Stefan Bauer: Towards Learning Causal Representations & Interactive Benchmarks
Time: Tue 2022-09-06 11.00
Location: 3721, Lindstedtsvägen 25, and Zoom
Video link: Meeting ID: 621 8808 6001
Participating: Stefan Bauer (KTH)
Many questions in everyday life as well as in research are causal in nature: How would the climate change if we lower train prices or will my headache go away if I take an aspirin? Inherently, such questions need to specify the causal variables relevant to the question. A central problem for AI and many application areas is thus the discovery of high-level causal variables from low-level observations like pixel values. While deep neural networks have achieved outstanding success in learning powerful representations for prediction, they fail to explain the effect of interventions. This is reflected in a limited ability to transfer and generalize even between related tasks. As a way forward to learn causal representations from data, this talk will describe our recent advances of combining interventions and causal structure with deep learning based approaches, as well as our efforts to create real-world benchmarks for the interactive learning paradigm. The proposed algorithms and frameworks are widely applicable, with use cases ranging from fairness in algorithmic decision making to experimental design in drug discovery.