[ISEA2011] Panel: Rob Saun­ders – Com­puter Vi­sion for Cu­ri­ous Ma­chines

Panel Statement

Panel: Surveillant Spaces: From Autonomous Surveillance to Machine Voyeurism

Com­puter vi­sion is a branch of ar­ti­fi­cial in­tel­li­gence that is con­cerned with de­vel­op­ing al­go­rithms to allow ma­chines to process and re­spond to vi­sual data.  The de­gree to which the early pi­o­neers in ar­ti­fi­cial in­tel­li­gence un­der­es­ti­mated the chal­lenges in­volved is often il­lus­trated with an anec­dote about Mar­vin Min­sky in­struct­ing a stu­dent to solve “the prob­lem of com­puter vi­sion” as a sum­mer pro­ject. The first break­throughs in com­puter vi­sion came from the field of com­pu­ta­tional neu­ro­science and the work of David Marr, build­ing low-level mod­els of the vi­sual cor­tex from the ground up. Nearly forty years later, the state-of-the-art in com­puter vi­sion is still very much in the process of con­struct­ing rel­a­tively prim­i­tive rep­re­sen­ta­tions from cap­tured im­ages. Nev­er­the­less, the re­search has pro­duced a wealth of tech­niques that can be ap­plied to suit­ably struc­tured scenes to ex­tract mean­ing­ful in­for­ma­tion.  Many of these tech­niques are sim­ple enough that they can be im­ple­mented by novice pro­gram­mers while more so­phis­ti­cated tech­niques have be­come read­ily avail­able through pro­gram­ming li­braries and off-the-shelf soft­ware. The avail­abil­ity of com­puter vi­sion tech­nol­ogy pro­vides a base for ex­per­i­ment­ing with ma­chine au­ton­omy in cre­ative do­mains. In this panel I will dis­cuss the pos­si­bil­ity of de­vel­op­ing au­tonomous ma­chine per­form­ers that take ad­van­tage of the ad­vances in com­puter vi­sion by first re­view­ing some of rel­e­vant low-level and high-level tech­niques and show­ing how these can be in­te­grated with ma­chine learn­ing sys­tems. In par­tic­u­lar, I will pre­sent this ex­plo­ration in re­la­tion to my re­search de­vel­op­ing self-mo­ti­vated (cu­ri­ous) agents and my col­lab­o­ra­tions with Petra Gemein­boeck ex­plor­ing the per­for­ma­tiv­ity of the gaze through the cre­ation of ma­chine aug­mented en­vi­ron­ments. In this work­shop-like ses­sion we will ex­plore the con­struc­tion of a self-mo­ti­vated ma­chinic voyeur, ex­am­ine what it sees, how it re­sponds and what dri­ves it.

  • Rob Saun­ders is Se­nior Lec­turer in De­sign Com­put­ing in the Fac­ulty of Ar­chi­tec­ture, De­sign and Plan­ning at the Uni­ver­sity of Syd­ney, AU. Rob’s re­search cen­tres around cre­ative ap­pli­ca­tion of com­put­ing and the com­pu­ta­tional mod­el­ling of cre­ativ­ity. Using tech­niques from ma­chine learn­ing, ro­bot­ics and sur­veil­lance he has ex­plored the role of cu­rios­ity in cre­ative processes and de­vel­oped mod­els of cre­ative sys­tems at in­di­vid­ual, so­cial and cul­tural lev­els. His mod­els of cu­ri­ous de­sign agents have demon­strated use­ful abil­i­ties for au­tonomous de­sign sys­tems, in­clud­ing prob­lem-find­ing and open-ended ex­plo­ration. Rob works with artists and de­sign­ers across a range of dis­ci­plines to sup­port and en­gage in the cre­ative ap­pli­ca­tion of com­put­ing and has ap­plied his re­search in the de­vel­op­ment of de­sign cus­tomi­sa­tion sys­tems, smart en­vi­ron­ments, in­ter­ac­tive in­stal­la­tions and ro­botic art­works.