Train the trainer: sustainable machine learning techniques
AI is taking up an ever more important role in software development, with core functionalities of software being replaced by machine learning models. However, the use of these kinds of adaptive techniques are not without drawbacks. They may use a lot of energy, introduce bias and have discriminatory effects.
Practical information:
Want to register?
- Prerequisites: Novice data scientists and AI engineers with a basic knowledge of Python, Scikit-learn, Pandas
- Price: free for SMEs, midcaps and nonprofits
Organised by:
What you will learn
- Students will learn to understand and improve the impact on people & the environment.
- How different machine learning models compare with regards to their impact on the environment.
- How to improve the performance of ML systems for minorities.
- How to assess fairness of ML techniques.
- How to lower the need for large amounts of data.
- Strengths and weaknesses of different evaluation metrics.
- Sources of bias that may introduce discrimination.
- How to improve the robustness of algorithmic outcomes.
- Awareness of potential secondary effects that are not modelled in data.
- Dangers of the “closed world” assumption.
Programme outline
The day looks as follows:
- 9:30 The AI mission is introduced to the students;
- 10:00 Students tackle the problem, in pairs, on their own;
- 11:00 The tutor helps students move forward and identifies potential problems;
- 12:30 The different approaches of the students are discussed, and quantitatively compared w.r.t fairness & impact on the environment.
13:00 BREAK
- 14:00 Techniques to assess and improve sustainability-related issues are explained.
- 15:00 Small exercises are done to empower the students.
- 16:30 Wrap-up and further reading
Practical
Cancellations: should be notified by email (Laetitia.dornano@vub.be). Cancellations notified 3 working days before the event are free of charge. In case of no-show or cancellation on the day itself, the fee of €105 will be charged to cover the costs. Replacement by a colleague is always possible provided that this is notified in advance by email (Laetitia.dornano@vub.be).
Teacher/speaker
Johan Loeckx
Johan Loeckx started as a micropreneur when he was 16. He received a MSc. in Electronic Engineering (2003), Msc. in AI (2004) and a PhD in 2010 from K.U.Leuven. He co-designed the Belgian health record encryption system, for which a patent was filed. He co-founded a Freinet school and is an active musician. Johan is professor and lab manager at the Artificial Intelligence Lab at the Vrije Universiteit Brussels.
Related courses
