Privacy-preserving transformation of continuous data

A non-transitory computer-readable medium comprising a program that, when executed by a processor of an apparatus, is configured to cause the processor to:

• transmit, by the apparatus to an electronic device, a performance parameter associated with a machine learning analytics task;
• obtain, from the electronic device by the apparatus, a set of time-series user data that is compressed, wherein:
• the set of time-series user data is compressed using a compression parameter that is determined based on (i) the performance parameter based on a level of accuracy for the machine learning analytics task and (ii) a privacy parameter based on an input from a user associated with the set of time-series user data, wherein the set of time-series user data is compressed using the compression parameter, and
• the set of time-series user data is compressed based on a reduction of a quantity of dimensions of the set of time-series user data according to a matrix, wherein each element in the matrix is from a normal distribution;
• transform, by the apparatus, the compressed set of time-series user data as a function of the privacy parameter for differential privacy when the compressed set of time-series user data was not previously transformed by the electronic device;
• perform, by the apparatus, the machine learning analytics task on the transformed set of time-series user data;
• provide, by the apparatus to the user associated with the set of time-series user data, a service based on an output of the machine learning analytics task;
• and
• coordinate, by the apparatus, with the electronic device, the performance parameter, the privacy parameter, and the compression parameter, so that a quality of the service does not degrade and a level of privacy and the level of accuracy are maintained.

An apparatus for transforming data, said apparatus comprising:

• a transmitter configured to transmit, to an electronic device, a performance parameter associated with a machine learning analytics task;
• a receiver configured to obtain a set of time-series user data that is compressed, wherein:
• the set of time-series user data is compressed using a compression parameter that is determined based on (i) the performance parameter based on a level of accuracy for the machine learning analytics task and (ii) a privacy parameter based on an input from a user associated with the set of time-series user data, and
• the set of time-series user data is compressed based on a reduction of a quantity of dimensions of the set of time-series user data according to a matrix;
• a memory configured to store the set of time-series user data;
• and
• a processor configured to:
• transform the compressed set of time-series user data as a function of the privacy parameter for differential privacy when the compressed set of time-series user data was not previously transformed by the electronic device, wherein each element in the matrix is from a normal distribution,
• perform the machine learning analytics task on the transformed set of time-series user data,
• provide, to the user associated with the set of time-series user data, a service based on an output of the machine learning analytics task, and
• coordinate, by the apparatus, with the electronic device, the performance parameter, the privacy parameter, and the compression parameter, so that a quality of the service does not degrade and a level of privacy and the level of accuracy are maintained.