A Well, With Two Buckets

May 15

From Breitkreutz et al. “Phenotypic and Transcriptional Plasticity Directed by a Mitogen-Activated Protein Kinase Network.” Genetics 165:997-1015 (November 2003)

"More artificial Finnish" (The Universe of Discourse)

The first part of this (now two-part) series of posts was mildly amusing — he was using Markov models to generate “Finnish-looking” nonsense text. Pretty straightforward. But in the second installment, things get interesting. Now he’s told, by some Finnish-speaking readers, about “vowel harmony.” There are two groups of vowels in Finnish which only ever appear in words with vowels of the same group — no vowels from different groups ever appear in the same word. (The vowels form a bipartite graph, and words only ever contain independent subsets of vowels from that graph, etc.) Dominus notes, obviously, that a higher-order Markov model would start to catch this — but you’d always be foiled when the word length exceeded the order of the model. So what to do? Are there ways of generating words, from a Markov (or similar) model, which are guaranteed not to have vowel disharmony? It’s a logical constraint “laid over” the underlying probabilistic model. It sounds like something that someone would have studied at one point, and also like something that could be applicable in other areas. Say: biological sequence analysis, which we might expect to be a combination of low-level “probabilistic properties,” but under the constraint of high-level “logical” restrictions.

“(I did once write to Jesse Sheidlower, an editor of the Big Dictionary, to ask his advice about whether “ballless” should be hyphenated.)” “More artificial Finnish” (The Universe of Discourse) And what was the answer??

May 6

“So just because a series converges, another one obtained by removing parentheses may not converge.” “Associativity in Series I” (The Unapologetic Mathematician) (An obvious point, perhaps. But I’ll confess that I’m really enjoying this series of John Armstrong posts that are essentially reviewing the basics of high-school calculus — admittedly from a slightly more advanced [college?] viewpoint.)

“One such approach, not mentioned by the authors, is based upon efficient or straightest possible convergence to the truth, where straightness is a matter of minimizing the number of times the method retracts earlier conclusions as well as the times (sample sizes) at which these retractions occur, an extension of ideas proposed by Hilary Putnam (1956). It can be shown that, given that convergence to the truth is feasible at all, the efficiently reliable causal discovery strategies are exactly the strategies that follow a version of Ockham’s razor at each stage of inquiry (Kelly 2007).” Kelly and Mayo-Wilson’s review of Harman and Kulkarni’s “Reliable Reasoning: Induction and Statistical Learning Theory”

May 2

“One more problem, which has been raised explicitly by Hajek in discussions and presentations of this paradox, seems to be solved by this theory. He has suggested that part of what makes the Pasadena game so troubling is that in addition to having no expectation itself, any action that has a shot at the Pasadena game as a possible outcome with non-zero probability will end up having no expectation. Thus, if (as some suggest) a rational agent shouldn’t assign probability zero to any epistemic possibility, then the “poison” of the Pasadena game will spread to every decision problem. When trying to decide between Indian and Chinese food for dinner, one can’t rule out the possibility that one will be forced to play Pasadena in either case, so one can’t calculate expected utilities, and classical decision theory will go silent.” Easwaran, “Dominance-Based Decision Theory.” (Both de Finetti and Savage have this construction, wherein events can be infinitely sub-divided. They usually do this by “appending a coin flip” in much the same was as Hajek/Easwaran suggest “appending” the Pasadena game. So is *this*, then, the source of (at least some) objections about “large” and “small” worlds in decision/utility theory?)

May 1

Easwaran, "Dominance-Based Decision Theory"

I get excited every time I find a new Easwaran monograph. I suppose I could just go and find a list of his publications, but that’d be too easy. It’s more fun to just stumble over them, and then realize how excellent they can be, all over again.

Apr 28

"I love Python" (Reasonable Deviations)

“Flat is better than nested?”

In the comments, someone suggests Ruby. Python, Ruby… I think we all know what the next step is going to be.

Apr 24

Putting the CART before the Horse-Race

Best. Post. Title. Evar.

Apr 17

“Roughly speaking, the strategy is to view the permanent of a n x n Bernoulli matrix recursively as a random signed combination of the permanent of its n-1 x n-1 minors formed from its first n-1 rows, which one can view as the “parents” of the original matrix. The idea is then to take a sort of “genetic” or “evolutionary” viewpoint, and ask how likely the trait of having a large permanent is of being passed from parents to children. The key point is that this trait is “dominant” rather than “recessive”: as long as just one of the parents has a large permanent, it is likely that the child will have large permanent as well (and if many of the parents have large permanent, the child is likely to have an even larger permanent).” On the permanent of a random Bernoulli matrix (What’s new)

A Well, With Two Buckets

“Once a certain degree of insight has been reached,” said Wylie, “all men talk, when talk they must, the same tripe.”

"More artificial Finnish" (The Universe of Discourse)

Easwaran, "Dominance-Based Decision Theory"

"I love Python" (Reasonable Deviations)

Putting the CART before the Horse-Race