This is a step towards an ongo­ing effort to Clean My Desk. And also to cut back on the friv­o­lous increase in the world’s heat I gen­er­ate by indis­crim­i­nately pick­ing books up at the library and haul­ing them back here (only to be told they’re over­due with­out being read). I’m not review­ing; rather, not­ing inter­est­ing arti­cles in the books.

I’ve already had some words, months back, about the ridicu­lous prices charged by cer­tain North­ern Euro­pean Pub­lish­ing Clans, and I don’t want you to con­sider buy­ing these books for an instant. The links to the left are offered more in the spirit of a pub­lic shame through lit­eral fact, not an appro­ba­tion. You could go to Ama­zon and buy some other crap—some use­ful crap, if you want. I wouldn’t mind that. Because Ama­zon would pay me, so I would be less mind­ing that, for the pay­ing. I am not at all about mind­ing paying.

But in gen­eral these are books that are for the bib­li­o­graphic padding of the con­trib­u­tors, not to be read, not even to be ref­er­enced phys­i­cally, but merely to be socked away on some shelf in a base­ment High Den­sity Stor­age, and to see the light of day only when some idiot (like me) gets a dose of undi­rected curiosity.

So: Expen­sive aca­d­e­mic press van­ity doorstops = dumb. But: Authors in said books = some­times very inter­est­ing. To kill π birds with one stone, I’ll call out some of the arti­cles and chap­ters and equa­tions that catch my eye, and briefly dis­cuss them. And offer links to free preprints online, as available.

I’m start­ing with the fat­test. Clear that space off quick.

[I’ll add the tele­graphic notices as I have time today. These books are due, after all.]

AI 2004: Advances in Arti­fi­cial Intel­li­gence is one of those catch-​​all pro­ceed­ings vol­umes that is full of this and that. As local Spe­cial­ist in This and That, I like it. Not all. Here are some con­tri­bu­tions that at least caught my eye:

• “Crit­i­cal dam­age report­ing in intel­li­gent sen­sor net­works” by Jiaming Li, Ying Guo, and Geoff Poul­ton. [not avail­able online!?] Wrap a space­craft in a “skin” of locally-​​connected sen­sor agents. When a lit­tle meteor or a way­ward space bolt strikes it, they yell at each other. How do you arrange them so that the col­lec­tive net­work struc­ture can under­stand (and com­mu­ni­cate) the dif­fer­ence between ran­dom fail­ure, minor dam­age and crit­i­cal dam­age? Espe­cially when you don’t know where the dam­age will be, and if it will affect the cru­cial “por­tal” com­mu­ni­ca­tor agents. You evolve a pheremone-​​directed sig­nal­ing route on the fly.
• “Com­bin­ing Bayesian net­works, k near­est neigh­bours algo­rithm and attribute selec­tion for gene expres­sion data analy­sis”, B. Sierra, E. Lazkano, J. M. Martínez-​​Otzeta, and A. Asti­gar­raga. [also not online!? sheesh.] Biol­ogy used to be so sim­ple, so ele­gant, so obser­va­tional. Now it’s bur­dened with data lack­ing knowl­edge, and all those years of com­plaint that “Math is hard; let’s do biol­ogy!” have wrought a fear­some slack, being taken up by folks in other dis­ci­plines. Like these. The prob­lem here: Gene expres­sion chips (AffyMetrix and oth­ers) result in thou­sands of data points for every exper­i­ment. Each of those 2000+ num­bers is (arguably) the expres­sion level of a cer­tain RNA species in vivo. How do you take a 2000–dimen­sional time­series, and recon­struct a genetic reg­u­la­tory net­work from it? The authors’ response (roughly) is an iter­a­tive variable-​​selection and learn­ing cycle: iden­tify a small set of salient (influ­en­tial explana­tory) genes from the mess, and add them to a data­base; build naive Bayes mod­els of the data­based gene dynam­ics using the com­plete dataset, in order to iden­tify new genes to add to the mix. Iterate.