Yeah, well, that was 3 weeks ago.

Today, Radeon’s new HD 4870 X2 graphics card has 1600 stream processors that crunch 2.4 Tera FLOPS.

Name: IBM Blue Waters
Peak performance: 10¹⁶ FLOPS
RAM: 620 TB
Location: University of Illinois
Delivery: 2011

For every synapse in an adult human brain this machine will have around 6 bytes of RAM and 100 floating point calculations per second.

First of all, I do latex -> dvips -> ps2pdf. Old school. If you’re a pdflatex person, I’ve heard that this works pretty well with lulu, but I can’t say from experience. You can figure out for yourself which parts of the following are relevant to you. Also I’m doing all this with Linux and Linux command line tools.

Page formatting
The first problem is to get the page size and formatting correctly set up. I went for lulu’s 9″x6″ case wrap hard cover. Essentially, that’s a “normal” sized book format, with a full colour hard cover with no dust jacket. I’m sure it’s possible to get one of the normal LaTeX document types to format the pages to some exact page size, but I never worked out how. I only seemed to be able to select preset sizes, none of which were 9″x6″. So I took a different route: I switched to the scrbook format which is part of KOMA-script. I was a bit hesitant about doing this to start with. Firstly, because I imagined it might be a pain to install, and secondly I was sure that it would conflict with a million little tricks I was doing in my document. Turned out to be plain sailing. I put “KOMA-script” into synaptic on my Ubuntu Linux machine and found that it was in the intexlive-latex-recommeded package and that I already had it installed.

Ok, so now we can switch to scrbook and set up the page size correctly for lulu:

\documentclass[10pt,twoside,openright]{scrbook}

\setlength{\paperwidth}{6in} % set size for latex
\setlength{\paperheight}{9in}
\special{papersize=6in,9in} % set size for ghostscript
\typearea[6mm]{1} % 6mm for spine

Amazingly, this change did not cause me any real problems. I didn’t like the chapter and section heading fonts to start with, but I actually grew to like them and kept them. If you want to mess around with fonts google for “The KOMA-script package” for a pdf on how all the options work for this document class.

I thought that the 6mm for the spine wouldn’t be enough. However, I decided to go with it to start with as somebody on the internet had suggested this, and yes indeed it was perfect in the printed copy. I also thought that the printed area of the page was slightly too high. On the printed book it still looks slightly too high to me, however I then compared it to a professional text book of the same page size and there the page spacing was identical. So, yeah, maybe it’s just me.

Fonts
The next major problem is sorting out the fonts. Just because your document prints fine on your printer does not mean that lulu’s book printer will also print it correctly. What you want is for all the fonts to be fully embedded into the document. There are a few tricks to make this happen. First up build your dvi file as per normal:

latex thesis.tex

Ok, now build the ps file using the pdf and G0 options. I’m not sure if the G0 option is really necessary, but I’ve never found it to cause any problem either:

dvips -Ppdf -G0 thesis.dvi -o

Now we get to the key step: forcing all the fonts to fully embed into the pdf:

ps2pdf13 -dEmbedAllFonts=true -dSubsetFonts=false -dPDFSETTINGS=/printer thesis.ps

(that’s one line of course) I used ps2pdf13 to force the output to version 1.3 = Acrobat 4.0 format. I heard that this tends to work best with lulu rather than newer formats. In any case it didn’t cause me any problems.

The first two options are obvious, embed all the fonts and don’t just subset embed them. Strangely, some fonts still won’t be embedded! The reason is that very common fonts are assumed to be present on all printers and so get skipped. Lulu doesn’t like this and demands these to be embedded also. With the third option to prepare for printing the remaining fonts will also be embedded. Problem fixed.

Ok, now let’s check that the fonts really are fully embedded:

pdffonts thesis.pdf

Under “emb” should all be “yes” and under “sub” should all be “no”.  Great, all your fonts are now lulu compliant. You can use any weird fonts you want with lulu, so long as you embed them. Crazy ass custom math symbols… no problem!

