These are my recent Pin​board​.in links:

• [1112.6209] Build­ing high-​​level fea­tures using large scale unsu­per­vised learning

  We con­sider the prob­lem of build­ing detec­tors for high-​​level con­cepts using only unsu­per­vised fea­ture learn­ing. For exam­ple, we would like to under­stand if it is pos­si­ble to learn a face detec­tor using only unla­beled images down­loaded from the inter­net. To answer this ques­tion, we trained a sim­ple fea­ture learn­ing algo­rithm on a large dataset of images (10 mil­lion images, each image is 200×200). The sim­u­la­tion is per­formed on a clus­ter of 1000 machines with fast net­work hard­ware for one week. Exten­sive exper­i­men­tal results reveal sur­pris­ing evi­dence that such high-​​level con­cepts can indeed be learned using only unla­beled data and a sim­ple learn­ing algorithm.
  
  image-​​analysis image-​​segmentation unsupervised-​​learning learning-​​by-​​doing feature-​​extraction nudge-​​targets

• [1105.0158] Detect­ing emer­gent processes in cel­lu­lar automata with excess information

  Many nat­ural processes occur over char­ac­ter­is­tic spa­tial and tem­po­ral scales. This paper presents tools for (i) flex­i­bly and scal­ably coarse-​​graining cel­lu­lar automata and (ii) iden­ti­fy­ing which coarse-​​grainings express an automaton’s dynam­ics well, and which express its dynam­ics badly. We apply the tools to inves­ti­gate a range of exam­ples in Conway’s Game of Life and Hop­field net­works and demon­strate that they cap­ture some basic intu­itions about emer­gent processes. Finally, we for­mal­ize the notion that a process is emer­gent if it is bet­ter expressed at a coarser granularity.
  
  emer­gence com­plex­ol­ogy cellular-​​automata signal-​​processing nudge-​​targets

• [1008.0901] Con­ver­gence to global con­sen­sus in opin­ion dynam­ics under a non­lin­ear voter model

  We pro­pose a non­lin­ear voter model to study the emer­gence of global con­sen­sus in opin­ion dynam­ics. In our model, agent $i$ agrees with one of binary opin­ions with the prob­a­bil­ity that is a power func­tion of the num­ber of agents hold­ing this opin­ion among agent $i$ and its near­est neigh­bors, where an adjustable para­me­ter $alpha$ con­trols the effect of herd behav­ior on con­sen­sus. We find that there exists an opti­mal value of $alpha$ lead­ing to the fastest con­sen­sus for lat­tices, ran­dom graphs, small-​​world net­works and scale-​​free net­works. Qual­i­ta­tive insights are obtained by exam­in­ing the spa­tiotem­po­ral evo­lu­tion of the opin­ion clusters.
  
  agent-​​based social-​​dynamics network-​​theory com­plex­ol­ogy nudge-​​targets

• [1110.4876] REBOUND: An open-​​source multi-​​purpose N-​​body code for col­li­sional dynamics

  REBOUND is a new multi-​​purpose N-​​body code which is freely avail­able under an open-​​source license. It was designed for col­li­sional dynam­ics such as plan­e­tary rings but can also solve the clas­si­cal N-​​body prob­lem. It is highly mod­u­lar and can be cus­tomized eas­ily to work on a wide vari­ety of dif­fer­ent prob­lems in astro­physics and beyond.
  
  sim­u­la­tion computational-​​science astro­physics numerical-​​methods sim­u­la­tor library open-​​source nudge-​​targets

• [1112.5908] Query Answer­ing under Match­ing Depen­den­cies for Data Clean­ing: Com­plex­ity and Algorithms

  Match­ing depen­den­cies (MDs) have been recently intro­duced as declar­a­tive rules for entity res­o­lu­tion (ER), i.e. for iden­ti­fy­ing and resolv­ing dupli­cates in rela­tional instance $D$. A set of MDs can be used as the basis for a pos­si­bly non-​​deterministic mech­a­nism that com­putes a duplicate-​​free instance from $D$. The pos­si­ble results of this process are the clean, “min­i­mally resolved instances” (MRIs). There might be sev­eral MRIs for $D$, and the “resolved answers” to a query are those that are shared by all the MRIs. We inves­ti­gate the prob­lem of com­put­ing resolved answers. We look at var­i­ous sets of MDs, devel­op­ing syn­tac­tic cri­te­ria for deter­min­ing (in)tractability of the resolved answer prob­lem, includ­ing a dichotomy result. For some tractable classes of MDs and con­junc­tive queries, we present a query rewrit­ing method­ol­ogy that can be used to retrieve the resolved answers. We also inves­ti­gate con­nec­tions with “con­sis­tent query answer­ing”, deriv­ing fur­ther tractabil­ity results for MD-​​based ER.
  
