“In this paper, we briefly review the con­cept of mem­ory cir­cuit ele­ments, namely mem­ris­tors, mem­ca­pac­i­tors and memin­duc­tors, and then dis­cuss some appli­ca­tions by focus­ing mainly on the first class. We present sev­eral exam­ples, their mod­el­ing and appli­ca­tions rang­ing from ana­log pro­gram­ming to bio­log­i­cal sys­tems. Since the phe­nom­ena asso­ci­ated with mem­ory are ubiq­ui­tous at the nanoscale, we expect the inter­est in these cir­cuit ele­ments to increase in com­ing years.”

“…The intrin­sic under­ly­ing struc­ture of the sys­tem is mod­eled by an epsilon-​​machine and its causal states. The deci­sional states are the emerg­ing pat­terns cor­re­spond­ing to the util­ity func­tion. In a com­plex sys­tems per­spec­tive, these pat­terns thus form a par­ti­tion of the lower-​​level sys­tem states that is defined accord­ing to the higher-​​level user’s knowl­edge. The tran­si­tions between these deci­sional states cor­re­spond to events that lead to a change of deci­sion. An algo­rithm is pro­vided so as to esti­mate the states and their tran­si­tions from data. Appli­ca­tion exam­ples are given for hid­den model recon­struc­tion, cel­lu­lar automata fil­ter­ing, and edge detec­tion in images.”

“A still more real­is­tic and sub­tle, but much more trou­ble­some sce­nario: Finan­cial Unde­tectable Jour­nal­is­tic Engi­neer­ing (FUJE). Finan­cial news jour­nal­ists could word the reports dif­fer­ently and send very dif­fer­ent sig­nals to the robot army. Here’re two actual news head­lines re. the May NFP num­ber (inci­den­tally, both are from the same out­let, same day, dif­fer­ent reporter — just a ran­dom google search):

US adds 431,000 jobs in May, unem­ploy­ment down to 9.7 pct
vs.

Despite Adding 431K Jobs, May Non-​​Farm Pay­roll Fig­ures Dis­ap­point
The first is fac­tual; the sec­ond con­tains more in-​​depth analy­sis. It takes an expe­ri­enced human to parse and rec­on­cile the two. You can see how robot read­ers may assign oppo­site signs to each.”

“Sparse mod­el­ing is a pow­er­ful frame­work for data analy­sis and pro­cess­ing. Tra­di­tion­ally, encod­ing in this frame­work is per­formed by solv­ing an L1-​​regularized lin­ear regres­sion prob­lem, com­monly referred to as Lasso or Basis Pur­suit. In this work we com­bine the sparsity-​​inducing prop­erty of the Lasso model at the indi­vid­ual fea­ture level, with the block-​​sparsity prop­erty of the Group Lasso model, where sparse groups of fea­tures are jointly encoded, obtain­ing a spar­sity pat­tern hier­ar­chi­cally struc­tured. This results in the Hier­ar­chi­cal Lasso (HiLasso), which shows impor­tant prac­ti­cal mod­el­ing advantages.…”

“… In this paper, we pro­vide a for­mal intro­duc­tion to rif­fled inde­pen­dence and present algo­rithms for using rif­fled inde­pen­dence within Fourier-​​theoretic frame­works which have been explored by a num­ber of recent papers. Addi­tion­ally, we pro­pose an auto­mated method for dis­cov­er­ing sets of items which are rif­fle inde­pen­dent from a train­ing set of rank­ings. We show that our clustering-​​like algo­rithms can be used to dis­cover mean­ing­ful latent coali­tions from real pref­er­ence rank­ing datasets and to learn the struc­ture of hier­ar­chi­cally decom­pos­able mod­els based on rif­fled independence.”

“Infer­en­tial sum­maries of tree esti­mates are use­ful in the set­ting of evo­lu­tion­ary biol­ogy, where phy­lo­ge­netic trees have been built from DNA data since the 1960’s. In bioin­for­mat­ics, psy­cho­met­rics and data min­ing, hier­ar­chi­cal clus­ter­ing tech­niques out­put the same math­e­mat­i­cal objects, and prac­ti­tion­ers have sim­i­lar ques­tions about the sta­bil­ity and ‘gen­er­al­iz­abil­ity’ of these sum­maries. This paper pro­vides an imple­men­ta­tion of the geo­met­ric dis­tance between trees devel­oped by Billera, Holmes and Vogt­mann (2001) [BHV] equally applic­a­ble to phy­lo­ge­netic trees and hieirar­chi­cal clus­ter­ing trees, and shows some of the appli­ca­tions in sta­tis­ti­cal infer­ence for which this dis­tance can be useful.…Our method gives a new way of eval­u­at­ing the influ­ence both of cer­tain columns (posi­tions, vari­ables or genes) and of cer­tain rows (whether species, obser­va­tions or arrays).”

