The simple use case is to display to the operator of a terminal the elapsed time in seconds to load a file of RDF data. The single assumption is that the data is encoded in JSON according to this Talis specification. The single constraint is that the load time must be less than 10 seconds for a 2.5MB load.

The non-functional capabilities of the solution are more interesting. I use the encodings in Paul Tarau’s Isomorphism library both to encode the data as numbers then later recover the original text. The functions to reduce the data to a directed graph (r2dig), the function that finds the indices (ndcs) and those that recover the text from the encoding (recover and decode) are the most original work. The solution includes a graph matching function indirectly related to the use case. This solution could be developed further as a triple store with various functions that take advantage of the encodings.

The high assurance reference speaks to use of the GHC type system as well the practice of verification through QuickCheck properties. This practice implies more than coding properties. It implies devising a verification approach that both avoids selection bias and proves mathematical properties specific to the solution. No doubt the Haskell type system supports the high assurance claim. With some excellent advice from David McBride, the Talis specification of the JSON encoding provided sufficient guidance to construct then refactor data types and associated instances of the FromJSON type class in Bryan O’Sullivan’s Aeson library.

The functional reference speaks to evolving my Haskell style.  The functional style became more natural to me as I further developed the solution. Haskell is the fourth programming language I’ve adopted in my career.

As important as the language are the practices in delivering a high assurance, high(er) performing solution. In addition to documentation with HUnit and verification with QuickCheck, those practices include the following: examining symmetric multiprocessor parallelism (smp), profiling, exploring heuristics (-K and -H flags) in the compiler and run time system, inlining, optimizing for fitness with a genetic algorithm.

Inlining and heuristics were the most useful practices to reduce execution time. A manual binary search, above and below abductive thresholds allowed the use of heuristics to reduced execution time.  Inlining the r2dig and ndcs functions reduced execution time from 10s to a little more than 5s on my Quadcore i7. Pure SMP was not fruitful. Early indications were that an SMP approach would  be comparable to a single thread. I did not explore Parallel IO. I welcome any suggestion on how to incorporate SMP.  I ran a half dozen simply configured Acovea sessions with some optimization. See configuration and reports.

You will find the files on github/hafr. I welcome feedback on refactoring, style, performance tuning, etc.

Looking ahead to 2012, I will be extending my high assurance capabilities by investing more time in Hets. The functional target is Information Flow. I’m still waffling on the Isabelle v. Agda decision.  You can do that in Washington.

Values, Types and Foundations is a broad survey presentation and demonstration. The presentation very generally describes a theoretical foundation for OMG standards such as the Meta Object Facility and the Object Constraint Language. By general I mean a computational, logical and philosophical overview through System F, the Calculus of Constructions, Martin-Lof Type Theory and dependent types. My goal was to give the talk that Philip Wadler would give to the OMG. The presentation also contains an informational brief on Vladimir Voevodsky’s Univalent Foundations program at the Institute for Advanced Studies.

The first four demonstrations use most of Oleg Kiseylov’s Interpreting Types as Abstract Values, Constructive Law of Excluded Middle, and Impredicativity Bites. I don’t add much of anything new to the demos. I compare and contrast the test cases Oleg provides and dump the stack traces to speed up the demonstration time. That’s way more than enough.

The last demonstration is based on Anton Setzer’s Interactive Theorem Proving for Agda Users.

Other than that I review some background on Types, Classes and Sets based on Peter Aczel’s What is a Set? Thanks to Cory Casanave for pointing out the importance of providing this background.

During the voice over and at the conference I contrasted the approach in Values, Types and Foundations with that of ISO 24707 Common Logic (CL) which is a type free logic. Pat Hayes describes CL as first order syntax and higher order semantics. Hmmm. I recommend Chris Menzel’s  Knowledge Representation, the World Wide Web and the Evolution of Logic for those  interested in the design goals that motivated ISO 24707.

I plan to step out a bit more in 2012. I will be connecting Values, Types and Foundations back to some topics I wrote about here on the Phaneron a few years ago such as  the Heterogeneous Tool Set and the Information Flow Framework. The wood shedding has paid off and I’m looking forward to catching up with a few old friends and mixing it up with some new folks.