  data­bases graph-​​theory algo­rithms nudge-​​targets

• The Wash­room Game by Jan Heufer :: SSRN

  This arti­cle analy­ses a game where play­ers sequen­tially choose either to become insid­ers and pick one of finitely many loca­tions or to remain out­siders. They will only become insid­ers if a min­i­mum dis­tance to the next player can be assured; their sec­ondary objec­tive is to max­i­mize the min­i­mal dis­tance to other play­ers. This is illus­trated by con­sid­er­ing the strate­gic behav­ior of men choos­ing from a set of uri­nals in a pub­lic lava­tory. How­ever, besides very sim­i­lar sit­u­a­tions (e.g. set­tling of res­i­dents in a newly devel­oped area, the selec­tion of food patches by for­ag­ing ani­mals, choos­ing seats in wait­ing rooms or lines in a swim­ming pool), the game might also rel­e­vant to the prob­lem of plac­ing bill­boards attempt­ing to catch the atten­tion of passers-​​by or sim­i­lar eco­nomic sit­u­a­tions. In the non-​​cooperative equi­lib­rium, all insid­ers behave as if they coop­er­ated with each other and min­i­mized the total num­ber of insid­ers. It is shown that strate­gic behav­ior leads to an equi­lib­rium with sub­stan­tial under uti­liza­tion of avail­able loca­tions. Increas­ing the num­ber of loca­tions tends to decrease uti­liza­tion. The removal of some loca­tions which leads to gaps can not only increase rel­a­tive uti­liza­tion but even absolute max­i­mum capacity.
  
  game-​​theory agent-​​based com­plex­ol­ogy eco­nom­ics nudge-​​targets

• [1109.0777] Effi­cient and Cor­rect Sten­cil Com­pu­ta­tion via Pat­tern Match­ing and Sta­tic Typing

  Sten­cil com­pu­ta­tions, involv­ing oper­a­tions over the ele­ments of an array, are a com­mon pro­gram­ming pat­tern in sci­en­tific com­put­ing, games, and image pro­cess­ing. As a pro­gram­ming pat­tern, sten­cil com­pu­ta­tions are highly reg­u­lar and amenable to opti­mi­sa­tion and par­al­leli­sa­tion. How­ever, general-​​purpose lan­guages obscure this reg­u­lar pat­tern from the com­piler, and even the pro­gram­mer, pre­vent­ing opti­mi­sa­tion and obfus­cat­ing (in)correctness. This paper fur­thers our work on the Ypnos domain-​​specific lan­guage for sten­cil com­pu­ta­tions embed­ded in Haskell. Ypnos allows declar­a­tive, abstract spec­i­fi­ca­tion of sten­cil com­pu­ta­tions, expos­ing the struc­ture of a prob­lem to the com­piler and to the pro­gram­mer via spe­cialised syn­tax. In this paper we show the decid­able safety guar­an­tee that well-​​formed, well-​​typed Ypnos pro­grams can­not index out­side of array bound­aries. Thus index­ing in Ypnos is safe and run-​​time bounds check­ing can be elim­i­nated. Pro­gram infor­ma­tion is encoded as types, using the advanced type-​​system fea­tures of the Glas­gow Haskell Com­piler, with the safe-​​indexing invari­ant enforced at com­pile time via type checking.
  
  domain-​​specific-​​language algo­rithms grid-​​computing nudge-​​targets

• What’s Chal­leng­ing About Paul? : Lawyers, Guns & Money

  It’s wrong to think of Ron Paul’s racism and his lib­er­tar­i­an­ism as two dis­tinct parts of his polit­i­cal per­sona, when in fact they are deeply tied together. White suprema­cists under­stand what Glenn, appar­ently, does not; the absence of Fed­eral author­ity makes it eas­ier for pri­vate actors and local gov­ern­ments to repress the civil and polit­i­cal rights of minori­ties. Paul’s lib­er­tar­i­an­ism emerged in a regional and cul­tural con­text that was deeply hos­tile to Fed­eral efforts at inte­gra­tion. The newslet­ters give strong indi­ca­tion that none of this is lost on Ron Paul. A notional Pres­i­dent Paul is just as likely to use the pow­ers of the office to gut Fed­eral enforce­ment of a wide range of civil lib­er­ties pro­tec­tions as he is to do any of the things that Glenn would like him to do.
  
  pol­i­tics lib­er­tar­i­an­ism racism con­ser­vatism pop­ulism