“It is now clear that cladis­tics can be applied and be use­ful to the study of galaxy diver­si­fi­ca­tion. Many dif­fi­cul­ties, con­cep­tual and prac­ti­cal, have been solved,. Sig­nif­i­cant astro­phys­i­cal results have been obtained and will be extended to larger sam­ples of galax­ies and glob­u­lar clus­ters. How­ever, many paths remain in the explo­ration of this new and large field of research.”

will I ever under­stand all the effort sta­tis­ti­cians put into what I con­sider a solved prob­lem? Pareto-​​GP is appar­ently utterly unknown, still

“A widely used tool in the study of risk, insur­ance and extreme val­ues is the mean excess plot. One use is for val­i­dat­ing a gen­er­al­ized Pareto model for the excess dis­tri­b­u­tion. This paper inves­ti­gates some the­o­ret­i­cal and prac­ti­cal aspects of the use of the mean excess plot.”

“Monte Carlo meth­ods play an impor­tant role in sci­en­tific com­pu­ta­tion, espe­cially when prob­lems have a vast phase space. In this chap­ter an intro­duc­tion to the Monte Carlo method is given. Con­cepts such as Markov chains, detailed bal­ance, crit­i­cal slow­ing down, and ergod­ic­ity, as well as the Metrop­o­lis algo­rithm are explained. The Monte Carlo method is illus­trated by numer­i­cally study­ing the crit­i­cal behav­ior of the two-​​dimensional Ising fer­ro­mag­net using finite-​​size scal­ing meth­ods. Fur­ther­more, advanced Monte Carlo meth­ods are described (par­al­lel tem­per­ing Monte Carlo) and illus­trated with non­triv­ial mod­els from the physics of glassy systems.”

“The beam-​​splitting method for fine-​​heterogeneity cor– rec­tion will inevitably mul­ti­ply beams to trans­port and thus will slow down dose cal­cu­la­tion. With the GDS al– gorithm, the dose con­vo­lu­tion is made only once after all the beams have been trans­ported, which min­i­mizes the impact of the beam mul­ti­pli­ca­tion on com­put­ing time. In fact, for the beams indi­vid­u­ally split into sev­eral tens, the cal­cu­la­tion time increased only by sev­eral times with the GDS. This algo­rith­mic frame­work will thus enable fast and accu­rate treat­ment plan­ning of heavy charged par­ti­cle ra– dio­ther­apy in the pres­ence of den­sity het­ero­gene­ity finer than the size of intrin­sic beam blurring.”

“Many fixed-​​parameter tractable algo­rithms using a bounded search tree have been repeat­edly improved, often by describ­ing a larger num­ber of branch­ing rules involv­ing an increas­ingly com­plex case analy­sis. We intro­duce a novel and gen­eral branch­ing strat­egy that branches on the for­bid­den sub­graphs of a relaxed class of graphs. By using the class of P4-​​sparse graphs as the relaxed graph class, we obtain effi­cient bounded-​​search tree algo­rithms for sev­eral para­me­ter­ized dele­tion prob­lems. For the cograph edge-​​deletion prob­lem and the triv­ially per­fect edge-​​deletion prob­lem, the branch­ing strat­egy yields the first non-​​trivial bounded-​​search tree algo­rithms. For the cograph ver­tex dele­tion prob­lem, the run­ning time of our sim­ple bounded search algo­rithm matches those pre­vi­ously designed with the help of com­pli­cated case dis­tinc­tions and non-​​trivial run­ning time analy­sis [16] and computer-​​aided branch­ing rules [7]”

“Ron Doer­fler has cre­ated a truly gor­geous 2010 cal­en­dar titled The Age of Graph­i­cal Com­put­ing. Ron has trans­formed nomog­ra­phy into a form of art.”

“Decid­ing whether a graph can be embed­ded in a grid using only unit-​​length edges is NP-​​complete, even when restricted to binary trees. How­ever, it is not dif­fi­cult to devise a num­ber of graph classes for which the prob­lem is poly­no­mial, even triv­ial. A nat­ural step, out­stand­ing thus far, was to pro­vide a broad clas­si­fi­ca­tion of graphs that make for poly­no­mial or NP-​​complete instances. We pro­vide such a clas­si­fi­ca­tion based on the set of allowed ver­tex degrees in the input graphs, yield­ing a full dichotomy on the com­plex­ity of the prob­lem. As byprod­ucts, the pre­vi­ous NP-​​completeness result for binary trees was strength­ened to strictly binary trees, and the three-​​dimensional ver­sion of the prob­lem was for the first time proven to be NP-​​complete. Our results were made pos­si­ble by intro­duc­ing the con­cepts of con­sis­tent ori­en­ta­tions and robust gad­gets, and by show­ing how the for­mer allows NP-​​completeness proofs by local replace­ment even in the absence of the latter.”

