Also reading this morning -- unforch, the refs in their PDF are messed up, and their source-code is missing some import, so I can't recompile it (from .tex source) unaided. GRrrrr.

Alan Willsky.

Oooh, a new notebook to read.

"The AI revolution basically came first to chess!" -- a thousand times, 'yes'! Also, point #5 could probably be expanded into an entire book on its own.

"We consider linear models for stochastic dynamics. To any such model can be associated a network (namely a directed graph) describing which degrees of freedom interact under the dynamics. We tackle the problem of learning such a network from observation of the system trajectory over a time interval $T$."

Yes, *that* David Mumford. Oddly enough, this reminds me how I met Mumford's niece at a soccer game this summer.

Gabe's new project.

More Mayo that I must own.

"So that gives us around 90k bits, which should require around 180k seconds of optimal memorization...! This would involve learning about two new vocabulary words per minute, which seems like a plausible rate. ... 50 hours is a rather remarkably small number. ... This is at least one order of magnitude better than commonly-observed performance in foreign-language learning. Why might this be? One possible explanation is that people are usually learning not only the vocabulary of the language, but also its alphabet, orthography, phonology, morphology, syntax, and pragmatics at the same time. A second ... is that typical vocabulary memorization is very badly structured..." -- and the second one's the one you go with? I'm about 50% convinced that this is an extended joke, in which case, bravo.

Via Andrew Gelman ( http://www.stat.columbia.edu/~cook/movabletype/archives/2010/01/what_can_search.html  ) 

I *really* need to buy Jonathan Wilson's book. And there's so much in this piece to really chew on, even if you're not into soccer (ahem, football) at all. Sports-as-evolution, culture, learning, etc etc etc.

"Children learn reading in a stepwise process: first, awareness that words are made up of phonemes or speech sounds (ba, da); then the discovery that there's a correspondence between these speech sounds and pairs or groups of letters. Later the child begins to recognize entire words, and after a few years, reading speed becomes independent of word length. Dehaene deplores the whole-language approach to teaching reading in which beginning readers are presented with entire words or phrases in the hope of fostering earlier comprehension of text. He cites research showing that children who first learn which sounds are represented by which letters, and how pairs or groups of letters correspond to speech sounds, make steadier progress and achieve better reading scores than those taught using the whole-language method." -- Must buy this book.

To be beaten with a copy of Orality and Literacy after a fair trial.

A history of soccer formations, phrased as the answer to the question of why Fabio Capello is a better coach for England than some people seem to want to believe. This is excellent, both from a sports perspective, and also from the perspective of a history of cultural and social evolution -- formations, and changes in them, are "how people learned to play the game."

"The sleep-dependent observational motor learning enhancement is at least similar to that previously reported for implicit and declarative memory. The apparent prerequisite of observing real movements indicates that subjects transfer experience obtained through observation of movements to subsequent self-initiated movements, in the absence of practice. Moreover, the consolidation of this transfer requires an early sleep window." -- Makes me think I should be taking more naps.

Papers on the multipole method and the fast gauss transform.

Tacit Knowledge #3. (via Mark Thoma's blog: http://economistsview.typepad.com/economistsview/2009/09/tacit-knowledge.html)

Tacit Knowledge #2.

Tacit Knowledge #1. Saving this so it's no longer sitting in the Chrome tabs.

Collected for an example in some thoughts about "tacit knowledge," culture, rules, and learning. But also because I want to make better coffee, and I desire a Chemex.

The original Temporal Differences learning paper, if I'm not mistaken? Time to re-read. (Alex, Bob, if you see this: ideas for a free-food-predictor start here, perhaps?)

[Bull. Amer. Math. Soc. 46 (2009) 255-308]

Statistical haptics.

Shaun, when you have a moment in the next week, do you want to^H^H^H^H^H^H would you be willing to talk about this?

"This paper considers the prospects for constructing a neoclassical theory of growth and international trade that is consistent with some of the main features of economic development. Three models are considered and compared to evidence: a model emphasizing physical capital accumulation and technological change, a model emphasizing human capital accumulation through schooling, and a model emphasizing specialized human capital accumulation through learning-by-doing." --- Doesn't look like *exactly* what I'm looking for, but it's a start.

"If the largest cities are able to churn over constant ptototypes it's because the abundance of density yields disproportionately large opportunities in the form of financing, know how, and other limited resources. The rural, on the other hand, typically has ample supplies of raw material and time. The rural ethos is to assemble what you have in the best way that you can and this kind of improvisation is is what was missing from Paso Robles. As a place that now thinks of itself as a city, Paso Robles looked to other, larger cities for its missing expertise and equipment rather than taking the imperative of the event to test something new. This opportunity for civitas was treated as a chance to consume." -- There's a lot more to say about *this* in particular. I think the idea of "cities-that-consume-or-produce" is very similar to "cultures-that-learn." These are attributes we normally think of ascribing to people; which other characteristics can be ascribed to corporate entities?

A basic review of naive Bayes modeling, bag of words representations, and all that.