Quiet on the Phaneron means there will be much to say. As advertised, the first six months of 2011 have been about thinking small. Yes, I’ve been in the woodshed again and truly enjoy the solitude. Much like an earlier period in my life when internalizing fingerings, modes and inversions required the quiet to allow careful concentration, so have these past six months allowed for examining nightly releases of the Glasgow Haskell Compiler, its Regular Shape Polymorphic Parallel Array (REPA) library and dependent types in a AGDA.

After having taken a pretty close look at topics in software development and information technology architecture over the past few decades I’m very excited to get under the hood with GHC and its back-ends. I hope to share some experiences by year end on optimizing GHC back-end selection and configuration management on my 4-core i7 using genetic algorithms as described by Don Stewart in Evolving Faster Haskell Programs (now with LLVM). I also hope to get access to some cooler hardware, possibly through some academic connections.

Closely examining REPA seems a good fit for my development in Haskell. Converting Paul Tarau’s list-based implementation of this bijective mapping between Godel numberings and Term Algebras in REPA is a valuable exercise in both learning this library and the tacit knowledge I need to become a journeyman on GHC. The next stage of this exercise will include verification with Quickcheck. I’ve also taken the opportunity to do a retrospective on REPA by examining some of the early work on parallel list notation [:e:] and state transformer monads.  There’s no shortage of great papers out there on Haskell. See my del.icio.us bookmarks. This retrospective will provide the right foundation for later investigation in Data Parallel Haskell. Seems there’s time before the next release.

It looks like I’ll drop my plans to build expertise in Isabelle/HOL in favor of Agda. I just don’t have the bandwidth to internalize that much. Agda allows me to leverage my Haskell investment.

If you read A Winter of Philosophy and Logic, you’ll also be interested to know Valdimir Voevodsky recently announced a special program to develop the Univalent Foundations of Mathematics along with Steve Awodey and Thierry Coquand. Voevodsky has taken an interest in Martin-Lof type theory. He’s demoed some proofs in Coq, but he’s also active on the Agda listserv. And Coquand & Co seem to be pro-Agda. Voevodsky’s stature in math circles could do a lot for this subject. I’ll be tracking this all pretty closely.

No, I don’t plan to become narrow-minded, but to think more clearly: to think as a logician. Until just a few years ago I was a big thinker, in fact a Very Big Thinker, also known as a possibilian. The approach worked worked quite well. I became suprisingly good at predicting technology trends and that was a pretty handy skill in my profession.

Before I talk more about thinking small, let me clarify my understanding of what it is to be a possibilian. David Eagleman loosley defines a possibilian as one who holds multiple hypotheses at the same time. In this very entertaining video from Poptech 2010, Eagleman describes the possibilian’s beliefs as knowing too little to commit to atheism while knowing too much to commit to a single religion. It would be an understatement to say that puts the possibilian in difficult position, not because uncertainity is untenable, but because the possibilian is left in a non-deterministic state. Later in the video Eagleman carefully differentiates possibilians from agnostics. Where agnostics are noncommital due to lack of evidence, possibilians experience the world free of dogma and full of awe. A possibilian who holds multiple hypotheses to be true that aren’t or can’t be evaluated and remains in a non-deterministic state, free of dogma and full of awe is the equivalent of a deer in the headlights in a world subject to natural selection.

Possibilians need survival tools and their Swiss Army knife is thinking small. Thinking small means resolving the meaning with the intended receiver of every term we use. Antipatterns to thinking small include compression to morphemes in communication patterns driven by smart phones; pervasive meme replication; and a disturbing pattern of widely used terms that remain undefined. Consider SOA after ten years. Whether we coin a new term or introduce a new term, we’re responsible for its meaning. This branch of logic is called grammar.

Thinking small also means resolving premise to conclusion. Observation, exploration and appreciation are surely part of the joy of life, but whether it’s reason, belief or instinct that allow us to conclude, conclude we must. Thinking small requires us to conclude based on the information available while not only acknowledging the fragile conditions on which we base our system of deduction, induction and abduction, but also accepting that pure chance is at play in our lives. This branch of logic is called critic.

Thinking small also means that we evaluate the strengths and weaknesses of our approach to conducting an inquiry. We understand the strengths and weaknesses of our conclusions and we attempt to resolve our own inevitable selection bias, self defeating, or contradictory beliefs by looking both inside and outside ourselves. This branch of logic is called methodeutic.

So I resolve here on New Years Day to think small in 2011.

