|Period:||period 4 (week 17 through 26, i.e., 26-4-2021 through 2-7-2021; retake week 28)
|Participants:||up till now 0 subscriptions|
|Schedule:||Official schedule representation can be found in MyTimetable|
|Contents:||This course will familiarize students with a growing set of concepts and techniques to develop and assess machine learning systems, so that these systems are fair and interpretable, and can be used in responsible ways.
More specifically, we will discuss methods to measure the fairness of machine learning systems (ranging from fairness metrics to the creation of challenging evaluation datasets) and machine learning techniques to reduce biases in ML systems (e.g., data augmentation, representation learning).
The course also covers different approaches to creating and evaluating interpretable or explainable ML approaches (e.g. post-hoc local explanations, influence functions, counterfactual explanations).
The impact of machine learning (ML) systems on our society has been increasing rapidly, ranging from systems that influence the content that we see online (e.g., ranking algorithms, advertising algorithms) to systems that enhance or even replace human decision making (e.g. in hiring processes). However, machine learning systems often perpetuate or even amplify societal biases—biases we are often not even aware of.
What’s more, most machine learning systems are not transparent, which hampers their practical uptake and makes it challenging to know when to trust (or not trust) the output of these systems.
The course will cover examples from various areas of AI. Given the expertise of the lecturers we will also zoom in on specific examples from natural language processing and multimodal affective computing research. Our discussion will also be informed by relevant literature from the social sciences. An interest in these areas is therefore desirable.
|Literature:||Most of the material we will read are academic articles. A few candidate readings are:
-“Why Should I Trust You?” Explaining the Predictions of Any Classifier, Ribeiro et al., KDD 2016
-A Unified Approach to Interpreting Model Predictions, Lundberg and Lee, NeurIPS 2017
-Fairness and Machine learning, Limitations and Opportunities, Solon Barocas, Moritz Hardt, Arvind Narayanan, https://fairmlbook.org/
-Gender Shades: Intersectional Accuracy Disparities in Commercial Gender Classification, Buolamwini and Gebru, Proceedings of Machine Learning Research 2018
-Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddings, Bolukbasi et al., NeurIPS 2016
-The Mythos of Model Interpretability, Lipton, ACM Queue, 2018
-Beyond Saliency: Understanding Convolutional Neural Networks from Saliency Prediction on Layer-wise Relevance Propagation, Li et al., Image and Vision Computing 2019.
|Course form:||There will be lectures and practical exercises. Students are also expected to discuss and present academic articles. The course also contains a group project.|
|Exam form:||Group project, midterm, paper presentations, paper review|
|Minimum effort to qualify for 2nd chance exam:||Om aan de aanvullende toets te mogen meedoen moet de oorspronkelijke uitslag minstens 4 zijn.|