Welcome to the Frontpage
Relative Consistency Proofs PDF Print E-mail
Written by Administrator   
Saturday, 27 December 2008 21:27

In the course of a proof, at each step, when you make an inference, you have to protect the inference from leading to a false conclusion, because the axioms and or rules of inference could very well be inconsistent.  What assurance do you have for the safety of the conclusion?  How has it been secured?  And where would you put a proof of the consistency of the axioms and the rules of inference?  Ultimately, perhaps this must reside in the mind of the prover.  It is here that the prover may be tricked.  If she wishes to verify some set of axioms and rules of inference, she must either do it in another proof system, or in her head (or perhaps I am mistaken).  Anyways, this is what a fresh pair of eyes yields upon rediscovering mathematical logic.
 
Raiders of the FTP Sites PDF Print E-mail
Written by Administrator   
Saturday, 08 November 2008 01:58

While the FRDCSA has a ridiculous number of projects ongoing at any
time, a recently developing project has taken my fancy because it
promises to offer good dividends.  The project, called "Raiders of the
FTP Sites", well, speaks for itself.  It is a system that searches FTP
sites for interesting artifacts and retrieves them.  It works by
correlating subject matter of interest with FTP sites, extracting a
recursive directory listing of the FTP site, and then performing
several analyzes of the contents.  While still a very immature system,
it has yielded up a labelled resume corpus which will help the
job-search program.
 
Oct 23 2008 Update PDF Print E-mail
Written by Administrator   
Thursday, 23 October 2008 17:39

Three new useful AI systems have been successfully configured and tested:

  • Sentiment Analysis using OpinionFinder-1.5
  • Query Expansion using LucQE
  • Natural Language to Logic Conversion using ResearchCyc

Unfortunately, I believe only LucQE is free software.  However, having these abilities will improve the analytic capabilities of the FRDCSA to help with the mission of promoting the use of free software.  For instance, the Sentiment Analysis system can determine who is expressing bad opinions about us, allowing us to automatically avenge them.  Lol.  Indeed it unfortunately could be used for that purpose.  However, it is intended to be used for MPQA (Multiple Perspective Question Answering), in order to get a better idea of subjective statements.

In other news, I am searching for research interest ontologies, in order to have an idea what to search for in terms of new systems.

 

Last Updated ( Thursday, 23 October 2008 17:39 )
 
Overview of Today's Work PDF Print E-mail
Written by Administrator   
Saturday, 27 September 2008 05:09

Now that I have this website, I am going to blog to it about what gets done at the FRDCSA.

I added a lot of new features to RADAR today.  First, radar-web-search has been extended.  This program originally allows for one to search for a topic on the net, say, "event extraction" after which it will build a large search for all the software on the net.  It then searches Yahoo and looks at each page, to see if any software is linked:

andrewdo@box:/var/lib/myfrdcsa/codebases/internal/event-system/IE$ radar-web-search "event extraction"
QUERY: "event extraction"  system OR java OR project OR library OR php OR web OR framework OR open OR manager OR linux OR engine OR net OR server OR management OR game OR tool OR tools OR client OR simple OR editor OR cms OR database OR\
 file OR generator OR software OR network OR xml OR python OR based OR source OR plugin OR data OR amp OR language OR application OR control OR online OR toolkit OR interface OR 3d OR irc OR eclipse OR free OR api OR windows OR code OR \
os OR perl OR virtual OR development OR gui OR driver OR content OR module OR mail OR image OR suite OR player OR portal OR monitor OR platform OR simulator OR script OR object OR log OR media OR text OR easy OR browser OR search OR ser\
vice OR viewer OR de OR chat OR remote OR parser OR mysql OR time OR bot OR mobile OR converter OR sql OR daemon OR tracker OR rpg OR programming OR test OR gnu OR environment OR class OR utility OR gnome OR compiler OR internet OR 0 OR\
 user OR utilities OR html OR package OR desktop

Result: #1
Url:http://www.pubmedcentral.nih.gov/articlerender.fcgi?tool=pubmed&pubmedid=15494078
http://www.pubmedcentral.nih.gov/articlerender.fcgi?tool=pubmed&pubmedid=15494078
Summary: With the explosion of molecular data, tools developed by computer scientists are ... BIND-The Biomolecular Interaction Network Database. Nucleic Acids Research. ...
Title: PASBio: predicate-argument structures for event extraction in molecular ...
$VAR1 = [];

Result: #2
Url:http://nlp.cs.nyu.edu/info-extr/
http://nlp.cs.nyu.edu/info-extr/
Summary: This system combines a web crawler (which searches for reports of outbreaks on a ... engine, and a data base browser to examine the extracted events (Proteus Project ...
Title: Proteus Project: Information Extraction
$VAR1 = [];