Now, let’s check that everything looks right. I suggest installing and using acrobat reader for this. It’s free and it’s the definitive viewer for pdf correctness.

acroread thesis.pdf

Now turn off the local fonts to ensure that you’re only viewing the document with the embedded ones. In my version of reader through the menu bar it’s:

edit->preferences->page display->use local fonts = false

At this point you might want to spend some time flicking through the pages of your document checking that things look the way you expect them too. Time well spent.

Page numbering
Now, let’s check that the pages are in the right places. Set to two page view with a cover page:

view->page display->show cover page during two up = true

view->page display->two up

At the start of your document you should see a single page on the screen. This will be your title page. There will be a blank page before this from the printing process, like in any normal book. The title page will be a right side page of course. Page down and you’ll now see two pages: the left page on the left of the screen and the right page on the right. This left page normally contains a copyright notice etc. Now keep flicking down through your document. The thing to check is that your left pages have the page number on the top left (or bottom left) and the right pages have the page number on the right. In other words, make sure that the left pages are really on the left side and the right pages are really on the right side. If not, you’ve got your page numbering messed up. Odd numbered pages go on the right, even on the left.

If your document is all looking good, then go ahead and upload the pdf to lulu.com. With all the fonts fully embedded lulu shouldn’t complain.

Cover
The final problem is to make up a cover design. I used gimp to produce a single image for both the back and front cover. Follow the lulu instructions to get the exact dimensions you’ll need in terms of “points”. You’ll need to write all these numbers down and do some arithmetic to figure out exactly where everything goes. You’ll have a spine width that depends on how many pages you have in the book, a bleed area where the printed colour cover wraps around the outside into the inside of the book etc. One thing to remember when designing the image in gimp is that when you first create a new canvas, set the size to “points” in the dimensions instructed by lulu, and also remember to set the dpi to 300 rather than 72 in order to get a good image quality. If I recall correctly lulu prints covers at 300 dpi.

The only part I found tricky in this process was the centring of things on the front page (which is on the right of the cover image).  The problem is that you have the edge of the spine, then you have a folding area, then you have the hard flat area of the front cover.  Where is the centre of this? From the spine to the right edge of the cover? Or from the edge of the folding area to the right edge? To start with I took the former and placed things half way between the edge of the spine and the right edge of the cover. However, when I got the first test printing of my book, while this is where things were printed, visually it appeared that things were a little off the to left due to the folding area. Thus, in my opinion, the visual centre of the cover is slightly to the right due to the presence of the folding region on the left next to the spine. My suggestion then is to shift things fractionally to the right. Just 1/16 of the cover width, which if you measure it out is really small, but still noticeable. If you do follow this suggestion, do measure it out as 1/8 the page width to the right would be too far.

Now, save the cover as a png and conver to a pdf using the image magick software (again in the Ubunut depositories if you haven’t got it installed):

convert cover.png cover.pdf

Check again with acroread and then upload the cover pdf to to lulu.

Finally, print a test copy to check everything is working correctly before ordering a box of copies!

My thesis is now available at lulu.com.  As promised, it’s at cost, which works out at $18 plus shipping.  It’s all under a creative commons licence and in a few months I’ll put the pdf online for free.  I’ll also write a post shortly on all the tricks involved in publishing on lulu.com with LaTeX, in case you plan on doing something similar.