“We con­sider the prob­lem of choos­ing between sev­eral mod­els in least-​​squares regres­sion with het­eroscedas­tic data. We prove that any penal­iza­tion pro­ce­dure is sub­op­ti­mal when the penalty is a func­tion of the dimen­sion of the model, at least for some typ­i­cal het­eroscedas­tic model selec­tion prob­lems. In par­tic­u­lar, Mal­lows’ Cp is sub­op­ti­mal in this frame­work. On the con­trary, opti­mal model selec­tion is pos­si­ble with data-​​driven penal­ties such as resam­pling or $V$-fold penal­ties. There­fore, it is worth esti­mat­ing the shape of the penalty from data, even at the price of a higher com­pu­ta­tional cost. Sim­u­la­tion exper­i­ments illus­trate the exis­tence of a trade-​​off between sta­tis­ti­cal accu­racy and com­pu­ta­tional com­plex­ity. As a con­clu­sion, we sketch some rules for choos­ing a penalty in least-​​squares regres­sion, depend­ing on what is known about pos­si­ble vari­a­tions of the noise-​​level.”

“…This idea arises from pre­vi­ous works in which com­pu­ta­tional mod­els of self-​​assembly were sub­ject to evo­lu­tion­ary design in order to per­form the auto­matic con­struc­tion of user-​​defined struc­tures. Then, the aim of this paper is to present a novel method­ol­ogy for the auto­mated design of heuris­tics by means of self-​​assembly.”

“The rig­or­ous the­o­ret­i­cal analy­ses of algo­rithms for #SAT have been pro­posed in the lit­er­a­ture. As we know, pre­vi­ous algo­rithms for solv­ing #SAT have been ana­lyzed only regard­ing the num­ber of vari­ables as the para­me­ter. How­ever, the time com­plex­ity for solv­ing #SAT instances depends not only on the num­ber of vari­ables, but also on the num­ber of clauses. There­fore, it is sig­nif­i­cant to exploit the time com­plex­ity from the other point of view, i.e. the num­ber of clauses. In this paper, we present algo­rithms for solv­ing #2-​​SAT and #3-​​SAT with rig­or­ous com­plex­ity analy­ses using the num­ber of clauses as the para­me­ter. By ana­lyz­ing the algo­rithms, we obtain the new worst-​​case upper bounds O(1.1892m) for #2-​​SAT and O(1.4142m) for #3-​​SAT, where m is the num­ber of clauses.”

“Phy­lo­ge­netic trees and net­works are leaf-​​labelled graphs that are used to describe evo­lu­tion­ary his­to­ries of species. The Tree Con­tain­ment prob­lem asks whether a given phy­lo­ge­netic tree is embed­ded in a given phy­lo­ge­netic net­work. Given a phy­lo­ge­netic net­work and a clus­ter of species, the Clus­ter Con­tain­ment prob­lem asks whether the given clus­ter is a clus­ter of some phy­lo­ge­netic tree embed­ded in the net­work. Both prob­lems are known to be NP-​​complete in general.…”

“… Kanso claims there is no effi­cient attack against the ASG(r,s) since r and s are kept secret. In this paper, we present an Alter­nat­ing Step Gen­er­a­tor, ASG, model for the ASG(r,s) and also we present a new and effi­cient alge­braic attack on ASG(r,s) using 3(m+n) bits of the out­put sequence to find the secret key with O((m^2+n^2)*2^{l+1}+ (2^{m-1})*m^3 + (2^{n-1})*n^3) com­pu­ta­tional com­plex­ity. We show that this sys­tem is no more secure than the orig­i­nal ASG, in con­trast to the claim of the ASG(r,s)‘s constructor.”

seems like one could search for heuris­tics (and full algo­rithms) which can tell you whether two con­fig­u­ra­tion matri­ces might describe the same skew con­fig­u­ra­tion, &c

“The paper is writ­ten in the form of intro­duc­tion to the sub­ject, with much of the mate­r­ial acces­si­ble to advanced high school stu­dents. How­ever, in the part of the sur­vey con­cern­ing con­fig­u­ra­tions of lines in gen­eral posi­tion in the three-​​dimensional space the expo­si­tion is free from any back­ground restric­tions. We have added few new results, fixed few mis­prints and ter­mi­no­log­i­cal inac­cu­ra­cies and expanded the ref­er­ence list. Notice that some of the results pre­sented in the paper appeared in other papers with­out appro­pri­ate references.”

