Call for Projects -- TOPML Phase 2
Deadline: 31st of December, 2024
The TOPML project studies the interactions among desirable properties of machine learning: What is the effect of your concept of privacy on the fairness of models and decisions? How do privacy and fairness impact the resource efficiency of methods? (This surely requires more computation.) If transparency is required, how does this interact with the wish for privacy? According to systems and software engineering, privacy, fairness, resouce efficiency, and transparency are all non-functional properties of machine learning. The function is learning: getting measurably better by experience, possibly building models of oneself, a domain, or the world along the way. Privacy, fairness, resource efficiency, and transparency are constraints under which learning should take place. It is known from systems and software engineering that these constraints or non-functional properties often are conflicting. This is also case in machine learning. The topic of TOPML is to study these conflicts.
If these research questions sound intriguing to you, consider submitting a project proposal for the Phase 2 of the TOPML project. TOPML stands for "Trading Off non-functional Properties of Machine Learning" and is financed by the Carl Zeiss Foundation (2022-2028).
Researchers in TOPML research methodologies in Computer Science, Law and Ethics to achieve sensible trade-offs between critical "TOP" properties of ML and AI systems: fairness, interpretability, privacy and resource efficiency. The project started in 2022 and currently involves 6 PhD students and over 10 faculty members at the departments of Computer Science, Philosophy and Law. In 2025, 4 new subprojects are scheduled to start. If you have a research question in this topic, we would be thrilled to hear about it! If selected, you would then receive funding to hire an early-stage researcher (PhD student or Post-Doc, TV-EL 13 100%), or be involved in co-supervision if that would be your preference.
Why Join TOPML
* TOPML researchers have published in top-ranked conferences and journals, even at an early stage in their career.
* You would join an interdisciplinary network of researchers, mostly but not exclusively early-stage, with a clear focus on themes of responsible development of AI methods and application.
* TOPML has a dedicated budget to outreach and networking, and can support your endeavors in further establishing your network.
Themes and Topics
We are looking for projects that deal with the intersection of two or more TOP properties from an application-driven or methods-driven perspective, and we welcome interdisciplinary perspective from other scientific communities. A project skeleton is available here.
Here is a non-exhaustive sketch of the TOP properties in TOPML:
* Fairness: ensuring that models do not discriminate against certain groups of people, not only at a statistical level but also e.g. during the data collection process. Studying the impact of AI systems on underprivileged communities.
* Transparency: ensuring that we can make sense of AI systems, in all their components including the data. Developing new ways to understand the complex interactions between humans and "black box" systems.
* Privacy: ensuring that AI development will not infringe people's privacy rights, under different definitions such as differential privacy, secure multi-party computation. Studying how privacy risk is perceived and evaluated by people.
* Resource-Efficiency: the development of new techniques and methods that can lead to sustainable development of AI systems. Understanding the ramifications of AI's environmental impact for people and communities.
Application Procedure and FAQs
Who can apply? Any faculty member at JGU, even if on a fixed-term contract. We would prefer that you be employed at least until 2028, or have concrete plans to remain at JGU. We require that every proposal has at least one PI or co-PI based at the Computer Science (Informatik) department at JGU, due to the nature of our funding.
How can I apply? Download the project proposal skeleton here and then send your proposal via email at topml-management@uni-mainz.de.
Fourth year funding. If you intend to use the funding to hire a PhD student, we kindly ask you to clarify whether you would have the resources to fund their fourth year of studies, as TOPML cannot cover it at this time.
I am a student, can I apply for a PhD? We suggest you reach out to one of the TOPML PIs (or any faculty at JGU which fits the requirements above) and discuss your interest in the project. Then, you could potentially contribute to the project proposal that they are going to write.
Deadline is the 31st of December, 2024.
Timeline
* Project Proposal Deadline: 31st of December, 2024
* Final Notification of Acceptance: 1st of May, 2025
* Start of Employment for new hires: 1st of July, 2025