CORALOG

Compression Contest and Marctar's Axe

2006-08-15T13:49:00.000-04:00

Marcus Hutter has posted a 50,000 Euro prize contest for compressing (a subset of) Wikipedia. While I applaud Dr. Hutter for his generous contribution to stimulating AI research, and I do believe that cognition often can be viewed as compression (as championed by Gerard Wolff), I do have one reservation with this idea:

The idea that a smaller theory is a better theory can be traced at least to Ockham's Razor. When I applied The Cruncher to the frames of a (very low resolution, but continuitous) movie, I found that the best compression was when the 1st frame was described, then every frame was described in terms of the previous frame (much the way that mpegs work). However, I don't think this is how people do it. I don't record my day by remembering when I woke up then every instant in terms of the previous instant. If I did, I'd have to spend a good deal of time "unpacking" my day to tell you what I had for supper (and presumably less time to tell you what I had for breakfast).

Perhaps a more extreme example would be Euclidean Geometry. This can be compressed down to the 5 postulates (plus some inference rules), but I don't think anyone rederives a commonly used lemma every time they use it.

Therefore, I propose an alternate to Ockham's Razor, which I'll call Marctar's Axe: The best model is that which has the smallest average query time (in terms of steps of computation to answer it). By "query" I mean things like "How many times have you seen this pattern?" or "What is the usual outcome of this pattern?".

My suspicion is that compression will usually fall out of Marctar's Axe, but there will be some cases where Marctar's Axe will tell you to "cache" results and thereby trade in a lot of computation time for a little bit of memory.

Human Subject Research

2006-05-19T22:38:00.000-04:00

I recently had some dealings with the UMBC Institutional Review Board. All university research involving human participants and animal subjects is placed under review by this board to ensure some level of security in privacy, safely, and general well formed research methodology. We can, essentially, blame behavioral scientists that couldn't control themselves or their research (e.g. The Stanford Prison Experiment). During this review, I came across a recent undertaking by an MIT Media Lab professor, Deb Roy. A New Scientist article, Watch language grow in the 'Baby Brother' house, details the addition of audio and video recording devices to Professor Roy's home placed to capture the linguistic growth and development of his son.

Back during my time in the UMass Linguistic department, Tom Roeper gave me several notebooks full of observations made when his children were younger to transcribe. I'm hoping that more researchers take advantage of new technologies and share the collected data. Now, who is going to be the first to do what Deb Roy is doing but in a bilingual environment?

Without Numbers

2006-05-12T11:21:00.000-04:00

Pirahã, a language spoken in Brazil by peoples number in the hundreds, lacks quantification and subordinate clauses. Daniel Everett, a Manchester researcher, has studied Piraha extensively and even supplies a sample of Pirahã sung speech and a gloss of a "Killing the Panther" story.

"Pirahã is a member of the Mura language family. It has been traditionally referred to as Mura-Pirahã but this obscures the distinction between the Mura language family and the Mura and Pirahã languages. The Mura language family includes Pirahã as well as the extinct dialects of Bohurá, Yaháhi, Mura and Torá. It has been suggested that the Mura family is member of the Macro-Chibcha phylum but there is insufficient evidence to support this view. Pirahã is the only language spoken in the region of the Maici. It has no external affiliates and is in danger of becoming extinct as well. Numerous attempts have been made to learn and document this language since first contact with the Pirahã in the 1850's. Most efforts were inadequate or relatively unsuccessful as the language is extremely difficult and unusual. The Everetts have devoted the most time, collected the most data and been the most successful at learning the language and providing linguistic analyses for several interesting features of Pirahã." (link)

In 2004, Peter Gordon of Columbia University published an article in Science, Numerical Cognition Without Words: Evidence from Amazonia, and recently brought linguistic relativity back into the mainstream media with recent coverage in a Spiegel article about Pirahã. Daniel Casasanto counters with a letter to Science with two main points: Gordon's work lacked a control, baseline performance group and Gordon demonstration fails to support the causality claim. Another letter provides additional criticism.

Cabspotting for Inference

2006-04-10T13:03:00.000-04:00

Here is one of the Exploratorium's coolest projects: Cabspotting (BoingBoing and Metafilter discussions). The visualization plots in real-time a cab (a circle glyph that is filled in only if the cab has a passenger with a comet trail where the cab has come from) in San Francisco and prior trips (phantom, wispy lines).

