BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:On the Two-fold Role of Logic Constraints in Deep Learning - Dr Ga
 briele Ciravegna\, Inria – Université Côte d'Azur
DTSTART:20220429T150000Z
DTEND:20220429T160000Z
UID:TALK173735@talks.cam.ac.uk
CONTACT:Pietro Lio
DESCRIPTION:In the last few years\, Deep Learning (DL) has achieved impres
 sive results in a variety of problems ranging from computer vision to natu
 ral language processing. Nonetheless\, the excitement around the field may
  remain disappointed since there are still many open issues.\nTo mitigate 
 some of these problems\, we consider the Learning from Constraints framewo
 rk. In this setting learning is conceived as the problem of finding task f
 unctions while respecting the constraints representing the available knowl
 edge. We provide an application in the Active Learning scenario where Firs
 t-Order Logic knowledge is converted into constraints and their violation 
 is checked as a guide for sample selection.\nAlso\, we propose to employ d
 omain knowledge to defend from Adversarial Attacks since it provides a nat
 ural guide to detect adversarial examples. While some relationships are kn
 own properties of the considered environments\, DNNs can also autonomously
  develop new relation patterns. Therefore\, we also propose a novel Learni
 ng of Constraints formulation which aims at understanding which logic cons
 traints are satisfied by the task functions. This allows explaining DNNs\,
  otherwise commonly considered black-box classifiers.\nIn a first case\, w
 e propose a pair of neural networks\, where one learns the relationships a
 mong the outputs of the other one and provides First-Order Logic (FOL)-bas
 ed descriptions. In a second case\, we propose an end-to-end differentiabl
 e approach\, extracting logic explanations from the same classifier. The m
 ethod relies on an entropy-based layer which automatically identifies the 
 most relevant concepts. It enables the distillation of concise logic expla
 nations in several safety-critical domains\, outperforming state-of-the-ar
 t white-box models.\n\nhttps://cl-cam-ac-uk.zoom.us/j/99805544705?pwd=cXR6
 MTlaeXd6VmEreVdQSmFRblBtUT09\n
LOCATION:Zoom + presence (lecture theatre 2) 
END:VEVENT
END:VCALENDAR