BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:The model is simple until proven otherwise - Anita Faul\, LSC\, Un
 iversity of Cambridge
DTSTART:20190522T120000Z
DTEND:20190522T130000Z
UID:TALK123838@talks.cam.ac.uk
CONTACT:Chris Richardson
DESCRIPTION:Machine Learning and AI have enjoyed an unprecedented rise\nin
  popularity. In academia as well as industry\, they are often viewed\nas t
 he future solution to all problems. However\, systems have become\nso comp
 lex that it is no longer humanly comprehensibly\, how an\nalgorithm arrive
 s at an answer\, see for example "AAAS: Machine\nlearning 'causing science
  crisis'"\n(https://www.bbc.co.uk/news/science-environment-47267081)\n\nIn
  some cases\, companies refuse to disclose the proprietary algorithm.\nThi
 s has lead to controversies such as the COMPAS algorithm giving\nscores on
  the likelihood to re-offend. The organisation ProPublica\nclaims that the
  software exhibits racial bias\n(https://www.propublica.org/article/how-we
 -analyzed-the-compas-recidivism-algorithm)\nwhich the company disputes\n(h
 ttp://go.volarisgroup.com/rs/430-MBX-989/images/ProPublica_Commentary_Fina
 l_070616.pdf).\n\nAnother example is Amazon's gender bias recruitment tool
 \n(https://www.bbc.co.uk/news/technology-45809919). Partly to blame is\nth
 e data used to train algorithms. If the data is biased\, then the\nalgorit
 hm will be. More seriously\, it might exacerbate the bias\, since\nalgorit
 hms distill the essential distinguishing features. If these are\nthen high
 ly correlated with black - white\, male - female\, we have a\nproblem.\n\n
 While humans can also have bias\, they are also capable of realizing\nthei
 r world view is too simplistic. The talk presents work in progress\nof inc
 reasing the complexity of a model\, if the data suggests more\nfeatures ar
 e necessary to model the data. This approach aides to\nunderstand the "bla
 ck magic" inside the "black box".
LOCATION:JJ Thomson Seminar Room\, Maxwell Centre\, Cavendish Laboratory
END:VEVENT
END:VCALENDAR