 

So, that's what it does.  But a problem it was having is it only looks one layer deep for tar.gz and zip files and the like.  I wanted it to look further, but that would have been bandwidth and time expensive, searching all the links.  So what I did was to download a dataset from:

http://cybermetrics.wlv.ac.uk/database

which contained a large dataset of web links.  I then rated the last dir or file of the url that linked to a set of files for how many "desireable" files were there.

i.e.  in the above url it would be "database"

17.0028327481393        jars    40      9       211
14.3298883847943        download.html   116     168     498
12.9734278116825        edit.html       67      33      67
10.8830909612802        patches 8       2       368
10.4918735220202        Debug   38      16      42
10.0194507146122        golem   29      6       30
8.97441185481296        canaries13      20      0       20

So I added that and now it can speculatively search 1 extra ply.  It has already helped to find some new software.

 

Secondly, I went ahead and added the ability to search within pdf and other documents and extract the URLs from them, so that research papers (which often link to systems, or at least name them) can be searched as well.  This is a separate script that will integrate eventually with radar.

/var/lib/myfrdcsa/codebases/internal/radar/scripts/get-software-by-searching-publications.pl

 

Additionally, in unrelated news, I worked on better calendar integration.  Instead of having to enter the details manually, I wish to be able to simply highlight some text that contains the event details and have it extract the information.  So I searched around for my old event dataset which turned out to be called sa-tagged.  It only had 48 or so items IIRC.  Not a lot to develop a good model.  Recalling a conversation with Dayne that he would like to go with CRF (conditional random field) learners in the resume analysis domain, when I downloaded mallet which I assume McCallum used on sa-tagged, I noticed it had CRF so I decided to use that.  Setting it up is not a completely trivial, I have to write the feature processor.  I know how to do everything, except add additional linguistic features - for that I am looking elsewhere - perhaps to Stanford's NE recognizer since it has one - if I can figure out how to access it - should be easy but need to look first.  Getting tired now.  So, if anyone is interested, tomorrow, I will hold a seminar on doing this.  That's it for now.  Happy hacking!

 
Motivation for Our Programme of Software Conglomeration PDF Print E-mail
Written by Administrator   
Friday, 26 September 2008 00:50

Let us give a rough outline of the logic behind this decision.

First is that one must have a vision, that is an understanding of a goal, such as artificial intelligence.

In my case, I could clearly see situations, usually social, in which I could imagine a computer performing a very complex set of tasks to arrive at some socially productive answers.

Unlike people untrained in computers and logic, I could not specifically disprove that a computer could do this. Besides, the risk of failure to achieve this goal is very large. <<<opportunity cost>>>

So I began to try to create a system. Like many who first start out, I fell into basic traps. I thought that I could write metaprograms, programs that wrote other programs which were more capable than themselves. Of course this ignores fundamental computer science concepts. It is based off practical work with computers, not theoretical. Of course, the lisp language is specifically crafted for this kind of metaprogramming, and it is a very basic possibility for automata to rewrite themselves, so like most beginner AI ideas they are hardly novel. So the idea is actually quite good, but it is not complete, it is only a fragment. It is necessary, but not sufficient.

In fact, most beginner AI ideas sound quite fuuny to people who know what has been developed. A beginner AI researcher will come along and say - we can have programs that map out all the words that people use and make a meaning for it (a dictionary?). Or maybe a more advanced suggestion - we could have mysql databases of available software - with all the dependencies recorded (apt-get?). So it is really quite funny listening to someone say basically - we could have round things that we could roll around on, as if the idea were new.

There are many other basic traps that I fell into. Another interesting one involved trying to model systems from data. I thought that I could have a system that observed interactions and "abducted" models of them. Well, this is quite true and there are many good uses of this, what is called model-checking (although model checking usually works from specifications and not observations). But ultimately, what was most damaging was that I had completely ignored the issue of what I would make my models out of. I thought programs, and this is fine, but really what is necessary is the notion of an object language to be learned. Of course, since these ideas are not well defined, there are many related concepts which are categorized by them. It covers quite a broad range.

So, the high point of my research came when I studied existing research extensively, which I had previously tried not to do for fear of confirmation bias. I am happy to say that I do not feel that I have been biased (any more than necessary) by my studies, but only that I came across results which tended to make my own notions more rigorous.

Also, during this period of intense study, where I would attend graduate classes from 8 in the morning until 6 at night, and then read at night, I was really fulfilling my neurological needs as an Asperger. Although I didn't know it at the time, my obsessive need to collect every fact, which is similar to my obsessive need to look at every possible move on a chess board, was really a fulfillment of my needs.

