Updated memory demo

I’ve updated the human-like memory demo to illustrate how clusters can be generated from taxonomic and thematic knowledge. See:

 https://www.w3.org/Data/demos/chunks/memory/ <https://www.w3.org/Data/demos/chunks/memory/> 

A button is provided to trigger the start goal, as a means to initiate clustering. The clusters are then shown in the log field below.

In the process of doing this work, I discovered a few things:

First, that using @do put in a rule to write a chunk works much better when you specify the id of the chunk you want to update.

Second, put needs to copy the properties in the module chunk buffer as well as any properties set explicitly in the action - the latter take precedence.

Third, sometimes you need to take care to clear the module chunk buffer. This is the case when you have a rule that applies when @do get failed to retrieve a matching chunk.

Fourth, the algorithm for stochastic recall needs to reduce the amount of noise for very recently formed memories, as you generally have good recall in such cases. I’ve chosen to progressively reduce the standard deviation of the normal distribution for noise for intervals smaller than the half-life for the forgetting curve.

Fifth: remembering is an active process, and works better if you are careful with the encoding and consolidation of new memories. This involves clustering and techniques like the memory castle.

I am now looking forward to resuming work on semantic processing for natural language.

Dave Raggett <dsr@w3.org> http://www.w3.org/People/Raggett
W3C Data Activity Lead & W3C champion for the Web of things