The Vrije Universiteit Amsterdam is seeking young talents interested in doing a PhD trajectory titled Conversations with the Architect – Large Language Models for Designing Energy-Efficient ACPSs.
Here is some info about the position:
- Required qualification: Master’s degree in Computer Science, Robotics, or related study
- Salary: €3,400/month (gross) + mobility allowance + family allowance
- Start date: February 2025
- Duration: 4 years
- University: Vrije Universiteit Amsterdam
- Supervisor: Ivano Malavolta
- Research group: Software and Sustainability - S2
- Mobility: possibility to spend up to 12 months in partner universities in Spain, Norway, Italy, and Switzerland.
- Application link: apply here and choose PhD project 12
- Application deadline: October 15, 2024
- Further info: https://www.innoguard.eu
At the end of this message, you will find more details about the project.
Cheers and spread the voice, thank you!
Ivano
The project aims to exploit LLMs to better support ACPS developers in developing energy-efficient software for ACPSs. At the core of the project lies the concept of architectural tactic, i.e., design decisions that influence the achievement of system qualities and can be reused across projects. For example, a tactic for energy efficiency is to offload computationally-expensive algorithms from a battery-powered robot to the Cloud. Tactics have been studied and successfully used in areas like big-data cybersecurity and Cloud-based systems, but they have never been used in conjunction with LLMs. LLMs will be used for recommending architectural tactics for ACPSs either conversationally to developers at development time or programmatically to the ACPS itself at runtime. Architectural tactics in ACPSs are highly domain- and context-dependent, they can have side effects, and can come with non-trivial complex trade-offs. The ability to process large amounts of data and internalise implicit cross-domain knowledge of LLMs makes them excellent candidates for managing architectural tactics. The objectives of the project include:
- to build a knowledge base of concrete, repeatable, and quantifiable architectural tactics for energy-efficient ACPSs,
- to integrate such knowledge base into different LLMs for providing timely recommendations about applicable tactics at development time,
- to develop an approach for automatically applying and composing architectural tactics in the context of ACPSs,
- to develop a self-adaptive approach where ACPSs autonomously apply tactics to their software architecture in response to changes in their measured quality of service (e.g., energy degradation).
Examples of scientific studies related to the topic:
- Ivano Malavolta, Katerina Chinnappan, Stan Swanborn, Grace Lewis, Patricia Lago (2021). Mining the ROS ecosystem for Green Architectural Tactics in Robotics and an Empirical Evaluation. In Proceedings of the 18th International Conference on Mining Software Repositories, MSR, pp. 300–311, New York, NY. (pdf)
- Milica Dordevic, Michel Albonico, Grace Lewis, Ivano Malavolta, Patricia Lago (2023). Computation Offloading for Ground Robotic Systems Communicating over WiFi - An Empirical Exploration on Performance and Energy Trade-offs. Empirical Software Engineering, 28(140), pp. 1573–7616. (pdf)
- Ye Yuan, Jingzhi Zhang, Zongyao Zhang, Kaiwei Chen, Jiacheng Shi, Vincenzo Stoico, Ivano Malavolta (2024). The Impact of Knowledge Distillation on the Energy Consumption and Runtime Efficiency of NLP Models. In 3rd IEEE/ACM International Conference on AI Engineering - Software Engineering for AI, CAIN 2024, Lisbon, Portugal, April 14-15, 2024. (pdf)