Olivia Caramello_SyntacticLearningViaToposTheory_watermark.pdf

1、Syntacticlearning viaTopos TheoryOlivia CaramelloGeneralprinciples for afuture AIMathematicalLogicTopos TheoryToposes forsemanticinformationSyntacticlearningFor furtherreadingSyntactic learning via Topos TheoryOlivia Caramello(University of Insubria-Como,and Grothendieck Institute)New AI Theory work

2、shopRL China,24 November 2023Syntacticlearning viaTopos TheoryOlivia CaramelloGeneralprinciples for afuture AIMathematicalLogicTopos TheoryToposes forsemanticinformationSyntacticlearningFor furtherreadingThe key features of true intelligenceIn artifi cial intelligence and beyond,unifi cation and com

3、putationare two facets of the same phenomenon:the more there isunderstanding,the less brute force is needed and calculationscan be simplifi ed.One should thus possibly design artifi cial learning systems insuch a way that computations become as conceptually inspiredand enlightening as possible,and,m

4、ost importantly,meaningful,to the human mind.In particular,we should shift to a newconception of information which is based on semantics andinvariants rather than on the classical set-theoretic foundations formathematics.One should also aim for strong forms of modularity and continuityensuring,in pa

5、rticular,technical flexibility,adaptability andresilience capabilities of the resulting systems.Despite the spectacular advances in deep learning systems of thelast years,we are still far from a form of artifi cial intelligenceenjoying these features,which in fact are quite characteristic ofhuman in

6、telligence.2/14Syntacticlearning viaTopos TheoryOlivia CaramelloGeneralprinciples for afuture AIMathematicalLogicTopos TheoryToposes forsemanticinformationSyntacticlearningFor furtherreadingUnifi cation and computationA general principle is that one should aim for the smallestpossible number of gene