This week Erik Arakelyan (Researcher @ ARM, PhD Student @ CopeNLU) will present a paper from EMNLP'2020 — A Diagnostic Study of Explainability Techniques for Text Classification. In the work, the authors develop a set of diagnostic properties for explainability technique evaluation, while giving empirical and intuitive justifications for the chosen properties. A broad spectrum of explainability methods is discussed, with further discussion of their shortcomings in terms of explainability value and completeness.
I will strongly encourage everyone to join the meeting, especially those seeking to attain better intuition on how deep learning models work.
See you at 3pm on Saturday, Zoom meeting link: https://zoom.us/j/315197427Best,