How can data scientists shed light into the black box of neural networks? How can researchers explain the decisions an algorithm took?
Anna Arias Duart is a Ph.D. candidate in artificial intelligence at the Universitat Politècnica de Barcelona and investigates how algorithms takes devcisions. She works with image recognition and visually highlights the parts of an image which were used for classification.
Links on Episode 47
- Anna’s research on a dataset with images of high resolution to understand the parts of the image considered important by the models
- Anna’s paper on creating a metric to to compare different explainability methods
- Anna’s current research on using feature attribution methods – the focus – to find biases in the datasets and therefore in the models (section 6)