Harry Chen Thinks Aloud

Soon we will say hello to 2006, and bye-bye to 2005. It has been an exciting year for me and my family.

2005 event highlight:

• Joined Image Matters LLC — my first full-time job after finishing my PhD.
• Bought our first house in Columbia, MD.
• Got serious about managing my personal finance — 401k, IRA etc.
• Switched main research focus from pervasive computing to geospatial semantic web.
• Switched my blogs from Blogger.com to WordPress

Interesting blogs in 2005:

• New Job, New Life, and New Goals
• My asset allocation in 2006 will include China
• SemanticWorks 2006 still needs some work and follow-up
• The whole brain atlas
• Japanese sorts their trash. Can we sort ours?
• My problem with President Bush
• PhD Students Backup Your Thesis

2006 outlook:

• Apple will announce iPod pico
• Google Earth and Yahoo! Maps will feature semantic markups that allow users to query geospatial information
• More people will know RDF is not XML
• Microsoft will attempt to fight off an Ajax-version of the OpenOffice software by giving away MS Office for free
• Problems in Iraq will continue to eat up US taxpayer money.

To be continued…

Few days ago I wrote about misconceptions that surround OWL reasoning. I received a good comment from Max V?lkel –

Stating a rule like the one in your example has nothing to do with any
kind of OWL. OWL Lite, OWL DL and OWL Full are three sets of defined
semantics (you can think of it as three sets of rules). With regard to
performance: OWL Full is simply undecidable in some cases, which means,
ther is never an answer to some queries. Btw. not all reasoners are
based on FOL theorem provers. KAON2 is different.

First, Max seems to suggest that OWL reasoning has nothing to do with different sub-languages of OWL. Second, Max kindly point out that OWL-Full reasoners need not to be built on FOL theorem provers. I agree with him on both points. However, there is something that I want to clarify.

I agree that given an OWL ontology, the kind of sub-language that it belongs to is a matter of semantic representation, which is not dictated by any kind of inference rules (such as the ones that I have defined). However, the OWL language type of an ontology does matter in building real world applications.

When building an ontology for a particular application, it is important to know how this ontology will be used. If the ontology is used to support logical inference, it is important to know the kind of logical inference that the application will need to support — is it within the capability of a target reasoner?

The use of a DL reasoner is typical in many applications today because developers can easily find an off-the-shelf DL reasoner. Reusing an existing DL reasoner often helped to reduce the development efforts in building logical inference support. In my experience, using OWL-DL alone is insufficient to support the needs of many real-world applications. As an engineer, I used to be afraid of telling people that my system exploits OWL reasoning that is beyond OWL-DL because I was afraid that people will criticize my system’s non-deterministic behavior or performance.

Over the years, I learn that there is nothing wrong with using OWL reasoning that is beyond DL. In fact, it is almost desirable to do so for many different reasons — this will be a subject for some other time. When I wrote “Misconceptions about OWL Reasoning“, I intended to share my learned lesson with the readers, so that they don’t have to rediscover what I have already discovered — it’s okay to use whatever the kind of OWL sub-languages and reasoning as long as they help you to solve real-world problems.

People love blogs. It’s only a matter of time for companies to figure out how to exploit this exciting technology to increase their revenues. Amazon.com is one of those smart companies. According to this IHT article, in Nov. 2005, Amazon.com begins a new program called Amazon Connect. In this program, it has recruited a group of about a dozen authors to blog about books and whatever they want to share with the readers. For example, Meg Wolitzer has an Amazon Blog.

I think Amazon Connect is a great program — it provides a new infrastructure for the authors to interact with their readers.

The US economy is growing at a healthy rate. What should we as IT employees expect in 2006? According to this BW article, the insight is as the follows.

• Shifts in the economy will continue to benefit highly educated workers at the expense of less-educated ones, who are more vulnerable to automation and competition from cheap foreign labor. Management consultants, architects, engineers, and the like ought to do well.
• Employment and pay in high tech could accelerate in ‘06 as business finally gets over the late ’90s tech bubble.

