?url_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Adc&rft.title=Differentially+Private+Mixture+of+Generative+Neural+Networks&rft.creator=Acs%2C+G&rft.creator=Melis%2C+L&rft.creator=Castelluccia%2C+C&rft.creator=De+Cristofaro%2C+E&rft.description=Generative+models+are+used+in+an+increasing+number%0D%0Aof+applications+that+rely+on+large+amounts+of+contextually%0D%0Arich+information+about+individuals.+Owing+to+possible+privacy%0D%0Aviolations%2C+however%2C+publishing+or+sharing+generative+models+is%0D%0Anot+always+viable.+In+this+paper%2C+we+introduce+a+novel+solution+for%0D%0Aprivately+releasing+generative+models+and+entire+high-dimensional%0D%0Adatasets+produced+by+these+models.+We+model+the+generator%0D%0Adistribution+of+the+training+data+by+a+mixture+of+k+generative%0D%0Aneural+networks.+These+are+trained+together+and+collectively+learn%0D%0Athe+generator+distribution+of+a+dataset.+Data+is+divided+into+k%0D%0Aclusters%2C+using+a+novel+differentially+private+kernel+k-means%2C+then%0D%0Aeach+cluster+is+given+to+separate+generative+neural+networks%2C+such%0D%0Aas+Restricted+Boltzmann+Machines+or+Variational+Autoencoders%2C%0D%0Awhich+are+trained+only+on+their+own+cluster+using+differentially%0D%0Aprivate+gradient+descent.+We+evaluate+our+approach+using+the%0D%0AMNIST+dataset+and+a+large+Call+Detail+Records+dataset%2C+showing%0D%0Athat+it+produces+realistic+synthetic+samples%2C+which+can+also+be+used%0D%0Ato+accurately+compute+arbitrary+number+of+counting+queries.&rft.subject=Data+models%2C+Kernel%2C+Data+privacy%2C+Privacy%2C+Neural+networks%2C+Training%2C+Standards&rft.publisher=IEEE&rft.contributor=Raghavan%2C+V&rft.contributor=Aluru%2C+S&rft.contributor=Karypis%2C+G&rft.contributor=Miele%2C+L&rft.contributor=Wu%2C+X&rft.date=2017-12-18&rft.type=Proceedings+paper&rft.language=eng&rft.source=+++++In%3A+Raghavan%2C+V+and+Aluru%2C+S+and+Karypis%2C+G+and+Miele%2C+L+and+Wu%2C+X%2C+(eds.)+Proceedings+of+the+17th+International+Workshop+on+Digital+Mammography%3A+10th+International+Workshop%2C+IWDM+2010%2C+Girona%2C+Catalonia%2C+Spain%2C+June+16-18%2C+2010.++(pp.+pp.+715-720).++IEEE%3A+Louisiana%2C+USA.+(2017)+++++&rft.format=text&rft.identifier=https%3A%2F%2Fdiscovery.ucl.ac.uk%2Fid%2Feprint%2F1574604%2F21%2F1709.04514v1.pdf&rft.identifier=https%3A%2F%2Fdiscovery.ucl.ac.uk%2Fid%2Feprint%2F1574604%2F&rft.rights=open