BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:From Machine Learning to Machine Reasoning: Deterministic Neural S
 yllogistic Reasoning (Part 1).  - Tiansi Dong
DTSTART:20250317T170000Z
DTEND:20250317T174500Z
UID:TALK228844@talks.cam.ac.uk
CONTACT:Pietro Lio
DESCRIPTION:This talk was recorded https://www.youtube.com/watch?v=9hnM9C4
 xHeM\n\nIn my last talk (https://talks.cam.ac.uk/talk/index/228790)\, I sh
 ow four methodological limitations that prevent machine learning systems f
 rom reaching the rigour of syllogistic reasoning. They cannot achieve the 
 rigour\, not because of insufficient amount of training data\, instead\, t
 o achieve the rigour\, they shall not use training data. What kind of neur
 al networks can be? Neural networks use vector embedding\, which is a sphe
 re embedding with zero radius. In this talk\, I will show the four limitat
 ions can be completely avoided by promoting vector embedding into sphere e
 mbedding with non-zero radius and the criterion of  achieving deterministi
 c neural reasoning\, namely\, for any satisfiable reasoning\, there is a c
 onstant number of M that the neural network shall correctly construct a mo
 del within M epochs. I will introduce a novel neural network\, Sphere Neur
 al Network (SphNN)\, which explicitly represents geometric objects\, here 
 spheres\, and introduces the method of syllogistic reasoning by constructi
 ng Euler diagrams in the vector space. Instead of using training data\, Sp
 hNN uses a neighbourhood transition map to transform the current sphere co
 nfiguration into the target. SphNN is the first neural network that achiev
 es deterministic human-like syllogistic reasoning in one epoch (M=1).
LOCATION:Lecture Theatre 2\, Computer Laboratory\, William Gates Building
END:VEVENT
END:VCALENDAR