Table of Contents

Preface
Thesis outline
Prerequisite knowledge
Acknowledgements

1 Nature and Measurement of Intelligence
1.1 Theories of intelligence
1.2 Definitions of human intelligence
1.3 Definitions of machine intelligence
1.4 Intelligence testing
1.5 Human intelligence tests
1.6 Animal intelligence tests
1.7 Machine intelligence tests
1.8 Conclusion

2 Universal Artificial Intelligence
2.1 Inductive inference
2.2 Bayes’ rule
2.3 Binary sequence prediction
2.4 Solomonoff’s prior and Kolmogorov complexity
2.5 Solomonoff-Levin prior
2.6 Universal inference
2.7 Solomonoff induction
2.8 Agent-environment model
2.9 Optimal informed agents
2.10 Universal AIXI agent

3 Taxonomy of Environments
3.1 Passive environments
3.2 Active environments
3.3 Some common problem classes
3.4 Ergodic MDPs
3.5 Environments that admit self-optimising agents
3.6 Conclusion

4 Universal Intelligence Measure
4.1 A formal definition of machine intelligence
4.2 Universal intelligence of various agents
4.3 Properties of universal intelligence
4.4 Response to common criticisms
4.5 Conclusion

5 Limits of Computational Agents
5.1 Preliminaries
5.2 Prediction of computable sequences
5.3 Prediction of simple computable sequences
5.4 Complexity of prediction
5.5 Hard to predict sequences
5.6 The limits of mathematical analysis
5.7 Conclusion

6 Temporal Difference Updating without a Learning Rate
6.1 Temporal difference learning
6.2 Derivation
6.3 Estimating a small Markov process
6.4 A larger Markov process
6.5 Random Markov process
6.6 Non-stationary Markov process
6.7 Windy Gridworld
6.8 Conclusion

7 Discussion
7.1 Are super intelligent machines possible?
7.2 How could intelligent machines be developed?
7.3 Is building intelligent machines a good idea?

Appendix
A Notation and Conventions
B Ergodic MDPs admit self-optimising agents
B.1 Basic definitions
B.2 Analysis of stationary Markov chains
B.3 An optimal stationary policy
B.4 Convergence of expected average value
C Definitions of Intelligence
C.1 Collective definitions
C.2 Psychologist definitions
C.3 AI researcher definitions

Bibliography
Index

]]> http://www.vetta.org/2008/07/machine-super-intelligence/feed/ http://www.vetta.org/2008/06/kolmogorov-solomonoff-aixi-etc-questions/ http://www.vetta.org/2008/06/kolmogorov-solomonoff-aixi-etc-questions/#comments Wed, 25 Jun 2008 16:14:45 +0000 Shane Legg http://www.vetta.org/?p=65 Many people seem to have questions about Kolmogorov complexity, Solomonoff induction, algorithmic probability theory, AIXI, the universal intelligence measure and so on.  I don’t always have time to watch all the email lists where these things get discussed, but if you do have any questions, concerns, etc. that you’d like to put to me, feel free to post a question below and I’ll try to answer it.

Two years later…

This PC has 3 nVidia GTX 280 graphics cards, costs about $3,000 and is rated at 3 T FLOPS.

The award is in “recognition of [my] efforts to improve AI theory” and is worth CAD $10,000.  This will certainly help my budget over the next two years while I study at the Gatsby Unit in London.  So, thank you to SIAI Canada, and to all the Canadians whose donations made this money available!

Speaking of my research, after a long weekend of final edits, corrections, formatting, indexing, embedding fonts and other complexity (I’ll write a blog post about what I had to do at some point), I’ve finally uploaded my thesis “Machine Super Intelligence” to lulu.com and have ordered a test copy.  Once I’ve checked that everything is ok I’ll let you know where copies can be ordered.  Copies should be USD $18 plus shipping for a 200 page casewrap hardcover.  Probably about in a month…

And in other news, I’ve been awarded a grant to study machine learning and theoretical neuroscience at the Gatsby Unit, University College London.  I should start that in January and be there for two years.  For the meantime I’ll remain at the Swiss Finance Institute working on portfolio choice models.

The next generation of Nvidia GPUs will support enhancements such as double precision floating point in order to make them more suitable for general purpose highly parallel computation. There will also be cards with no graphics interface and greater maximum RAM designed specifically for low cost supercomputing applications.

Next milestone…. 10¹⁶ FLOPS.  Perhaps in 2012?