Artificial Intelligence Series (Vol. 2):

This is the second release of a series on AI, with the platform serving as a support for a course this semester.

New controllers

The course is based on AI a Modern Approach by by Stuart Russell and Peter Norvig, and as part of a new AI Rest service, the Java code for the text book was ported using IKVM and new controllers were added for propositional logic, first order logic and probabilistic models

As with dynamic Constraint satisfaction problems included in vol.1, new controllers services include:

• Demos with default entities that can be edited online (propositional and first order logic knowledge bases, bayesian networks, hidden markov models).
• Multiple inference strategies, which can be passed as web service parameters to the demos, and support the coursework.
• Open actions, where the demo entities are replaced by dedicated parameters passed over Http by the web services client. those should ease integrating AI components in your own applications.

The collection of postman requests was updated to include demos of the various strategies available.

Future releases will include additional controllers for semantic web, Natural language processing, Neural networks and genetic algorithms.

Neural Networks

Neural Networks and other machine learning techniques are a work in progress. The final version will integrate the following frameworks:

• AIFH and Encog by Jeff heaton
• Infer.Net, the powerful probabilistic programming library by Microsoft
• Cntk (computational network toolkit): Microsoft's deep learning Framework.

Game of Go

The game of Go is quite a hot topic in AI.

As part of our teaching session was a project to train a Go AI with deep neural networks. We were very lucky to have one of Microsoft CNTK's main contributors, Frank Seide, spend a significant amount of time helping us get our configuration right and work around the latest version's issues.

The final environment will be part of next version together with the coming stable releases of CNTK and other machine learning material, yet as part of this release are included:

• a sample web service to generate a training dataset from a set of professional go records leveraging GnuGo and Gotraxx, together with a sample dataset of 1500 game records.
• A corresponding convoluted neural network training configuration, together with a sample trained model, and methods to run the model to play or to generate influence maps.

Asset manager editor

In order to simplify editing Skin source files online, a control adapter was introduced to add a file editor to the asset manager. It works by hijacking the existing module creator's source editor with color syntax.

String filter includes a Markdown parser

Markdown, the language used on Wikipedia and this site, is increasingly supported by online platforms as a simple yet convenient alternative to rich text editors, allowing advanced users to format their published content while keeping control on the resulting Html.
As part of new string filtering capabilities leveraged by several components, a Markdown parser is now available.
It is illustrated in the default firewall html filtering rule.

Highlights

• New: Artificial Intelligence Rest services (vol.2): Propositional reasoning and First Order logic with dynamic knowledge bases and inference procedures: Resolution, DPLL, Walksat, Model elimination, Otter etc.
• New: Artificial Intelligence: Probability with Bayesian networks and hidden markov models, exact and approximate inference: rejection sampling, Likelihood weighting, Gibbs sampling, Forward backward, Fixed lag smoothing, particle filtering etc.
• New: Artificial Intelligence: Service dedicated to the Game of Go, leveraging GnuGo, Gotraxx and CNTK to generate training sets and to run deep convoluted neural networks
• New: Call Service Action: invoke dynamic hosted web API services server side, without the overhead of an Http Request.
• New: Control Adapter adds a text editor to Asset manager by hijacking the module creator source editor, handy for skin editing
• Improved: Html filter local transformer parses Markdown syntax: enable Wikipedia formatting where you need it
• Improved: Expression Filter supports tokens and markdown parsing
• Improved: Run Program action fixed, localized and supports custom working directory
• Improved: Added Recursive A star and Greedy Best first search strategies
• Improved: CSV serializer updated to latest FileHelpers version + supports escaped delimiter (e.g tabulations)
• Fixed: Restored Search Result Scraper bot parameters and corresponding user bots parameter overrides

Min Requirements:

• .Net 3.5+ (remove .Net 4.0 assemblies), .Net 4.0 for Web API services
• DotNetNuke 4.8.1+, Web services available from DNN 7.0
• Aricie Shared 1.9.3+

 

---

Jesse - CTO Aricie