When Tim Berners-Lee published the W3C Linked Data design issue in July of 2006 he introduced four rules that foster a “post semantic” Web. The term Semantic Web had become less associated with the value proposition of broad societal adoption and more closely associated with narrowly defined enabling technologies. To make a long story short, some rebranding was in order. The Linked Data meme rebrands the Semantic Web. More importantly the Linked Data rules redefine the necessary level of commitment to participate. They encourage a wider audience and allow for wider choice in the enabling technologies, both on the up and downside.

The rebranding continued in November of 2007 with introduction of the term Giant Global Graph (GGG). The GGG is a non-technical term that represents the societal value of the enabling technologies in this “post semantic” Web. These technologies are the “tools which allow us to break free of the document layer” while we “cede control for greater benefit.” John Sowa has proposed these tools will not be limited to RDF and SPARQL. And when the government releases data back to the public, it bootstraps ceding control on the GGG.

In Working Ontologist, Dean Allemang and Jim Hendler introduce the AAA Slogan: “Anyone can say Anything about Any topic.” But that doesn’t mean the technologies underlying the GGG are one inconsistent RDF graph. Linked Data encourages information to flow through a web of constraints on the GGG. These constraints preserve semantics locally on the GGG. RDF semantics is one such constraint, but not the only one. SQL semantics might be another; Common Logic another; smaller logics like SHIOND, or LJT are others. Together the constraints form a web of their own. And this web of constraints are the regularities through which information flows. Think of the GGG as WEBS of data, not THE web of data.

So, how will this all happen? There are signs that its happening already: Drupal‘s rise in popularity. Some signs are good: publishing RDFa straight from a CSV. Some anti-patterns are emerging: extracting otherwise useful data, converting it into RDF, serializing it across the wire in a highly inefficient form, disregarding inference, stripping the RDF out and publishing it as a Web page. Our general understanding of what’s possible remains limited. Is our only choice a Semantic Web Architecture [based on] A Stack or Two Towers? I don’t think so. Are Datalog rules our best option and do we have to drop logic from the Semantic Web Layer Cake? I think not.

It’s all quite early just yet, but I think its possible to suggest a few recommendations for both the public and private sectors – some social, some technical – to allow information to flow across the GGG based on a web of constraints:

1. Identify publishing and linking roles for Linked Data initiatives. They are not the same. Be able to describe the activities you assign to each of those roles;
2. Acknowledge that we’re working at the edge. Where we need specifications for publishers and linkers, develop edge specifications. Edge specifications reassemble existing vocabularies like Dublin Core and XBRL and provide new terms only when needed. Avoid the urge to develop yet another core specification.
3. Recognize that regularities are what allows information to flow. Model theories, boolean valuations and proof interpretation are what creates these regularities. Without regularities we’re just hacking up something on a global scale. And that’ll delay achieving the potential of the GGG;
4. Use RDFS semantics instead of vocabulary, or terminology based approaches. These approaches retain the limitations of original intentionality. There are some great patterns for linking open data serendipitously in Working Ontologist. Create your own patterns.
5. Avoid death by layering. There were some known risks to the layer cake approach when it was devised and they’ve proven themselves unavoidable. LBase describes an embedded semantics a lot like what some smart folks are advocating for Common Logic;
6. Avoid anti-patterns like intervention and curation. Enterprise Information Management is a valid, separate discipline than Linked Data. The value proposition of Linked Data implies something a little short of serendipity. If you find yourself tempted to do lots of curation, its not Linked Data.
7. Study Institutions and functional languages like Haskell. Functional languages are very convenient for transforming among languages, logics, models and theories. Institutions formalize and standardize those transformations.

Technology trends, at least in my small corner of the world, appear stable for at least the next five years.

I didn’t follow hardware trends much until a few months ago when one of my machines went down. Hardware trends are pretty easy to read. I have a new Nehalem i7 720 on order. It cost a few bucks, but my private cloud in a box is on a UPS truck somewhere.

Scale it? you bet. Haswell: double the cores, double the threads by default in 2013.

Luckily a year ago I made a good choice in programming languages: Haskell. Language trends are harder to read. Think there’s a practical use for Second order Lambda Calculus with Polymorphism? I do. Category theory is embedded in the core libraries of the language which makes incorporating information flow and the logical environment of the Information Flow Framework a natural fit.

I’ve been active recently on the Object Management Group’s Architecture Ecosystem Special Interest Group (AE-SIG) as well as the Ontolog Forum’s Sharing and Integrating Ontologies (SIO-Dev) activity.

