Here is a paper titled "A Grounded Deep Symbolic Neural Network for Perception" that I am planning to submit to the NeSy 2022 workshop. It is one of a number of workshops in the Second International Conference on Learning and Reasoning (IJCLR 2022) in Windsor, UK on 28th – 30th September. Papers are due by May 31st. I would appreciate any constructive feedback you would like to make.
https://www.adaptroninc.com/sites/default/files/inline-files/Grounded_Deep_Symbolic_Neural_Network_for_comments.pdf Abstract Both animals and artificial intelligent agents rely upon the identification of types of objects and events during perception. It is a categorization process, which senses, recognizes and encodes invariant features. Binary neurons (binons) are general-purpose artificial neural nodes for representing properties, objects, events and relationships between them. Non-symbolic binons are used in short-term memory to represent core sensory properties such as position, intensity and time and ones derived from them. Ratios derived from these properties are converted into invariant symbolic categories using a novel discretization algorithm based on the Weber-Fechner psychophysical laws. Symbolic binons are combined to form deep hierarchical neural networks that comprise long-term memory. It contains spatial and temporal binons representing the shape and contrast patterns for categories of objects and events. They are grounded on the core and derived properties. Empirical evidence of their successful use in classifying handwritten digits was provided by Martensen in 2013[1]. The neural networks are 100% symbolic, transparent, compositional, scalable and sparse. Learning is continuous and unsupervised. ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T842d98b5a3b0de4d-M15de4154a8d9d55f0d73a200 Delivery options: https://agi.topicbox.com/groups/agi/subscription
