020.00010.
[4] A. Palm, A. Metzger, and K. Pohl, “Online Reinforcement Learning for Self-adaptive Information Systems,” in Advanced Information Systems Engineering, Cham, 2020, pp. 169–184. doi: 10.1007/978-3-030-49435-3_11.
[5] J. Van Der Donckt, D. Weyns, F. Quin, J. Van Der Donckt, and S. Michiels, “Applying deep learning to reduce large adaptation spaces of self-adaptive systems with multiple types of goals,” in Proceedings of the IEEE/ACM 15th International Symposium on Software Engineering for Adaptive and Self-Managing Systems, New York, NY, USA, Jun. 2020, pp. 20–30. doi: 10.1145/3387939.3391605.
[6] T. Bureš, I. Gerostathopoulos, P. Hnětynka, and J. Pacovský, “Forming Ensembles at Runtime: A Machine Learning Approach,” in Leveraging Applications of Formal Methods, Verification and Validation: Engineering Principles, Cham, 2020, pp. 440–456. doi: 10.1007/978-3-030-61470-6_26.
[7] T. Bures et al., “A language and framework for dynamic component ensembles in smart systems,” Int J Softw Tools Technol Transfer, vol. 22, no. 4, pp. 497–509, Aug. 2020, doi: 10.1007/s10009-020-00558-z.
[8] R. Hennicker, M. Wirsing: A Dynamic Logic for Systems with Predicate-Based Communication. ISoLA (2) 2020: 224-242
[9] I. Gerostathopoulos, D. Skoda, F. Plasil, T. Bures, and A. Knauss, “Tuning self-adaptation in cyber-physical systems through architectural homeostasis,” Journal of Systems and Software, vol. 148, pp. 37–55, Feb. 2019, doi: 10.1016/j.jss.2018.10.051.
[10] T. Bureš et al., “Targeting uncertainty in smart CPS by confidence-based logic,” Journal of Systems and Software, vol. 181, p. 111065, Nov. 2021, doi: 10.1016/j.jss.2021.111065.
[11] I. Gerostathopoulos, D. Skoda, F. Plasil, T. Bures, and A. Knauss, “Tuning self-adaptation in cyber-physical systems through architectural homeostasis,” Journal of Systems and Software, vol. 148, pp. 37–55, Feb. 2019, doi: 10.1016/j.jss.2018.10.051.