Last Sunday I gave the following presentation to the Northwest AGI Forum. Grounding Invariant Features with Binons for Perceptual Constancy
Both animals and artificial intelligent agents rely upon the representation of invariant features of objects and events for perceptual constancy during recognition. Perception is a categorization process that senses, encodes and recognizes such features. It produces percepts that are grounded on the senses. Binons (binary neurons) are general-purpose artificial neural nodes for representing relationships between things. Perception binons are able to represent properties such as position, intensity, spatial and temporal repeat counts and time. Features such as delay, distance, width, duration, speed, size, shape, edges, contrast, and more are derived from them. Shape and contrast patterns allow for types of things to be recognized independent of their varying features. This presentation covers the subjects of: * Perceptual constancy * Property features versus part features * Non-symbolic features (interval and ratio scale) * Symbolic features (nominal and ordinal) * Converting non-symbolic properties into symbolic ones * The symbol grounding problem * General purpose design of senses and sensors * Perception = sensing, encoding, and recognition * Deriving invariant features from core sensory properties * Binary Neurons (Binons), spatial and temporal * Deep overlapping compositional hierarchies in perception * Weber-Fechner’s Law and the Just Noticeable Difference (JND) * Contrast and shape patterns * Representing objects and events It's available at: http://www.adaptroninc.com/BasicPage/invariant-features ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T8366cc740ec68376-M1295159505af886484d69a4e Delivery options: https://agi.topicbox.com/groups/agi/subscription
