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

• Tar­get Expres­sion Exam­ples — Eureqa Formulize

  ‘The “Tar­get Expres­sion” in the field at the top of the Set Tar­get tab tells For­mulize what type of model to search for. By default, the tar­get expres­sion is an equa­tion where y (or, if there’s no y, what­ever vari­able is in col­umn A) is mod­eled as a func­tion of all other vari­ables. To edit the tar­get expres­sion, click on it, then make the desired alter­ations. Use the spe­cial func­tion f(…) to spec­ify the part of the equa­tion that For­mulize will attempt to fill in; For­mulize will search for the for­mula f(…) using the vari­ables you put inside the parentheses.’
  
  for­mulize eureqa genetic-​​programming symbolic-​​regression mod­el­ing doc­u­men­ta­tion

• A New Solu­tion to the Puz­zle of Sim­plic­ity — PhilSci-​​Archive

  “Explain­ing the con­nec­tion, if any, between sim­plic­ity and truth is among the deep­est prob­lems fac­ing the phi­los­o­phy of sci­ence, sta­tis­tics, and machine learn­ing. Say that an effi­cient truth-​​finding method min­i­mizes worst-​​case costs en route to con­verg­ing to the true answer to a the­ory choice prob­lem. Let the costs con­sid­ered include the num­ber of times a false answer is selected, the num­ber of times opin­ion is reversed, and the times at which the rever­sals occur. It is demon­strated that (1)always choos­ing the sim­plest the­ory com­pat­i­ble with expe­ri­ence and (2) hang­ing onto it while it remains sim­plest is both nec­es­sary and suf­fi­cient for efficiency.”
  
  via:cshalizi occam’s-razor sim­plic­ity model-​​discovery expla­na­tion philosophy-​​of-​​science

• [1206.4599] A Uni­fied Robust Clas­si­fi­ca­tion Model

  “A wide vari­ety of machine learn­ing algo­rithms such as sup­port vec­tor machine (SVM), min­i­max prob­a­bil­ity machine (MPM), and Fisher dis­crim­i­nant analy­sis (FDA), exist for binary clas­si­fi­ca­tion. The pur­pose of this paper is to pro­vide a uni­fied clas­si­fi­ca­tion model that includes the above mod­els through a robust opti­miza­tion approach. This uni­fied model has sev­eral ben­e­fits. One is that the exten­sions and improve­ments intended for SVM become applic­a­ble to MPM and FDA, and vice versa. Another ben­e­fit is to pro­vide the­o­ret­i­cal results to above learn­ing meth­ods at once by deal­ing with the uni­fied model. We give a sta­tis­ti­cal inter­pre­ta­tion of the uni­fied clas­si­fi­ca­tion model and pro­pose a non-​​convex opti­miza­tion algo­rithm that can be applied to non-​​convex vari­ants of exist­ing learn­ing methods.”
  
  clas­si­fi­ca­tion algo­rithms lumpers-​​and-​​spliters-​​sittin-​​in-​​a-​​tree

• CUDA Down­loads | NVIDIA Devel­oper Zone

  This release of the CUDA Toolkit  enables devel­op­ment using GPUs using the Kepler archi­tec­ture, such as the GeForce GTX680. Fea­ture and func­tion­al­ity builds on the foun­da­tion of the CUDA 4.1 release which intro­duced: A new  LLVM-​​based CUDA com­piler 1000+ new image pro­cess­ing func­tions Redesigned Visual Pro­filer with auto­mated per­for­mance analy­sis and inte­grated expert guidance
  
  CUDA GPU pro­gram­ming library MacOS

• [1206.2057] Fin­ish­ing Flows Quickly with Pre­emp­tive Scheduling

  “Today’s data cen­ters face extreme chal­lenges in pro­vid­ing low latency. How­ever, fair shar­ing, a prin­ci­ple com­monly adopted in cur­rent con­ges­tion con­trol pro­to­cols, is far from opti­mal for sat­is­fy­ing latency require­ments. We pro­pose Pre­emp­tive Dis­trib­uted Quick (PDQ) flow sched­ul­ing, a pro­to­col designed to com­plete flows quickly and meet flow dead­lines. PDQ enables flow pre­emp­tion to approx­i­mate a range of sched­ul­ing dis­ci­plines. For exam­ple, PDQ can emu­late a short­est job first algo­rithm to give pri­or­ity to the short flows by paus­ing the con­tend­ing flows. PDQ bor­rows ideas from cen­tral­ized sched­ul­ing dis­ci­plines and imple­ments them in a fully dis­trib­uted man­ner, mak­ing it scal­able to today’s data cen­ters. Fur­ther, we develop a mul­ti­path ver­sion of PDQ to exploit path diver­sity. Through exten­sive packet-​​level and flow-​​level sim­u­la­tion, we demon­strate that PDQ sig­nif­i­cantly out­per­forms TCP, RCP and D3 in data cen­ter envi­ron­ments. We fur­ther show that PDQ is sta­ble, resilient to packet loss, and pre­serves nearly all its per­for­mance gains even given inac­cu­rate flow information.”
  
  queuing-​​models engineering-​​design algo­rithms performance-​​measure nudge-​​targets

• [1206.2216] Com­plex Sys­tems Sci­ence: Dreams of Uni­ver­sal­ity, Real­ity of Interdisciplinarity

  “Using a large data­base (~ 215 000 records) of rel­e­vant arti­cles, we empir­i­cally study the “com­plex sys­tems” field and its claims to find uni­ver­sal prin­ci­ples apply­ing to sys­tems in gen­eral. The study of ref­er­ences shared by the papers allows us to obtain a global point of view on the struc­ture of this highly inter­dis­ci­pli­nary field. We show that its over­all coher­ence does not arise from a uni­ver­sal the­ory but instead from com­pu­ta­tional tech­niques and fruit­ful adap­ta­tions of the idea of self-​​organization to spe­cific sys­tems. We also find that com­mu­ni­ca­tion between dif­fer­ent dis­ci­plines goes through spe­cific “trad­ing zones”, ie sub-​​communities that cre­ate an inter­face around spe­cific tools (a DNA microchip) or con­cepts (a network).”
  
  via:cshalizi com­plex­ol­ogy pro­fes­sion­al­iza­tion network-​​theory disappointed-​​by-​​lack-​​of-​​Abbott-​​ref citation-​​networks