Module Name: IRIS 0.3
Release Date: February 11, 2006
Organization(s): SRI International

Brief Description:
Release 0.3 is the first version of IRIS released to the public under an LGPL license. At this point, IRIS is considered to be a fairly early stage research prototype (by no means a commercial product). It should be stable enough for use by a motivated user, and should provide enough of a framework where programmers can begin integrating with it.

IRIS is being developed as a small part of the CALO project. As such, we will try to be responsive to the open source community (particularly those who are using IRIS for machine learning purposes), but please understand that we have limited bandwidth.

What's New in 0.3:

IRIS 0.3 has focused primarily on infrastructural improvements (and not much on new features or UI):

• Plugin Framework - We moved from our own custom plugin framework to Spring, which is supported by a larger community.

• Knowledge Base - The KB for IRIS is now stored in JENA, improving the query language and overall flexibility of the architecture. Full-text search (lucene) has been seamlessly integrated with relational query. Integrated learning algorithms can now be granted their own KB for storing hypotheses, rather than being forced to write unvalidated hypotheses to a single global KB. Transactional POJOs are now supported, allowing the developer to control when data is written and where it is written to.

• Learning Framework - Classifiers are now declaratively configured in Spring XML configuration files and classifier declarations support SPARQL for member set specifications. This allows adding or changing suggestions classifications without requiring any programming. Clustering is now declaratively configured in Spring XML configuration files making it very easy to add new clustering algorithms. The UI for a clustering algorithm's runtime parameters can be fully configured in XML without requiring any programming. Clustering and results analysis is supported for any text document and a new textextraction package can be used to extract text and metadata from .doc, .pdf, and .ppt files.

• Ontology - The IRIS Ontology has been re-implemented. The result is more concise, consistent and requirements-focused. Compared with the CALO Y2 ontology, representations should be simpler for humans to understand, less computationally complex, yet equally expressive. We have also improved the km to owl translator. The ontology documentation is now produced using OWLDOC which is an off the shelf tool for producing Javadoc style documentation pages.

• Instrumentation & Automation - A new ontology-based event framework allows for instrumentation events and automation tasks to be linked directly to objects in the knowledge base. Events and tasks are defined in a centralized event ontology and can be dispatched both internally in IRIS and externally to other CALO components via the new Pub/Sub Messaging framework.

• IRIS UI - We have added a few new "themes" to IRIS, focusing on better management of screen real-estate. We also fixed several important bugs. Still much to do here...

• Performance - We now measure the performance of IRIS and generate weekly reports on startup and shutdown time, execution speed, memory, and file space usage. For this release, we have made a number of improvements in each of these areas, but we are still far from "commercial quality".

Known Issues:

The most important issues to know about are:

• PDFs can crash IRIS (#IRI-351): If you have not upgraded to Acrobat 7.0, opening a PDF file can freeze IRIS.
• Speed (#419, #464, #467): Installing IRIS is slow. IRIS is very slow on startup the first time, only a bit faster subsequent times, and during use, has functionality that needs to be dramatically optimized.