Now, let's consider the above screen capture that I made of their Flash client. There's a clump of cabs on Market Street and near The Embarcadero which is clearer on the Google map of San Francisco. The southernmost cab seems to be going from Market Street to pick up someone and the westernmost cab was doing the same and has picked up the passenger. From these data we can infer "cab intent," major areas of commerce, and traffic dynamics. For example, consider the amount of time and distance it takes the cab to pick up a passenger and with a bit of taxicab geometry we might discover short cuts.

You can email the fine folks from their FAQ page which provides more project extensions. I just emailed them for access to their API and Flash client. Perhaps I'll strap GPS to something and display that and more.

More Squiggly Lines

2006-03-24T15:29:00.000-05:00

I realized that someone could use spectrograms to compose as well as recognize sounds. That is, with enough training, I could think of a sound (virtually any sound), then imagine what the picture of the sound must look like and draw it. It's like mastering an instrument: The Pickettheremin Musical Whiteboard.

With this in mind, I threw together a quick script so that I could point my webcam at a whiteboard, press a button on Lappidactus, then listen to the (...ahem) beautiful muzak [sic] produced.

Below is my first composition (both the sound and its spectrogram). It's called "Squiggly Lines I: John Cage Shooting a Laser Shotgun".

A Story of Predator Drones, Signal Intelligence, GPS, and 20th Century Human Intervention

2006-03-19T10:35:00.000-05:00

I was reading an article in the April 2006 issue of The Atlantic on the NSA and their purported ability to snoop on all information, everywhere. The piece is a bit of hype given the domestic "wiretapping" (they mean to say signal intelligence gathering, apparently) press of late. One page of the article had an inset on the 2002 predator drone strike in Marib, Yemen that killed Qaed Salim Sinan al-Harethi. Harethi, the article says, was a high priority target. Below is an image of a predator drone like the one in the attack.

The story goes on to say that when a cell phone was used, a monitoring team saw the alert, located the call using GPS, and began listening. The people involved made the decision to attack the target after listening to the speakers on the phone. This time, Harethi's driver was using the phone and only when Harethi began giving directions to his driver did the staffer realize the target was present.

Let's consider this problem. Harethi was using any number of cellphones and cards between cell phones to avoid eavesdropping or tracking. The NSA was able to be alerted to any one of the phones/cards being used and immediately listen to the conversation. The NSA located the cell phone using GPS. The CIA launched an attack from an unmanned Predator drone carrying Hellfire missles. There's much to be said about these technological feats, but there's more to be said and done about the 20th century approaches to the rest of the operation.

A human being, trained to speak in the language of the enemy, had to be listening in real-time in order for this mission to succeed. The human had to listen to a conversation containing at least three people, separate who was saying what and when, and determine that Harethi was one of the speakers. There is no way, today, to get a computer to do these human tasks, but it poses a nice playground for research.

Math For Programmers

2006-03-18T19:09:00.000-05:00

I know this site is supposed to cover issues of Cognition, Robotics, and Learning (as it applies to Artificial Intelligence), but I thought this blog was particularly valuable anyway. Steve Yegge did a nice evaluation of the current state of mathematics teaching in the US and why the modern programmer uses very little of what they learned in school. He proposes an alternative methodology for learning math and suggests everyone try to learn a bit more in their free time.

Graceful robots

2006-03-15T15:32:00.000-05:00

Here's a video of four QRIO robots dancing. They have been programmed beautifully.

Speaking of robots - jabberwacky.com has chatbots that you could chat up. They are kind of slow but worth a try! The conversation could get a bit edgy though - the chatbots are afflicted with severe ADD. :)

More on artificially generated music

2006-03-15T10:30:00.000-05:00

I'd come across GenJam a couple of years ago. It is a software that learns to play jazz solos. The creator of this software is Al Biles As the name indicates it is essentially a Genetic Algorithm.

The website has a few snippets from a jam session between the creator and GenJam. Some of the results are actually very listenable (i.e., if you are "into" jazz music) I recommend "LadyBug" because there are distinct pieces played by the software and the creator and is a good showcase of the software's "talent".

He actually plays regular gigs using that software as a part of his Virtual Quintet.

Learning Elmospeak

2006-03-13T12:28:00.000-05:00

Toys are fun and toys that you can modify or extend (e.g. LEGO) expand everyone's imaginations. After seeing someone's attempt to hack Elmo, I had to try myself.