“Evo­lu­tion­ary adap­ta­tion is the process that increases the fit of a pop­u­la­tion to the fit­ness land­scape it inhab­its. As a con­se­quence, evo­lu­tion­ary dynam­ics is shaped, con­strained, and chan­neled, by that fit­ness land­scape. Much work has been expended to under­stand the evo­lu­tion­ary dynam­ics of adapt­ing pop­u­la­tions, but much less is known about the struc­ture of the land­scapes. Here, we study the global and local struc­ture of com­plex fit­ness land­scapes of inter­act­ing loci that describe pro­tein folds or sets of inter­act­ing genes form­ing path­ways or mod­ules. We find that in these land­scapes, high peaks are more likely to be found near other high peaks, cor­rob­o­rat­ing Kauffman’s “Mas­sif Cen­tral” hypoth­e­sis. We study the clus­ters of peaks as a func­tion of the rugged­ness of the land­scape and find that this clus­ter­ing allows peaks to form inter­con­nected networks.…”

“The feed-​​forward rela­tion­ship nat­u­rally observed in time-​​dependent processes and in a diverse num­ber of real sys­tems –such as some food-​​webs and elec­tronic and neural wiring– can be described in terms of so-​​called directed acyclic graphs (DAGs). An impor­tant ingre­di­ent of the analy­sis of such net­works is a proper com­par­i­son of their observed archi­tec­ture against an ensem­ble of ran­dom­ized graphs, thereby quan­ti­fy­ing the {\em ran­dom­ness} of the real sys­tems with respect to suit­able null mod­els. This approx­i­ma­tion is par­tic­u­larly rel­e­vant when the finite size and/​or large con­nec­tiv­ity of real sys­tems make inad­e­quate a com­par­i­son with the pre­dic­tions obtained from the so-​​called {\em con­fig­u­ra­tion model}. In this paper we ana­lyze four meth­ods of DAG ran­dom­iza­tion as defined by the desired com­bi­na­tion of topo­log­i­cal invari­ants (directed and undi­rected degree sequence and com­po­nent dis­tri­b­u­tions) aimed to be preserved.…”

“In this paper, we con­sider the prob­lem of block­ing mali­cious traf­fic on the Inter­net, via source-​​based fil­ter­ing. In par­tic­u­lar, we con­sider fil­ter­ing via access con­trol lists (ACLs): these are already avail­able at the routers today but are a scarce resource because they are stored in the expen­sive ternary con­tent address­able mem­ory (TCAM). Aggre­ga­tion (by fil­ter­ing source pre­fixes instead of indi­vid­ual IP addresses) helps reduce the num­ber of fil­ters, but comes also at the cost of block­ing legit­i­mate traf­fic orig­i­nat­ing from the fil­tered pre­fixes. We show how to opti­mally choose which source pre­fixes to fil­ter, for a vari­ety of real­is­tic attack sce­nar­ios and oper­a­tors’ poli­cies. In each sce­nario, we design opti­mal, yet com­pu­ta­tion­ally effi­cient, algo­rithms. Using logs from Dshield​.org, we eval­u­ate the algo­rithms and demon­strate that they bring sig­nif­i­cant ben­e­fit in practice.”

“We give an algo­rithm that com­putes the final state of cer­tain growth mod­els with­out com­put­ing all inter­me­di­ate states. Our tech­nique is based on a “least action prin­ci­ple” which char­ac­ter­izes the odome­ter func­tion of the growth process. Start­ing from an edu­cated guess for the odome­ter, we suc­ces­sively cor­rect under– and over­es­ti­mates and prov­ably arrive at the cor­rect final state. The degree of speedup depends on the accu­racy of the ini­tial guess.
Deter­min­ing the size of the bound­ary fluc­tu­a­tions in inter­nal diffusion-​​limited aggre­ga­tion is a long-​​standing open prob­lem in sta­tis­ti­cal physics. As an appli­ca­tion of our method, we cal­cu­late the size of fluc­tu­a­tions over two orders of mag­ni­tude beyond pre­vi­ous sim­u­la­tions. Our data strongly sup­port the con­jec­ture that the fluc­tu­a­tions are log­a­rith­mic in the radius.”