The Triadic Continuum

I spend a bit of time blogging about the importance of Persistent Context – this being a data store that contains up-to-the-second contextualized information. Each new observation is Semantically Reconciled, and its relationship is evaluated against all previous observations. What is learned incrementally builds on the historical context and if a contextualized observation warrants action, this insight is published. Persistent context is absolutely essential to delivering real-time situational awareness – something I have been calling Perpetual Analytics. This is not window-based Stream Processing.

Introducing the Triadic Continuum.

Ever heard of the Triadic Continuum? Probably not. Me either, until a few days ago that is … when this article in DMReview: "The Best New BI Invention You’ve Never Heard Of" made its way into my inbox.

This article describes what appears to be a pretty similar concept to persistent context … and thus I got somewhat excited. While reading this article, I tried to imagine what Jane Mazzagatti and her team are doing. It is great news that work of this type has received some attention.

Here are some excerpts of interest to me and related thoughts:

[… instead of data and information being "found," "analyzed" or "discovered," it is already there waiting to be "realized."] [… discover more complex, less easy-to-find relationships in vast amounts of real-time data … answers to queries may change as more and more data are introduced into the structure - similar to how we are able to change our perceptions and decisions based on … new facts.]

I wonder if they have started thinking beyond assembling data for future recall … but also performing data finding data and relevance finding the user processing at ingestion. The principle being: at the exact moment new observations (data) are ingested into a persistent context data store … also happens to be the cheapest moment computationally to detect relevance (insight). [More here: Sensing Importance: Now or Never]

[… theoretically, each individual particle of data occurs only once within the structure …]

I wonder if they have been playing with the concept of also making disambiguation assertions during the ingestion process. [More here: Entity Resolution Systems vs. Match Merge/Merge Purge/List De-duplication Systems]

[… a computer data structure that is self organizing - in other words, a data structure that naturally organizes new data by either building on the existing data sequences or adding to the structure as new data are introduced.]

This makes me wonder if they have been working on the ripple effect that can happen when a new observation invalidates an earlier disambiguation or relationship assertions.  [More here: Sequence Neutrality in information Systems]

[… the format and organization of the Triadic Continuum not only hold the representation of the data, but also the associations and relations between the data and the methods to obtain information and knowledge.]

First, it appears they are thinking of context in the same light as I do. [More here: Context: A Must-Have and Thoughts on Getting Some …]

Maybe I am reading too much into this, but I also wonder if the words "relations between the data and the methods to obtain information" have anything to do with commingling new observations and user queries into the same data space. [More here: What Came First, the Data or the Query?]

[… has developed and patented a new data structure …]

Having thought a lot about how data structures govern function, I do sometimes think about persisting the data in something other than an SQL database engine to get even higher throughput rates. Although, then I ponder how to make up for the freebies that come with industry standard SQL engines like transaction consistency, restartability, and a large community of trained DBA’s. So I wonder what thoughts this team has in this area. [More here: Big Breakthrough in Performance: Tuning Tips for Incremental Learning Systems]

No matter where these folks are in their evolution – first or tenth generation – I am real excited to hear about their work.

Hopefully, one day, I will have the chance to chat with these folks. Kindred spirits I suspect.

RELATED POSTS:

Accumulating Context: Now or Never

The Phone Call is Coming From Your House! Context is King

Enterprise Intelligence – My Presentation at the third Annual Web 2.0 Summit

It’s All About the Librarian! New Paradigms in Enterprise Discovery and Awareness

How to Use a Glue Gun to Catch a Liar

Intelligent Organizations – Assembling Context and The Proof is in the Chimp!

It Turns Out Both Bad Data and a Teaspoon of Dirt May Be Good For You

There Is No Such Thing As A Single Version of Truth

More Data is Better, Proceed With Caution

Comments

As usual, Jeff, you find some really interesting work to post about. I'm definitely interested in learning more about the Triadic Continuum.

[Dan Linstedt]

Hi Jeff,

I heard about this too, and at first glance it sounds promising. However when I looked at the patent application - what's really interesting is the actual functionality that they've built.

What I find is that the Data Vault architecture that I've constructed over the years meets the same mathematical modeling basis as this "traidic continuom." There are three parts in the Data Vault, Hub and Satellite establish context and business keys for a local component, and the Link is left to be "discovered, assigned, weighed" and acts as a vector between the contextual time based elements.

I'm wondering if you wouldn't share your thoughts about the Data Vault modeling architecture as well?

I'm real interested in talking to these folks to see just what their data modeling looks like... They have some great application on top of it, but I bet that we could use the Data Vault model with their application, and arrive at similar results.

As you know, the Data Vault model is also based on an interpretation of the neural cells in the brain.

Cheers,
Dan L

[Jane Mazzagatti]

Actually we have thought about many of these issues and more - the Pierce cognitive model covers everything from the rudimentary data structure we're currently experiementing with to self awareness - but so far we have only been able to implement a rudimentary K (knowledge) structure (lack of resources not understanding) and prototype a data analyzer to begin to demo the potential of data structure - John's book gives the details of how the rudimentary structure is created and some of the attributes - to answer some of your questions the K structure (patented as the interlocking trees data store) creates a structure from data that contains everything you would have if you created a set of tables with indexes and cubes - so you have every context captured in one structure as soon as you record the data - there are no calculated probabilities stored in the structure (only counts) so that as each new record is added every probability changes but no calculations are done until the particular probability is needed - and the only time structure is created is when some new experience is encountered - such as a new field variable (otherwise the old structure would be reused) so that the 'structure knows' immediately that something new has been encountered - also, the resulting structure for large data sets is smaller than the original data - Jane

[Michael Atlass]

When you suggested that "The principle being: at the exact moment new observations (data) are ingested into a persistent context data store … also happens to be the cheapest moment computationally to detect relevance (insight)."

I think you could profitably rephrase that in the context of the Triadic Continuum. The moment the data become part of the TC, it is connected to all past events of which it is related in a structured relationship that can lead back to that data at any time in the future when it is determined that that data is again needed to answer an inquiry. Further, as future events are linked to that data via their relationship, those events also become part of the answer to the query, according to their relevance.

[Ray Garcia]

Wouldn't the concept of Perpetual Analytics require that a valid simulation model exist for the system dynamic in order for it to receive new information and accurately produce a new system state that is useful for analysis and subsequent validation?

How would this be possible without a unifying theory on the rational application of evolutionary/genetic/neural programming, properly balanced with human judgment?