Casey points toward the innocuous file "temp.inf". In temp.inf we can see a grammar, of sorts, for Elmospeak. For me, I'm interested in what language Elmo speaks (I call it Elmospeak). With stripping out extraneous bits, we're left with a bunch of non-terminals of the form PL_# -> SOUND1 SOUND2 ... SOUNDN, where # is an integer between 1 and 620 and SOUNDN is some terminal associated with a particular sound (generally a word or generic phrase). Also, an [NMSEC] terminal "plays" silence for N milliseconds. Here is an example production:

PL_059 -> ONE [150MSEC] TWO [100MSEC] THREE

Further, there are GAME and STORY non-terminals that expand to any number of non-terminals and terminals. But, this has all been said before. Let's instead consider learning Elmospeak. Overall, Elmospeak is a regular language and regular only because of its perceived finiteness, I think. In my experience with Elmo, the same utterance is heard each time a conversation begins. From there, with some probability, a non-terminal is selected from a subset of non-terminals. The problem is that there doesn't appear to be any nondeterminism in Elmo by default. Right now I'm experimenting with adding nondeterminism to Elmo and seeing just how productive Elmo can be.

Squiggly Lines

2006-03-12T20:38:00.000-05:00

I claim that The Speech Recognition Problem is fundamentally The Vision Problem, and both are really just The Squiggly Line Problem. My hypothesis is that with a few weeks' training. I'll be able to translate spectrographs (no background noise) into their English phrase.

Tim and Bill each have $10 that says that I can't do this. Is anyone else interested in giving me their money (i.e., betting that I can't do this)? Unless your name is Bill or Tom, I'm gonna cap the amount that you give me to $1. Just to warn you, I've trained myself to read Suetterlin handwriting, which *I* once thought was impossible.

Here is a sample spectrogram of me saying "Chicky check, microphone check, chicky check-a, sibilance sibilance.", one of the coolest sounds ever (check out its spectrogram), and my java code that gives you a real-time spectrogram.

(After I lose my hearing from my iPod, I'll be using this software so that I can understand people.)

Generating Music

2006-03-09T22:29:00.000-05:00

At Playing the Market (via /.), music is generated based upon some physical phenomenon (e.g. the stock market) that need not result in classic musical styling, but rather, conform to the physical phenomenon. The goal is not to mimic, but to be artistic. Take a listen to Fibonacci's Random Walk (part 1) which is based on the Fibonacci sequence [ F(n) = F(n-1) + F(n-2) ] augmented with deterministically generated noise for an example of this approach. The result is eminently listenable, but other works on the album lack that quality.

Veering away from data inspired music, we look at music generated by the approach in Grammar Based Music Composition (see the example above). Jon McCormack transforms "string rewriting grammars based on L-Systems into a system for music composition" that seeks to aid in the composition of works. While not using the L-System approach, a sample of work generated from human speech offers a taste of his music.

amazing video game with evolution and AI

2006-03-06T15:08:00.000-05:00

Will Wright, of Sim City fame, has created a new video game named Spore, where players start with a single celled organism and help evolve it over many generations (using a "creature editor") through a sequence of more complex organisms that walk on land, mate to produce offspring, eat other creatures to survive, and so on. Eventually, the creatures form villages, cities, and leave their home planet to meet creatures evolved by other players. There is a fantastic video available where the developer shows off what the game can do. Text on the video's web page says "Will Wright talking at the Game Developer's Conference about 'Spore', which looks like it could possibly be the best video game ever". That may not be hyperbole.

The video contains interesting comments about "procedural verbs", like "bite" and "walk" and how they can be combined to produce "drag" = "bite and walk". It's unclear whether there is significant NPC AI, or whether the creatures are all the result of the work of other players.

A bit of cognition woo-hoo

2006-03-06T10:14:00.000-05:00

Here's a nice bit of cognition related food-for-thought I found on my friend's blog. Before you click on the link below, read the following. (If you don't do so, you'll miss the point!) There are two teams of people, one wearing white tees and other wearing black tees. Each team has a ball that they throw to another member of the same team. Your job is to find out the number of throws made by the white team members.

Did you find anything wierd? Not really? Please go back and watch the video again(trust me!). This time around, just watch the video relaxed.

