Abstract: Recently, academics as well as policy makers have written many papers, on responsible data science / AI. Moreover, many open-source packages for bias dashboards or tools for `fairness’ have been proposed. This session aims to provide attendees a broad overview as well as the specific technical background to use the available ` fairness’ tools. In addition, a governance framework describing the precise responsibilities of data scientists will be discussed.
1. The session starts with an overview of examples of data science applications that are considered unfair / unethical, as well as the main `driving sources’.
2. Hereafter, an overview of proposed policies and frameworks, as well as upcoming regulation is provided.
3. Next, the discussion will concentrate on `fairness’. An overview of the (academic) literature will be provided including an in-depth discussion of the similarities and dissimilarities between different approaches. The concepts will be illustrated by an application of open-source Python packages that provide so-called `bias-dashboards’. An open-source dataset will be used throughout. An overview of methods that try to enforce fairness by design is provided. Again, all concepts will be illustrated by a selection of open-source packages.
4. The session will be concluded by a discussion of the framework that de Volksbank (a Dutch retail bank) has developed for its data science activities.
Python (Jupyter notebook)
The session focuses on concepts and not on technical implementation. Mathematics will be used in order to provide clear definitions. The notebooks will be extremely easy to use -- they just serve as an illustration. The discussion of algorithms/models that try to enforce fairness-by-design via in-processing requires that attendees understand the core (supervised) learning concepts.
Bio: Daan Knoope works as an AI Engineer at de Volksbank, the Dutch parent company of several banks and mortgage providers. He has a background in Computer Science (MSc) and has specialized in Algorithmic Data Analysis. During his studies, he researched the application of Dynamic Bayesian Networks on practical use cases to help further the development of explainable AI. Currently, he is focusing on developing AI-models for the bank as well as providing fellow AI Engineers the tools they need to efficiently explore data and build production-ready models