What if software can learn and set its own goals?

What if software can be trained naturally?

What if software can recognize, predict, decide, and act?

What if software can have DNA and evolve?

What if that software, is your software?

GAIuS 3.0

Posted on by editor

Happy 6th Birthday, GAIuS!

Let’s take a brief look at your history.

It was January 10, 2008, six years ago today,  that the first version of GAIuS was saved into the code repository.  As an engineering project to discover the most general algorithms that can be used in any problem domain, there were five major restarts.  All of this happened within the first year.

That’s five big failures.  Five ways that, while searching for those elusive general algorithms, I had reached a point in the code that I realized it was still too specific to a set of problem domains.  It still wasn’t general enough.  The first three times, all of the code that had been written up to that point had to be dumped.  Back to the drawing board!  The last two times, some of it was able to be kept.  Ah, progress.

After those failures, the subsequent failures weren’t as catastrophic.  They didn’t require a complete re-thinking of the process.  Sure, many of the individual problems left to be solved were hard, but the path had become clearer.  The process is what is important.  This is the process – now known as the “GAIuS framework” – that extracts information from data.  Now that these steps were discovered, the rest was just a matter of programming.

On January 10, 2009, the Beta version of GAIuS was released.  This first, complete, implementation of the “GAIuS framework” was slow, clunky, and inefficient.  But, as a proof-of-concept, it worked.  People downloaded it, read the documentation, and gave positive feedback.  Shortly after, we were getting projects to apply it to different problem domains.

This is how the GAIuS engine, i.e. the software implementation of the theoretical framework, has been molded into what it is today.  GAIuS, as a process, hasn’t changed since that initial Beta version.  That process works.  Over the years, it has proven itself over a wide problem domain range from providing retail product recommendations, business predictive analytics, anomaly detection, multi-sensory robotics command & control systems, medical diagnosis, and machine vision.

What has changed are the improvements in the performance and efficiencies of the internal algorithms; and the API that allows developers to interact with agents created by the engine.

On December 21, 2012, GAIuS 1.0 “Nibiru” was released.  This was the first production-ready version.  Having proven itself on a few disparate verticals, undergone extensive quality testing, and qualified by notable scientists and engineers, “Nibiru” entered the workforce.

After iterating on the real-world, production environment experience, and customer feedback from Nibiru, GAIuS 2.0, codename “Apollo”, went into production last year.  Apollo’s major change was to the Knowledgebase storage technology, and improvements to the customer/developer interface.  The improved storage technology opened up Big Data capabilities to the engine like never before.   To the relief of developers, the simpler JSON-RPC was introduced alongside the existing XML-RPC interface, reducing the learning curve of training and querying the GAIuS engine.

On this 6th anniversary of GAIuS version 3.0, codename “Benji”, has been released into production systems.  Benji’s focus has been on major performance improvements and feature enhancements.

As always, all current Intelligent Artifacts customers automatically and seamlessly get the latest version applied to their current agents.

What’s new in the “Benji” release?

In 3.0, GAIuS manipulatives can now be attached in many-to-many connections.  This makes an agent’s topology much more flexible and efficient.  Both input and output manipulatives can be attached in this topology.  Furthermore, input manipulatives can now feed their results directly to output manipulatives, bypassing the primitive node completely.  This allows the primitive to process and make decisions on contextual information, while the actions can use input data as arguments for their outputs.

GAIuS’ API has also seen a facelift.  As of 2.0, the JSON-RPC was added alongside the original XML-RPC.  For 3.0, XML-RPC has been dropped.  Some new JSON-RPC API calls have been added while deprecated ones have been removed.  (Customers can review their documentation for details.)

Multiprocessing support has been added for some of the most resource intensive major algorithms.  Specifically, the pattern matching, model classification, and prediction algorithms are seeing at least an order of magnitude performance improvements on multi-processor systems.  This is in addition to improved efficiencies and performance to these algorithms.  These algorithms affect all GAIuS agents, so all customers will now see significant improvements in response times.

As part of their contracts, all current Intelligent Artifact customers have already had their agents tested and upgraded to GAIuS 3.0.  All agents have successfully passed testing and are seeing performance improvements of at least 10x.

What’s next?

With these advances to the GIAuS engine, our work in 2014 will focus on applications that require real-time processing of shorter period events.  In particular, we are excited to take the next steps in our work on machine vision, and financial analysis.

If the last 6 years are any indication, 2014 will prove to be another exciting year for GAIuS.

This entry was posted in News. Bookmark the permalink.

Comments are closed.

You must be logged in to post a comment.


GAIuS 3.0

Happy 6th Birthday, GAIuS!

Let's take a brief lo…


Nibiru Unleashed!

Did the Mayans predict the end of the world?  Perhaps, that depends on your definition. GAIuS 1.0 codenamed Nibiru has been re…