I’ll take type inference, theorems for free and the EU funded CASL in HETS.

But wait, there’s more.

Software transactional memory and shared memory parallelism in a pure functional programming language.

In the immortal words of Hannibal Smith: “I love it when a plan comes together.”

If you follow my tweets, you saw some of it unfolding. Readings from and about Russell’s theory of types provided insight into ramified type hierarchies that remain central to much of the Object Management Group and the World Wide Web Consortium‘s standards development. Proceeding through the development of type theory since Russell, constructive and intuitionistic logic became the central focus of my readings. I am very pleased with the amount of original material I was able to find on this subject. Brouwer’s Intuitionism and Formalism is just one example. My bookmarks are worth browsing for these original materials.

A longer journey into type theory took me through Martin-Lof’s impredicative and predicative type systems, Girard’s Paradox and Coquand’s Calculus of Constructions. These readings brought into focus as sharp as one might get at noon on a hot Summer’s day, the past five years of our team’s work on Model Driven Architecture and the Semantic Web. Following on Coquand’s work, I progressed deeper into algebraic specifications with the EU’s Common Algebraic Specification Language.

Most importantly, I made very good progress with tooling in each of these areas. Proof General is showing itself a very useful proof assistant. Haskell, Isabelle, oCAML and Coq all provide open source licensing that made it possible for me to get my hands dirty. It’s all free. Free as in Freedom !

I’m confident enough with what I was able to accomplish this Winter that I registered the methodeutic.com domain name.

But Winter has passed and it’s a warm Spring morning here in Alexandria. I’ll talk more about this work over the next few months. Right now the Spring weather’s so nice, I’m heading outside !

-Charles Peirce 1906

If you didn’t attend this year’s International Semantic Web Conference (ISWC 2009) the good news is that you now have the chance to see the Pat Hayes invited keynote called BLOGIC or Now What’s in a Link?.

Pat’s keynote is a must watch video. His keen insights into Web Portability, Names and Identification, the Horatio Principle, SameAs and Death by Layering will shape the future of the Web for the better. After Pat’s talk I had the pleasure of speaking with Dame Wendy Hall and Wendy believed Pat’s talk was precisely what Web Science is about.

In this post I won’t reiterate the main points of Pat’s talk. Be sure to watch the video. Regular readers on The Phaneron will recognize similar points already made here over the past few years. I will take the opportunity to elaborate on what I believe to be a few important lessons from Pat’s talk that are revealed through his incidental comments. These incidental comments speak volumes about what many of us experience throughout our careers whether in the workplace or working with standards organizations.

The lessons are these:

1. The question we’re asking has, in some surprising cases, already been answered. As computer scientists we have little opportunity to dedicate the time required to study the important work of the giants that came before us. Philosophy, logic and mathematics have a very large body of literature that takes years to truly understand. Peirce is just one of many whose writings contain answers to questions that we could otherwise ponder for decades only to arrive at the same answer. As Pat says to Tim in the video, Peirce solved the same problem as RDF with Existential Graphs in 1896. And he did it right!
2. Our feelings really are useful indicators of when something is or isn’t right. Pat talks about having a sense that Bnodes weren’t quite right during the specification of RDF. In hindsight Pat had the right intuition and he has proposed a backward compatible solution based on Peirce’s Sheets of Assertion. There’s a Myers-Briggs story to be told that’s especially meaningful to me. I’m an INFJ. Over the last few years I’ve come to recognize many occasions where my feelings were good indicators of the truth. Although a cursory reading of Myers-Briggs may lead one to believe feeling (F) and Thinking (T) are in opposition, they are not. Our feelings are as good indicators of the truth as is logic.
3. We often overlook advice only to learn later of its immense value. This happens when we’re just not ready to learn something. What I learned from Pat’s story that “John Sowa showed me Peirce, then he showed me Peirce again, then he showed me Peirce again” is that sometimes we’re just not ready to hear the advice we’re offered. I’ve been overlooking the advice to take Common Logic seriously for a few years now and Pat’s talk convinced me its time to take a serious look.

There’s another relevant point from Pat’s talk that should not go unmentioned. Pat’s talk was given at ISWC, but could equally well have been given to the Object Management Group. The OMG recently issued a request for proposals for a MOF to RDF Structural Mapping in Support of Linked Data. The contents of the RFP imply that the OMG faces many of the same challenges as W3C. The OMG would do well to study W3C’s lessons learned from RDF.