This is excellent news. As big companies begin to replace their old IT solutions and adopt new ones, we expect people with IT careers to do well in 2006.

In a recent CNN interview, Wired magazine co-founder John Batelle was asked the question “What is the next big thing on the Web?” His answer is the Semantic Web.

The idea to create a semantic Web where everything is described not by one researcher and his team but rather by all of us as we root about the Web. The idea is that we might get to the point where everything in the world of value is in the index correctly, whether it’s your car, your child or whether it’s a media object like a page or an audio file or whatever, or in this case a picture. And then you create these vast semantic attachments to everything and that becomes the seedbed for the next generation of search to crawl and make sense of.

I believe Semantic Web technology is more than just a new search technology. It will change the way we store, discover, and share information — among the humans and the machines.

The current OWL language specification purposefully defines three different OWL sub-languages, OWL-Lite, OWL-DL, and OWL-Full. Each of the sub-languages has different usage limitations and has different expressive power.

In general, I think this classification is helpful. For example, it helps reasoning engine developers to label their OWL reasoners, and it defines a clear guideline for the parser developers to create validators for a particular OWL sub-language.

Nevertheless, this classification could create misconceptions that lead people to wrongly interpret the properties of different OWL reasoning. For example, a common belief is that when you build an ontology, you should attempt to stay within OWL-DL because OWL-DL reasoners is more efficient that OWL-Full reasoners. Some others believe that the use of OWL-Full will always cause intractable computation performance in a reasoner.

It’s kind of like saying, “reasoning over OWL-DL is solving a problem with polynomial time complexity, and reasoning over OWL-Full is solving a NP problem”. I admit I was one of those people.

This belief is flawed.

There is little doubt that a highly optimized OWL-DL resaoner runs faster than a OWL-Full reasoner, especially if that reasoner is built on a first-order theorem prover. However, in reality, solving many practical problems often does not require a general purpose OWL-Full reasoner, even if the ontology is OWL-Full.

Here is an example,


(urn-x:p1 rdf:type foaf:Person),
(urn-x:p1 foo:reads bk:Comics),
(bk:Comics rdfs:subClassOf bk:Book).


Say I want to deduce all people who read comics are foo:ComicsReader. I define the following Jena rule:


(?p rdf:type foaf:Person),
(?p foo:reads bk:Comics)
=>
(?p rdf:type foo:ComicsReader)


I’m quite confident that the performance of this rule is no worse than a rule that reasons over the subClass-superClass relation between bk:Comics and bk:Book.

In summary, the classification of OWL sub-languages can be useful in guiding the development of OWL tools. However, the language reasoning support labels (i.e., DL, Full) could create misconceptions among the novice users of the OWL language.

A student lost her purse and a USB key with her only copy of the master thesis. What does she have to do to get her thesis back?

She was going to retrace his steps, go to every store he hit. She would talk to security guards, check lost-and-found, scour the parking lots.

So that day, she drove to Greenbelt, and as soon as she parked she saw a big trash bin behind a Wendy’s. She started pulling out broken-down boxes. She didn’t care about the trash, even if it was greasy slop from a fast-food place. “No cockroach, no rat, no creature from the dark was going to keep me from my jump drive,” she said. “Nothing is as bad as the thought of rewriting that thesis.”

And there, at the bottom, was her black leather purse. She unzipped it, reached in, and felt her fingers close around — her jump drive.

Man, I’m glad I didn’t have to go through this like her.

We used to believe that it’s difficult to “teach” computers to play good chess. With advancements in AI, it turned out that our old belief was false. It’s relatively easy to build computers that can beat chess masters.

CNN.com runs an article that describes the latest chess war between the humans and the machines.

This is what Veselin Topalov has to say about his enemy:

“I find it fun playing computers. The only problem is that the psychological duel does not exist. You cannot bluff. You cannot count on unforced errors. You have to find a special strategy completely different from what you would do against humans.”

So just how good is our computer players?

Experts say machines out-compute people by a rate of around 200 million moves per second to one…