In my last talk (https://talks.cam.ac.uk/talk/index/228844), I introduced the criterion of deterministic neural reasoning, the method of reasoning through model construction and inspection, and proposed a novel neural network, Sphere Neural Network (SphNN), which reasons syllogistic statements by constructing and inspecting Euler diagrams. SphNN does not use training data, instead, it uses a transition map of neighbourhood relations. In this talk, I will present three control process (1. neighbourhood transition without constraint; 2. constraint neighbourhood transition; 3. neighbourhood transition with restart) and prove that the whole control process will successfully construct an Euler diagram in one epoch (M=1). With this proof, SphNN becomes the first neural network that reaches the symbolic-level of syllogistic reasoning.
https://www.youtube.com/watch?v=PFCHg-DAnEs