The RDF/OWL community continues to work through important issues in realism and nominalism and I thought it useful to relate an important conversation during a break between sessions. I asked Peter Patel-Schneider how the description logic community came to use the terms Concept and Role. Peter told me there was a perceived need in the 80s within the description logic community to differentiate the language of description logics from the language of object oriented programming. The sentiment was simply that the terms they used should seem more related to logic. It was nothing deeper than that. The object oriented programming community had already chosen the terms Class and Property. Those terms were out. And there’s no evidence the description logic community understood the significance of the term Concept as it relates to semiotics, realism and nominalism.

I was also part of a conversation in which someone was asked to reflect on the philosophical implications of ontology imports. Of course I was intrigued and listened carefully to the response. A question this general is difficult to answer but it, as well as the very nice presentation on the Simple Knowledge Organization System (SKOS) given by Rinke Hoekstra, confirmed my intuition that government, industry and the Semantic Web community is seeking a deeper understanding of how realism and nominalism apply to implementation and interoperability. A better understanding of model theory can help, but controlled vocabularies, ontology importation and more generally ontology alignment within and across domains can also be better informed by identifying semiotics as domain.

Figure 1.

Since OWL ED and the International Semantic Web Conference (ISWC) 2009, there’s been a useful exchange on the SKOS archive called Using DBPedia Resources as skos:Concepts. Pat Hayes brings to light some important issues in his comments which begin here.  You can find my comments related to figure 1, the interpretant triangle here.

The evidence from OWL Experiences and Directions as well as the SKOS archive indicate that better understanding the significance of realism and nominalism will remain important for many in 2010.

After some reorientation, I am becoming more comfortable working with higher order logic and Haskell. The Isabelle/HOL distribution is very nicely done and the setup on my OpenSolaris machine was really quite painless. Contrary to some opinions about Haskell literature, there’s a lot of great information available. Unlike Visual Basic, Java or C# you won’t be spoon-fed by a publisher or vendor. I highly recommend the AJT Davie’s Introduction to Functional Programming Systems Using Haskell. It’s one of those little books that provide the reader with enough of a perspective of the subject to cause the reader to develop their own approach and understanding. Thanks to Mr. Goodchord, one of my guitar teachers, who many years ago for helped me realize how useful these little books can be. The real joy is in developing the deep understanding that comes by educating oneself on a rich and fruitful subject like functional programming the value of which can last a lifetime.

]]> http://phaneron.rickmurphy.org/2009/11/observations-from-owl-experiences-and-directions-2009/feed/ 2 http://phaneron.rickmurphy.org/2009/10/open-government-linked-open-data/ http://phaneron.rickmurphy.org/2009/10/open-government-linked-open-data/#comments Fri, 23 Oct 2009 15:06:10 +0000 rickmurphy http://phaneron.rickmurphy.org/?p=37 It was about this time a year ago when it became clear to me that that the Obama campaign’s fact sheet called Connecting and Empowering All Americans Through Technology and Innovation implied developing a government collaboration platform based on Linked Data. Shortly after the election in November and while serving as an invited expert on W3C’s electronic government interest group I wrote the Open Government: Linked Open Data use case that called for “new set of information technology architecture principles that align open government with citizen engagement in a networked society enabled by linked open data.”

Now a year later I am excited to recommend the Building Semantic Web Applications for Government half-day tutorial at International Semantic Web Conference (ISWC) 2009. This tutorial builds on the work that Dean Allemang, Chief Scientist at TopQuadrant, lead for our team the the General Services Administration as well as TopQuadrant’s emerging Ontologies for Electronic Government. In Dean’s own words “The charter of government is to be responsive to the people. As such, government information belongs in the hands of the governed.”

Today, Linked Data is the ONLY Internet-scale collaboration platform for open government. As I have written previously the long-term success of Linked Data for Open Government depends on an information life-cycle where publishing and linking co-evolve with minimal intervention. In the sort-term and to paraphrase John Sheridan of the UK’s Office of Public Sector Information, shiny objects and tiny cost allow the groundswell of innovation to reach a tipping point.

While you’re at ISWC09 be sure to attend Dean’s half-day tutorial on Monday October 26 from 2:00pm to 6:00pm.