“So, let’s get down to the nitty gritty. If con­sumer debt was $13.8 tril­lion at the end of 2008 and the banks have since writ­ten off 5.66% of that debt, total write-​​offs were $800 bil­lion. If total con­sumer debt now sits at $13.5 tril­lion, then con­sumers have actu­ally taken on $500 bil­lion of addi­tional debt since the end of 2008. The con­sumer hasn’t cut back at all. They are still spend­ing and bor­row­ing. It is beyond my com­pre­hen­sion that no one on CNBC or in the other main­stream media can do sim­ple math to fig­ure out that the delever­ag­ing story is just a Big Lie.”

“I strongly sus­pect these pat­terns are dri­ven mostly by cus­tomers, i.e., that more accu­rate pro­fes­sion­als would be less suc­cess­ful in inspir­ing con­fi­dence by oth­ers in them. If you are a suc­cess­ful pro­fes­sional, that is prob­a­bly in part because of your unjus­ti­fied arrogance.”

“We all have pre­con­ceived ideas about what hum­ming­birds’ lives are like, but so much of their world is imper­cep­ti­ble to the human eye. Film­maker Ann Prum describes the break­through sci­ence and lat­est tech­nolo­gies that allowed her and the crew to reveal incred­i­ble new insights about these aer­ial athletes.”

“The remain­ing charts com­pare mar­ket per­for­mance since 2000 with the equiv­a­lent elapsed time fol­low­ing the peak in 1929. As the final chart shows, the cur­rent real total return over the past decade is worse than the per­for­mance over the equiv­a­lent time­frame dur­ing the Great Depression.”

“Here is the trailer for Inside Job, Charles Ferguson’s upcom­ing doc­u­men­tary about the finan­cial cri­sis of 2008. Looks like inter­est­ing and well-​​done stuff.”

“I think every body who’s breathed the air around eco nom ics gets the the sis that money is an eco nomic prod uct sub ject to sup ply and demand like any other. But to actu ally see it bro ken down as analy sis of dis crete things — a fiat cur rency backed by the full faith and credit of the US gov’t but whose weight and mate ri als and cost and dura bil ity and shape all turn out to be cru cial to its suc cess or fail ure — man, it’s another thing altogether. ”

“We under­stand the dynam­ics of the world around us as by asso­ci­at­ing pairs of events, where one event has some influ­ence on the other. These pairs of events can be aggre­gated into a web of mem­o­ries rep­re­sent­ing our under­stand­ing of an episode of his­tory. The events and the asso­ci­a­tions between them need not be directly experienced-​​they can also be acquired by com­mu­ni­ca­tion. In this paper we take a net­work approach to study the dynam­ics of mem­o­ries of his­tory. First we inves­ti­gate the net­work struc­ture of a data set con­sist­ing of reported events by sev­eral indi­vid­u­als and how asso­ci­a­tions con­nect them. We focus our mea­sure­ment on degree dis­tri­b­u­tions, degree cor­re­la­tions, cycles (which rep­re­sent incon­sis­ten­cies as they would break the time order­ing) and com­mu­nity structure.…”

“he motil­ity of the worm nema­tode \textit{Caenorhabditis ele­gans} is inves­ti­gated in shal­low, wet gran­u­lar media as a func­tion of par­ti­cle size dis­per­sity and area den­sity ($\phi$). Sur­pris­ingly, we find that the nematode’s propul­sion speed is enhanced by the pres­ence of par­ti­cles in a fluid and is nearly inde­pen­dent of area den­sity. The undu­la­tion speed, often used to dif­fer­en­ti­ate loco­mo­tion gaits, is sig­nif­i­cantly affected by par­ti­cle size dis­per­sity for area den­si­ties above $\phi \geq 0.55$, and is char­ac­ter­ized by a change in the nematode’s wave­form from swim­ming to crawl­ing in dense poly­dis­perse media \textit{only}. This change high­lights the organism’s adapt­abil­ity to sub­tle dif­fer­ences in local struc­ture between monodis­perse and poly­dis­perse media.”

“A short sur­vey is pro­vided about our recent explo­rations of the young topic of noise-​​based logic. After out­lin­ing the moti­va­tion behind noise-​​based com­pu­ta­tion schemes, we present a short sum­mary of our ongo­ing efforts in the intro­duc­tion, devel­op­ment and design of sev­eral noise-​​based deter­min­is­tic mul­ti­val­ued logic schemes and ele­ments. In par­tic­u­lar, we describe clas­si­cal, instan­ta­neous, con­tin­uum, spike and random-​​telegraph-​​signal based schemes with appli­ca­tions such as cir­cuits that emu­late the brain’s func­tion­ing and string ver­i­fi­ca­tion via a slow com­mu­ni­ca­tion channel.”

“We con­sider prob­lems of Bayesian infer­ence for a spa­tial epi­demic on a graph, where the final state of the epi­demic cor­re­sponds to bond per­co­la­tion, and where only the set or num­ber of finally infected sites is observed. We develop appro­pri­ate Markov chain Monte Carlo algo­rithms, demon­strat­ing their effec­tive­ness, and we study prob­lems of opti­mal exper­i­men­tal design. In par­tic­u­lar, we demon­strate that for lattice-​​based processes an exper­i­ment on a spar­si­fied lat­tice can yield more infor­ma­tion on model para­me­ters than one con­ducted on a com­plete lat­tice. We also prove some prob­a­bilis­tic results about the behav­iour of esti­ma­tors asso­ci­ated with large infected clusters.”

