ABSTRACT
When do the Knet and other internal attributes get updated (e.g., energy, patterns, frequency, Inet)? Responses occur at each attentive time point, which is related to the variable attention pulse rate. Before a response, tokenization (hierarchical tokenization) and patternization occur. However, updating Knet and other data tables will not occur until the response is finished, or possibly even later. The purpose of delaying the updating of data tables is to improve Zda's real time performance. Internal attributes such as energy are updated immediately after each response, or in real time.
After a reflex, update the Reflexon frequency only, unless the next attentive time is within a threshold Cr, likely requiring another immediate reflex. Note that a 2-gramton is a 2-gram with the second token actionable. When the frequency of a 2-gramton reaches a threshold, the 2-gramton becomes a Reflexon. A 2-gram, 2-gramton, and reflexon are associations, but the reverse is not necessarily true.
After a fast-thinking response, the Knet will be updated using the corresponding pattern:
Update the frequency of the exactly matched pattern in Knet, the distributive reward, and recency.
Update the frequency of the new similarity-matched pattern (desensitisor) with frequency one in Knet, the distributive reward, and recency.
Create the imitation-induced new pattern with frequency one, the distributive reward, and recency.
Create the creativity-induced new pattern with frequency one.
After a slow-thinking response, the hierarchical tokenization of an event-string of up to 16 randomly selected elementary tokens occurs before patternization, and at the higher token level, patternization occurs in three forms:
Update the frequency of the higher level tokens from hierarchical tokenization in Knet, the distributive reward, and recency.
Update the frequency of the new similarity-matched pattern (desensitisor) with frequency one in Knet, the distributive reward, and recency.
Create the imitation-induced new pattern with frequency one, the distributive reward, and recency.
Create the creativity-induced new pattern with frequency one.
In deep-thinking, patternization occurs before the response, including the patternization that could occur in slow-thinking, and repatternization of Knet (induction with desensitisors, deduction with sensitisors). Any pattern (including high-level tokens from hierarchical tokenization) update will update frequency, the distributive reward, and recency. The response can be based on n-token ahead predictions. From deep-thinking with recursion, Zda can derive many learning methods, statistical models, and can create humanized agents.
