BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Neural Symbolic Interpretability - Pietro Barbiero (IBM Zurich)
DTSTART:20260212T160000Z
DTEND:20260212T170000Z
UID:TALK244324@talks.cam.ac.uk
CONTACT:Mateja Jamnik
DESCRIPTION:Neuro-symbolic (NeSy) interpretability provides a formal langu
 age for controlling deep neural networks (DNNs) and ensuring that their be
 havior satisfies human desiderata. We will present the architectural condi
 tions that enable NeSy control in DNNs and\, based on these conditions\, i
 ntroduce a general blueprint for instantiating NeSy-interpretable reasoner
 s. We illustrate this paradigm using two representative examples: verifiab
 le and causally transparent concept-based models.\n\nPietro is a Swiss Pos
 tdoctoral Fellow and Ellis member at IBM Research. Previously\, he was a p
 ostdoc at Universita' della Svizzera Italiana and received his PhD at the 
 University of Cambridge. His research focuses on the mathematical foundati
 ons of interpretability and on developing causally transparent models to g
 o beyond the current accuracy-interpretability trade-off.
LOCATION:Lecture Theatre 2\, Computer Laboratory\, William Gates Building
END:VEVENT
END:VCALENDAR