How on earth did you miss that before? Are we missing out on some eye doctor appointments? :)

Music To My Ears

2006-03-01T21:40:00.000-05:00

According to this story on PR Leap, an Isreali company (could it be the people putting porn on Sprocket?) has announced the release of a new product to stream mp3s to 3G cell phones. "MusicGenome, the leading expert in Artificial Intelligence (AI) and Music Cognition, announced today the release of Smart DJ – a new innovative fun product for the global cellular market. Smart DJ offers 3G mobile phone users the ultimate in digital music experience, by creating a new online entertainment service that accurately analyzes listeners’ music preferences and then delivers a stream of mp3 songs totally in-sync with the listener’s taste with 80% success rate." It does not discuss the specifics of the technology employed, but it does claim to be better than products that employ Collaborative Filtering. The company's website does not appear to be compatible with Firefox for Mac OS X.

Speech Accent Archive

2006-03-01T21:33:00.000-05:00

I just happened across an interesting data source: the George Mason Speech Accent Archive. Accent and dialect classification can improve speech recognition approaches and automatically labeling native tongues would make our friends by the airport happy. The archive, as of 1 March 2006, contains 502 recordings of individuals reciting the same paragraph of English text. Here's an excerpt:

"Ask her to bring these things with her from the store: Six spoons of fresh snow peas, five thick slabs of blue cheese, and maybe a snack for her brother Bob."

One of my favorites is a recording made by a native Wolof speaker from Senegal. At UMBC, we offer language courses in Wolof that I am considering auditing.

Watch out for that port-scanner half a planet away...

2006-03-01T14:08:00.002-05:00

It was a cold Wednesday morning. The inhabitants of Coral and Maple were busy with their daily bustle of intense research activity. Dr. Marie desJardins walks in to the lab and says "Could someone shut this window on sprocket?" Curious we assemble around our good ol'server (sprocket). Lo and behold! There is a Safari window open with a fairly graphic picture from a porn website in Israel open!

A whirl of activity ensues. Disturbing questions are asked - Who did this? How to prevent it? How to track this down? Was it just a prank or a malicious attack? Are Mac servers (gulp! gulp!) vulnerable? (Incidentally, slashdot carried this post on two days back)

The obvious suggestions are all made and shot down in turn. Concerned lookers-on (self included) throw in their tuppence-worth nonetheless - check Safari's history, check network usage, check VNC's logs... After a frantic morning of grepping through logs and other thingmajigs, our sys-admins track the attack down to an IP address in Israel. They confirmed it wasn't malicious. (Why would someone want to remote login to a computer and watch porn on it?)

Moral of the story: Logout of your computer when you leave.

Towers of Hanoi with Onion Rings

2006-02-28T21:23:00.000-05:00

The Coral and Maple labs will occasionally go out to lunch together. Not that long ago, we all ate at Red Robin where they have ketchup smotherable onion rings that we ordered. Each onion on the skewer gets progressively larger and, as we are computer scientists, we immediately thought about the Towers of Hanoi. Our server kindly gave us two additional skewers and after a few odd looks, we were solving the problem.

We need to reinact the solution as we were all hungry and wanted to eat more than perform recursive moves. Keep an eye peeled for our next computer science canonical example in food.

iPod Hearing Loss

2006-02-28T19:46:00.000-05:00

This semester at UMBC, the CSEE Department is offering a class called "Topics in Signal Processing: Digital Signal Processing for Speech and Hearing Disorders." Being that some of the research in the Coral Lab centers on language learning (involving speech processing) and that body of work, in general, relies on noise-free input, this class has made me consider not only a noisy environment, but also noise that can be introduced during language comprehension and processing. Ironically, in the course, we study technologies and techniques that can restore or improve hearing (e.g. cochlear implants) using some of the same engineering that can cause hearing loss (e.g. abuse of standard speakers).

On a college campus, it's a rare day to not see those little white bud headphones presumably attached to one of the myriad iPod models. Recently a lawsuit was filed against Apple Computer, Inc. asserting the device does not adequate limit sound intensity and can lead to hearing loss. The best I could find on technical iPod specifications and advice was that the frequency response is between 20Hz and 20,000Hz and that unknown experts are warning about the dangers of prolonged, high intensity iPod listening. There is more to come...

Rock, Paper, Scissors

2006-02-23T20:57:00.000-05:00

Dated -- 25 January 2005

A post via email to the CORAL Mailing List:

According to a post on Slashdot, "New Scientist is reporting that UK
researchers have created a computer (story) that can learn rock, paper, scissors (research site) by observing humans. CogVis uses visual information to recognize events and objects in addition to learning by observing."