"At the risk of sounding self-contradictory, I also think that when the student owns the task of learning, they will learn enormously better than when the task is imposed. This does not imply that the student discover everything, however. It merely implies that they need to discover the need for the things that they learn. Students who need to learn something can learn from almost any source, even from something as currently unfashionable as, say, sitting quietly through traditional lecture."

"The Protocol Informatics project is a software framework that allows for advanced sequence and protocol stream analysis by utilizing bioinformatics algorithms. The sole purpose of this software is to identify protocol fields in unknown or poorly documented network protocol formats. The algorithms that are used perform comparative analysis on a series of samples to better understand the underlying structure of the otherwise random-looking data." --- So... Smith-Waterman, but used to analyze network protocol data? That's ... neat. Sort of like that genetic-programming-for-file-parsing stuff from a week or two ago.

"For some time now, I’ve been experimenting with algorithms for solving proportional analogies." Haven't had enough time to read the underlying paper yet (apparently, my family and girlfriend are not cool with me reading papers at the dinner table on Christmas Eve? who knew...), but this looks totally sweet. Time to add Apperceptual to the RSS reader (I'm not sure why I hadn't done that already).

"Overall, the researchers found that S.R.D. could complete nearly all of the tasks just as accurately as the others, although anecdotally it took her about 5 to 10 seconds longer than the people who had never experienced blindness. There were two exceptions. She wasn't quite as good at recognizing faces (though she still was more than 75 percent accurate), and she had a peculiar difficulty in judging gaze direction."

I (honestly) don't find this quote damning at all. This is exactly how being a graduate student generally works (at least, around here): someone sits you down and says, "you are now our expert in implementing dynamic programming methods for calling segments in microarray data." Or, "you are now our expert in motif discovery methods that take sequence conservation into account." Or, whatever. And you totally aren't an expert, at first. But by the time you graduate, you know a lot more than you did when you started. This anecdote just makes me think that Jane Jacobs had a really poor understanding of Learning and its role in Culture -- which I don't think she did. So something else must be wrong.

The course notes to Sanjay Mahajan's IAP course.

"The Marine Corps teaches you how to be miserable. This is invaluable for an artist." Reading this, I was thinking, "this must be the function of graduate school, too." And then I got to the end and, yes, there is the requisite mention of graduate school.

The original paper behind the "gut instinct" and mathematics article in the NYT.

The Log mentions the "gut instinct & mathematics" article in the NYT, along with a link to the original paper.

Quoting (and disagreeing with) Harvey Mansfield on grade inflation. "To know it would require a large year-by-year database of the actual work done by Harvard students (including, presumably, evidence of their classroom participation), and the matching grades. I’d be very surprised if the administrators have such a database, access to which they are jealously guarding. Mansfield ought to know: if one existed he would have been asked to contribute to it. But he does not even seem aware that that is what would be needed." --- But there's something else going on here, too. There are at least three or four different reasons why grade inflation could be bad, right?

Papers from the International Joint Conference on AI. Including several papers in the "Learning" section that I never got around to reading...

"Here, we present a computational model that learns structures of many different forms and that discovers which form is best for a given dataset. The model makes probabilistic inferences over a space of graph grammars representing trees, linear orders, multidimensional spaces, rings, dominance hierarchies, cliques, and other forms and successfully discovers the underlying structure of a variety of physical, biological, and social domains." To read. If it were someone other than Josh Tenenbaum in the author list, I'd be less interested...

An excerpt of a description of Kenneth Pike eliciting information about a new language from a native speaker without a translator -- "monolingual elicitation." Benzon at the Valve links to this, asking "what did Pike know that Quine didn't?" But that seems like a complete misreading of Quine, who's describing how the narrowing process (I think he terms this, asking "ostensive" questions) narrows down some intermediate translation without ever permanently settling the question. Would someone like Pike ever really dispute that? Anyway, I think language-learning-games like this are probably a great lab for thinking about science too -- "nature" as the native speaker.

"We also sketch how a regret-based algorithm can be used as an alternative to Bayesian averaging in the context of inferring latent random variables. "

Ordering and notation in high-school algebra. See, what I take away from this is that it must be *very hard* to design algebra tests correctly (to actually test what you want to test). Test-writers probably don't even recognize their own mistakes.

"I suspect his motivation was not so much to understand the world as it was to find new ideas to explain. The act of discovery was not complete for him until he had taught it to someone else. "

"Whether kids played board games correlated with number line success, but playing video games did not."

"...it is a burning pity that our lives are not long enough and not sufficiently free of annoying obstacles, to study all things with the same care and depth as the one we now devote to some favorite subject or period." Nabokov on scholarship.

Just when I think I've managed to worm my way out from under the dsquared cult... he pulls me back in. It's a great post.

PsycARTICLES - American Psychologist - Vol 34 Iss 7 Page 571. Gah, why won't MIT give me access to this? Time to track it down in the library. (Via Andrew Gelman).

To read. (Valiant, but no Vapnik?)