BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Interpretable machine learning for critical evaluation of scientif
 ic ML models - the case of reaction prediction - David Peter Kovacs
DTSTART:20201026T170000Z
DTEND:20201026T173000Z
UID:TALK153304@talks.cam.ac.uk
CONTACT:Bingqing Cheng
DESCRIPTION:In this talk\, I will present an approach for interpreting ML 
 models and will illustrate it through the example of the Molecular Transfo
 rmer\, the state-of-the-art model for reaction prediction. I will outline 
 a framework to attribute predicted reaction outcomes both to specific part
 s of reactants\, and to reactions in the training set. Furthermore\, I wil
 l demonstrate how to retrieve evidence for predicted reaction outcomes\, a
 nd understand counterintuitive predictions by scrutinising the data. Addit
 ionally\, I will point out ”Clever Hans” predictions where the correct
  prediction is reached for the wrong reason due to dataset bias. Finally I
  will illustrate how the reported accuracy of models can be much higher th
 an it is in reality due to not appropriate train-test splitting.\nFor furt
 her details see: https://chemrxiv.org/articles/preprint/Quantitative_Inter
 pretation_Explains_Machine_Learning_Models_for_Chemical_Reaction_Predictio
 n_and_Uncovers_Bias/13061402
LOCATION:virtual ZOOM meeting ID: 263 591 6003\, Passcode: 000042\, https:
 //us02web.zoom.us/j/2635916003?pwd=ZlBEQnRENGwxNmJGMENGMWxjak5nUT09
END:VEVENT
END:VCALENDAR