We build a global matrix of co-views for training Identify K-Neighbors based on the Similarity Matrix Predict favorite TV programs based on their ownĬonsider N users, M programs and binary ratingsīuild a co-views matrix C, where Cab is the number of The sum of two sketches results in the sketch of the union of Mapping a stream of values (of length T) into a matrix of size Intuition: Approximate statistics are acceptable inĪn estimate of an item’s frequency in a stream Traffic statistics for the Tor NetworkĪvailable tools for computing private statistics are Statistics about mass transport movementsģ. Recommender systems for online streaming servicesĢ. Large-scale collection of contextual information oftenĮssential to gather statistics, train machine learning models,ġ. Will I be sued for sharing customer data? Government to increase the volume, timelines,Īnd quality of cyber threat information shared Suspicious activities deliberately mimic normal behaviorīut, malevolent actors often use same resources How much understanding required from users? Most personalized medicine tests in < 1 secondĮncryption can’t guarantee security past 30-50 yrs (2) Testing facilities keep test specifics (“secret sauce”) (1) Individuals don’t disclose their entire genome Individuals remain in control of their genomeĪllow doctors/clinicians/labs to run genomic tests, whileĭisclosing the required minimum amount of information, i.e.: Chills and Thrills of Whole Genome Sequencing. Progress aggravates fears of discriminationĭe-identification and obfuscation are not effective : Private Substring/Pattern Matchingĭiscloses ethnicity, disease predispositions (including mental) : Private Set Intersection Cardinality-only Privately intersect sets – 2,000 items/secĪuthorized Private Set Intersection (APSI) Won IARPA APP challenge, basis for IARPA SPAR Works also for Authorized Private Set Intersection : extends to arbitrarily malicious adversaries : scales efficiently to very large setsįirst protocol with linear complexities and fast crypto )?Ĭlient needs to prove that inputs satisfyĪuthorizations issued by appropriate authorityĪuthorizations need to be verified implicitly.Quantify information disclosure from output of f( Only the required minimum amount of information should When parties with limited mutual trust willing or Protect personally identifiable informationĪctively enabling scenarios otherwise impossible w/o Increase privacy of users, groups, and/or organizations Privacy-preserving Information Sharing: Tools and Applications