“At the most fun­da­men­tal level, com­put­ers are an assem­bly of gates that are used to per­form the basic oper­a­tions required to exe­cute a pro­gram. For prob­lems in the prob­a­bil­ity domain, even the val­ues used in these most basic oper­a­tions are not con­strained to be either a 0 or a 1. Instead, the basic gates must deter­mine the prob­a­bil­ity that a bit is a 1, or the prob­a­bil­ity that it is a 0.
Lyric’s gates are designed to model rela­tion­ships between prob­a­bil­i­ties natively in the device physics. For this rea­son, Lyric can per­form math­e­mat­i­cal oper­a­tions in the prob­a­bil­ity domain with just a hand­ful of tran­sis­tors – cre­at­ing power and area sav­ings of more than 10X over tra­di­tional implementations.”

“Zipf’s law seems to be ubiq­ui­tous in human lan­guages and appears to be a uni­ver­sal prop­erty of com­plex com­mu­ni­cat­ing sys­tems. Fol­low­ing an early pro­posal made by Zipf con­cern­ing the pres­ence of a ten­sion between the efforts of speaker and hearer in a com­mu­ni­ca­tion sys­tem, we intro­duce evo­lu­tion by means of a vari­a­tional approach to the prob­lem based on Kullback’s Min­i­mum Dis­crim­i­na­tion of Infor­ma­tion Prin­ci­ple. Using a for­mal­ism fully embed­ded in the frame­work of infor­ma­tion the­ory, we demon­strate that Zipf’s law is the only expected out­come of an evolv­ing, com­mu­nica­tive sys­tem under a rig­or­ous def­i­n­i­tion of the com­mu­nica­tive ten­sion described by Zipf.”

“We engi­neer an algo­rithm to solve the approx­i­mate dic­tio­nary match­ing prob­lem. Given a list of words $\mathcal{W}$, max­i­mum dis­tance $d$ fixed at pre­pro­cess­ing time and a query word $q$, we would like to retrieve all words from $\mathcal{W}$ that can be trans­formed into $q$ with $d$ or less edit oper­a­tions. We present data struc­tures that sup­port fault tol­er­ant queries by gen­er­at­ing an index. On top of that, we present a gen­er­al­iza­tion of the method that eases mem­ory con­sump­tion and pre­pro­cess­ing time sig­nif­i­cantly. At the same time, run­ning times of queries are vir­tu­ally unaf­fected. We are able to match in lists of hun­dreds of thou­sands of words and beyond within microsec­onds for rea­son­able distances.”

“The effects of sev­eral non­lin­ear reg­u­lar­iza­tion tech­niques are dis­cussed in the frame­work of 3D seis­mic tomog­ra­phy. Tra­di­tional, lin­ear, $\ell_​2$ penal­ties are com­pared to so-​​called spar­sity pro­mot­ing $\ell_​1$ and $\ell_​0$ penal­ties, and a total vari­a­tion penalty. Which of these algo­rithms is judged opti­mal depends on the spe­cific require­ments of the sci­en­tific exper­i­ment. If the cor­rect repro­duc­tion of model ampli­tudes is impor­tant, clas­si­cal damp­ing towards a smooth model using an $\ell_​2$ norm works almost as well as min­i­miz­ing the total vari­a­tion but is much more effi­cient. If gra­di­ents (edges of anom­alies) should be resolved with a min­i­mum of dis­tor­tion, we pre­fer $\ell_​1$ damp­ing of Daubechies-​​4 wavelet coefficients.…”

“Assume that we observe a large num­ber of curves, all of them with iden­ti­cal, although unknown, shape, but with a dif­fer­ent ran­dom shift. The objec­tive is to esti­mate the indi­vid­ual time shifts and their dis­tri­b­u­tion. Such an objec­tive appears in sev­eral bio­log­i­cal appli­ca­tions like neu­ro­science or ECG sig­nal pro­cess­ing, in which the esti­ma­tion of the dis­tri­b­u­tion of the elapsed time between repet­i­tive pulses with a pos­si­bly low signal-​​noise ratio, and with­out a knowl­edge of the pulse shape is of inter­est. We sug­gest an M-​​estimator lead­ing to a three-​​stage algo­rithm: we split our data set in blocks, on which the esti­ma­tion of the shifts is done by min­i­miz­ing a cost cri­te­rion based on a func­tional of the peri­odogram; the esti­mated shifts are then plugged into a stan­dard den­sity esti­ma­tor. We show that under mild reg­u­lar­ity assump­tions the den­sity esti­mate con­verges weakly to the true shift dis­tri­b­u­tion. The the­ory is applied both to sim­u­la­tions and to align­ment of real ECG signals.…”

“Moti­va­tion: Sec­ond gen­er­a­tion sequenc­ing tech­nol­ogy makes it fea­si­ble for many researches to obtain enough sequence reads to attempt the de novo assem­bly of higher eukary­otes (includ­ing mam­mals). De novo assem­bly not only pro­vides a tool for under­stand­ing wide scale bio­log­i­cal vari­a­tion, but within human bio-​​medicine, it offers a direct way of observ­ing both large scale struc­tural vari­a­tion and fine scale sequence vari­a­tion. Unfor­tu­nately, improve­ments in the com­pu­ta­tional fea­si­bil­ity for de novo assem­bly have not matched the improve­ments in the gath­er­ing of sequence data. This is for two rea­sons: the inher­ent com­pu­ta­tional com­plex­ity of the prob­lem, and the in-​​practice mem­ory require­ments of tools.”

“Whereas a con­ven­tional NAND gate out­puts a “1” if nei­ther of its inputs match, the out­put of a Bayesian NAND gate rep­re­sents the odds that the two input prob­a­bil­i­ties match. This makes it pos­si­ble to per­form cal­cu­la­tions that use prob­a­bil­i­ties as their input and output.”