For those unaware of HTM: https://medium.com/@rockingrichie1994/understanding-hierarchal-temporal-memory-f6a1be38e07e
https://numenta.com/ "Temporal Memory is therefore the substrate upon which all neocortical functions are built. TM is probably the biggest difference between HTM theory and most other artificial neural network theories. HTM starts with the assumption that everything the neocortex does is based on memory and recall of sequences of patterns." "In order for a column in a Spatial pooler to exist it should be a winning column i.e the overlap score should be above some threshold value while non-winning columns are inhibited from learning. Only the winner columns can update their permanence values. Boosting helps to change the overlap score before the inhibition occurs giving less active columns a better chance to express themselves and diminishing columns that seem overactive . Boosting on better enables the learning of input data i.e it improves the efficiency. In other words we can say that the columns that have low overlap score are boosted so that they can better express themselves and all the columns with higher overlap score are inhibited because they are expressing themselves too much." I used to say in my notes like 3 years ago "winning match" lol. Not sure if I saw them say it first but their design sounds very similar to mine, at least. ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/Tcf8fbceae6bd0acd-M5e7c650bedd401dfd4cb4d54 Delivery options: https://agi.topicbox.com/groups/agi/subscription
