BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Sphere Neural-Network for Higher-level Cognition - Tiansi Dong
DTSTART:20230718T160000Z
DTEND:20230718T170000Z
UID:TALK203371@talks.cam.ac.uk
CONTACT:Pietro Lio
DESCRIPTION:To simulate higher-level cognition of our mind\, deep-learning
  needs to go beyond the statistic learning framework\, and reason with out
 -of distribution data. Here\, we present a qualitative extension by promot
 ing vectors into spheres as the input-output signals\, and come up to the 
 Sphere Neural-Network (SphNN). We show that the sphere has the representat
 ion power to introduce the methodology of reasoning by model construction 
 into the neural computing. This enables SphNN to determine\, without train
 ing data (therefore\, robust to out-of-distribution data)\, the validity o
 f long-chained syllogistic reasoning\, the microcosm of higher-level cogni
 tion\, by constructing Euler diagrams in the vector space\, with the worst
  computational complexity of O(N^2) (where N is the length of the chain). 
  A high-level process for model construction plans the neighbourhood trans
 ition towards the target model\, by using a deterministic neurosymbolic ca
 usal schema and habitual neural operations that gradually transform the cu
 rrent relation towards the neighbourhood relation. SphNN is the first neur
 al model that carries out explainable and deterministic syllogistic reason
 ing (Experiment 1)\, with theoretical proof\; Compared with ChatGPT in lon
 g-chained syllogistic reasoning\, SphNN achieves 100% for all 1200 tasks\,
  while ChatGPT achieved an average accuracy of 62.8% (Experiment 2). In ou
 r experiments\, ChatGPT may also give a correct answer with a false explan
 ation\, suggesting that ChatGPT may not truly construct the human-like mod
 els used in its answers. SphNN is able to evaluate the answer of ChatGPT b
 y constructing models\, and give feedback through prompts. This improves t
 he accuracy of ChatGPT from 80.86% to 93.75% in deciding the satisfiabilit
 y of atomic syllogistic reasoning (shown in Experiment 3). By fixing centr
 e orientations of spheres to pre-trained embeddings from LLMs\, e.g. ChatG
 PT\, Experiment 4 shows that pre-trained vector embeddings\, such as GLOVE
 \, BERT\, ChatGPT\, can very well approximate orientations of the sphere c
 entres. Experiment 5 shows rigorous logic deduction cannot be achieved thr
 ough supervised learning. SphNN demonstrates a new way of realising logic 
 deduction through explicit model construction in the vector space\, and si
 gnificantly narrows the gap between higher-level cognition and deep-learni
 ng\, and will enhance collaborations among neural computing\, classic AI\,
  and cognitive science to develop novel neural models that reach the expla
 inability and determinacy of higher-level cognition.\n\n----------\nTiansi
  Dong leads the new neurosymbolic research at Fraunhofer Institute IAIS \n
 \nhttps://www.iais.fraunhofer.de/en/research/artificial-intelligence/neuro
 symbolic-representation-learning.html\n\n\nhttps://cl-cam-ac-uk.zoom.us/j/
 93040206714?pwd=aFVNeExndzFWSk50b0dSWldSaXFndz09\n
LOCATION:Lecture Theatre 2\, Computer Laboratory\, William Gates Building
END:VEVENT
END:VCALENDAR