I made a few assumptions about my work on AI. First, without actually learning of the distinction between strong AI and weak AI, I realized that there were indeed two classes of AI, and that weak AI seems a prerequisite of strong AI. In fact I was relatively certain because strong AI would be capable of achieving weak AI on its own. So either way, weak AI would be developed.

Weak AI means to me problems that are solved by existing mathematical and computational technologies - that is, all problems that are computable by classical computers.

This begs this question, "what is the class of problems that computer can solve", and also this, "can they solve all formalizable problems about their own behaviour". In fact, my initial work centered on finding this program, but I did not know how complex this program would be.

What tended to startle people in the earlier part of last century was that this program did not in fact exist. It had been called by various names in the past - the Entscheidungsproblem, the Hilbert program. Several people simultaneously realized that this program did not exist. This included Church, Turing, Godel, Zermelo, Russell, King, Wittenstein, (even Socrates). Even Von Neumann was startled at this fact, which, in retrospect, is entirely obvious.

But this fact is not at all a "negative" result as people seem to claim. Indeed this fact presents the solution to us very clearly.

That is, for any program, there exists another program which is larger and which can solve a strictly larger class of problems.

So, based on this relatively simple analysis, I compared the efficiency of these ideas. Since it is known that the time complexity of solving problems, as well as the extent of computable problems, is strictly increased with certain additions to program length, it was rather obvious which additions would provide the greatest increases. These would be the software systems with the greatest capacity relative to other systems. They would not be systems that I wrote myself, but would be systems that other people, scores of other people, had written with lots of funding.

That is, our operative thesis is that the systems capabilities are increased the most by collecting the best software that others have written. Since we do not know, ahead of time, which software is most useful, we collect all of it, and apply decision methods and ratings and tests to all software, and package applications ordered by ease of packaging and utility after being packaged.

Many people have problems with the fact that we do not uniquely specify which problem we intend to solve. This is a characteristic requirement which Asperger's do not have. The fact is that we have proven the systems capabilities have increased. Furthermore, we simply make the claim that we can, transitively, solve every problem which the system we collected was designed to solve.

But in practice there are very few attempts at collecting software. The reason for this is most people are motivated to solve a specific goal. But solving a particular problem is usually a waste of time. It forces a long and complicated search (naturally because the system is weaker proof theoretically than a larger system). Collecting software on the otherhand, in leiu of automatically solving more problems later, is very important. But in order to solve more problems, a certain minimal rate is needed to have a self sustaining reaction.

That is why I implore you to help me find people willing to program packages for an indefinite period of time.

The cost of failure are infinite stacks, and we have proposed a most reasonable objective. Furthermore, the lack of such similar practices indicates a certain novelty, which may be attached to the probably of finding a gifted Asperger.

Of course, there is a limit to how much software we can collect. But we are far from that limit - and once we hit it, our situation will be quite different. We will have shared systems that are in many respects extremely capable. Until we collect this software, we realize, like Godel's S and S', that there are infinitely classes of theorems that we cannot prove with our systems.

Einstein, Godel, Newton - all the greats had Asperger's. It is ironic and unfortunate that by virtue of the very skills I have, and the skills I apparently don't have, the skills I have are ridiculed, and the skills I don't have are taken as an excuse to not send me to a school where I belong. Everything we see is a product of this misappropriation of talent. In highschool I worked at Fermilab as an experimental physicist, with a classmate. She went to Princeton as a physicist. But everyone thought I was lazy, because people didn't realize that my lack of executive skills, such as time management, allowed me to see things that other people don't. Now, daily my work, which is clearly of historical importance, is forsaken by those who claim to love me. Let actions speak louder than words. Let the truth out. Please stop murdering the truth. You are risking the entire stability of reality.

If somebody writes a 30 million dollar program and gives it away for free, and you don't have it, your genetic material will spontaneously mutate into that of an ameoba. That's like rejecting someone's offer to give you a space shuttle, on the grounds that you can't imagine why in the world you would need it. Moreover if you repeat this 5000 times, you are in trouble. No, I agree with your assessment, why not write it yourself? After all, there's nothing that 1000000 postdoctorate genius programmers could do in thirty years that you couldn't do in a week with C++, right?
 
More Articles...
<< Start < Prev 1 2 3 Next > End >>

Page 2 of 3

Login Form



Polls

FRDCSA should prioritize
 
Free template 'Feel Free' by [ Anch ] Gorsk.net Studio. Please, don't remove this hidden copyleft!