PhD Position in Green-and-Explainable Machine Learning

Do you enjoy figuring out how things work? Do you have the ambition of going beyond the limits of what is currently known? Do you like working with others and occasionally engage in teaching activities? The Informatics Institute is looking for an enthusiastic PhD candidate. Your research is part of the Complex Cyber Infrastructure (CCI) research group.

There is growing concern that black-box Machine Learning is not always suitable as its explainability is limited and its energy consumption might be prohibitively expensive. Both issues are interrelated, as enhancing the explainability of ML (ensuring that decisions made by ML are sufficiently motivated to humans by that same ML) may require more complex ML models or additional steps – such as local approximations with an interpretable model or simulations of “what if” scenarios, thus leading to increased energy consumption. Conversely, recent comprehensive research into explainable-by-design deep learning systems indicates that feed-forward neural architectures are intrinsically suitable for explainability, at the cost of worsened classification metrics (accuracy).

Thus, it is important to understand to what extent explainable ML, accurate ML, and reducing the energy consumption of ML could be simultaneously optimised. This way, we can establish best practices to guide ML developers in the implementation of ML models and users in the selection of ML tools that are accurate, explainable, and energy-efficient at the same time.

What are you going to do?

You will design and develop techniques that support formulating recommendations for ML tools that focus on human and energy-related aspects, while also accounting for accuracy. You could scope the vast ML landscape to the energy consumption and explainability aspects of Deep Neural Networks (DNNs) and lay the foundational work for Green-and-Explainable Machine Learning, i.e., the set of models that are highly energy efficient, highly explainable, and highly accurate. Concretely:

1. Design energy recommendations for DNNs. Several key research questions are:

• To what extent can DNNs be grouped in terms of their energy consumption and accuracy?
• To what extent do extra-functional properties, such as sensitivity to hyperparameters and (re)training time, influence the grouping?

2. Design explainability recommendations for DNNs. Several key research questions are:

• How can we describe explainability from different user perspectives?
• What would be the design of an actionable framework for the interaction between explainer and explainee?
• To what extent does the explainability requirement lead to more complex ML architectures?
• How can we efficiently characterise the relationship between accuracy and explainability?

3. Combine energy and explainability recommendations into Green-and-Explainable ML. Several key research questions are:

• What is the relation between DNNs’ energy consumption and their explainability?
• How can we efficiently explore the space of DNNs in terms of the three characteristics: energy consumption, accuracy, and explainability?
• What are suitable metrics for each of these so that we can compare DNNs?

What do you have to offer?

Your experience and profile:

• MSc in CS, AI or a closely related field;
• Strong interest in ML, energy efficiency, and explainability. Previous experience in these fields is a plus;
• Enthusiasm for the research process: studying research papers, solving complex problems, applying creative thinking, evaluation, reflection and disseminating findings via writing and oral presentations;
• The ability to critically analyse abstract models as well as concrete implementations;
• The ability to work individually as well as effectively in a team;
• Fluency in written and spoken English.

Our offer

A temporary contract for 38 hours per week for the duration of 4 years (the initial contract will be for a period of 18 months and after satisfactory evaluation it will be extended for a total duration of 4 years). The preferred starting date can be discussed, ideally before the 1st of September 2024. This should lead to a dissertation (PhD thesis). We will draft an educational plan that includes attendance of courses and (international) meetings. We also expect you to assist in teaching undergraduates and master students.

The gross monthly salary, based on 38 hours per week and dependent on relevant experience, ranges between € 2,770 in the first year to € 3,539 in the last year (scale P). UvA additionally offers an extensive package of secondary benefits, including 8% holiday allowance and a year-end bonus of 8.3%. The UFO profile PhD Candidate is applicable. A favourable tax agreement, the ‘30% ruling’, may apply to non-Dutch applicants. The Collective Labour Agreement of Universities of the Netherlands is applicable.

Besides the salary and a vibrant and challenging environment at Science Park we offer you multiple fringe benefits:

• 232 holiday hours per year (based on fulltime) and extra holidays between Christmas and 1 January;
• Multiple courses to follow from our Teaching and Learning Centre;
• A complete educational program for PhD students;
• Multiple courses on topics such as leadership for academic staff;
• Multiple courses on topics such as time management, handling stress and an online learning platform with 100+ different courses;
• 7 weeks birth leave (partner leave) with 100% salary;
• Partly paid parental leave;
• The possibility to set up a workplace at home;
• A pension at ABP for which UvA pays two third part of the contribution;
• The possibility to follow courses to learn Dutch;
• Help with housing for a studio or small apartment when you’re moving from abroad.