Many areas are now experiencing data streams that contain privacy-sensitive information.Although the sharing and release of these data are of great commercial value,if these data are released directly,the private user...Many areas are now experiencing data streams that contain privacy-sensitive information.Although the sharing and release of these data are of great commercial value,if these data are released directly,the private user information in the data will be disclosed.Therefore,how to continuously generate publishable histograms(meeting privacy protection requirements)based on sliding data stream windows has become a critical issue,especially when sending data to an untrusted third party.Existing histogram publication methods are unsatisfactory in terms of time and storage costs,because they must cache all elements in the current sliding window(sW).Our work addresses this drawback by designing an efficient online histogram publication(EOHP)method for local differential privacy data streams.Specifically,in the EOHP method,the data collector first crafts a histogram of the current SW using an approximate counting method.Second,the data collector reduces the privacy budget by using the optimized budget absorption mechanism and adds appropriate noise to the approximate histogram,making it possible to publish the histogram while retaining satisfactory data utility.Extensive experimental results on two different real datasets show that the EOHP algorithm significantly reduces the time and storage costs and improves data utility compared to other existing algorithms.展开更多
基金supported by the Anhui Provincial Natural Science Foundation,China(Nos.2108085MF218 and 2022AH040052)the University Synergy Innovation Program of Anhui Province,China(No.GXXT-2023-021)+1 种基金the Key Program of the Natural Science Foundation of the Educational Commission of Anhui Province of China(No.2022AH050319)the National Natural Science Foundation of China(Nos.62172003 and 61402008)。
文摘Many areas are now experiencing data streams that contain privacy-sensitive information.Although the sharing and release of these data are of great commercial value,if these data are released directly,the private user information in the data will be disclosed.Therefore,how to continuously generate publishable histograms(meeting privacy protection requirements)based on sliding data stream windows has become a critical issue,especially when sending data to an untrusted third party.Existing histogram publication methods are unsatisfactory in terms of time and storage costs,because they must cache all elements in the current sliding window(sW).Our work addresses this drawback by designing an efficient online histogram publication(EOHP)method for local differential privacy data streams.Specifically,in the EOHP method,the data collector first crafts a histogram of the current SW using an approximate counting method.Second,the data collector reduces the privacy budget by using the optimized budget absorption mechanism and adds appropriate noise to the approximate histogram,making it possible to publish the histogram while retaining satisfactory data utility.Extensive experimental results on two different real datasets show that the EOHP algorithm significantly reduces the time and storage costs and improves data utility compared to other existing algorithms.
