Multi-correlation difference privacy matrix decomposing method in non-independent identical distribution environment

• Main Authors: WANG LI'E, LIU PENG, CHU HONGGUANG, LI XIANXIAN, FU XINGCHENG
• Format: Patent
• Language: Chinese; English
• Published: 06.03.2018
• Subjects: CALCULATING; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; PHYSICS

The invention discloses a multi-correlation difference privacy matrix decomposing method in a non-independent identical distribution environment. According to the method, other attribute multi-correlation of data is considered, a correlation target disturbance mechanism is utilized to introduce related natures of the data into a model target function simultaneously, and meanwhile, the safety and effectiveness of the prediction result are guaranteed. The method mainly comprises two parts of correlation noise matrix calculation in which a generated random noise matrix satisfies a predicting result and satisfies difference privacy under a non-independent identical distribution hypothesis and the correlation difference privacy matrix decomposition and training process in which other attributemulti-correlation is introduced and a random noise matrix is added. On the condition that the data privacy safety is guaranteed, the predication precision can be improved as much as possible to offsetthe precision loss brough