Terminal Apparatus Probabilistic Encryption for Privacy-Preserving Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning systems that utilize behavior data from terminal apparatuses face privacy concerns as they often require large data sets and computation, leading to potential user privacy breaches, especially when deterministic encryption schemes are vulnerable to frequency attacks.
Innovation Solution
A method where the terminal apparatus determines different ciphertexts for each data type of behavior data, using a deterministic encryption scheme to encrypt behavior data in a way that the same types of data are encrypted with different ciphertexts, preventing frequency attacks and maintaining user privacy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If deterministic encryption scheme is used to encrypt behavior data, then machine learning is possible and user privacy is protected, but the system becomes vulnerable to frequency attacks based on statistical data
Solution Approach 1:
The patent changes the encryption parameter from deterministic to probabilistic encryption. This allows the same plaintext to be encrypted into different ciphertexts, preventing frequency attacks while still enabling machine learning through encrypted data. The probabilistic encryption scheme transforms the system from being vulnerable to statistical analysis to being secure against such attacks.
Solution Approach 2:
The patent introduces an intermediary mechanism (probabilistic encryption) between the plaintext behavior data and the ciphertext. This intermediary layer ensures that even though encryption is applied, the statistical properties are preserved enough for machine learning while adding sufficient randomness to prevent frequency attacks.
2Reliability
If probabilistic encryption scheme is used to encrypt system calls, then user privacy is protected, but machine learning becomes impossible
Solution Approach 1:
The patent adjusts the parameters of probabilistic encryption to strike a balance between privacy protection and machine learning capability. By carefully controlling the probability distribution and encryption parameters, the system preserves enough statistical structure for machine learning while maintaining security against frequency attacks.
Solution Approach 2:
The patent applies different encryption strategies to different parts of the data or uses localized probabilistic transformations that preserve certain statistical properties needed for machine learning while providing privacy protection. This allows selective preservation of useful information while encrypting sensitive patterns.
3Productivity
If behavior data is collected and transmitted to server for machine learning, then meaningful models can be generated, but user privacy may be compromised
Solution Approach 1:
The patent applies encryption to behavior data before transmission to the server. This preliminary encryption action ensures that privacy is protected from the outset, while the encrypted data retains enough structure for the server to perform machine learning and generate meaningful models without accessing raw user behavior information.
Solution Approach 2:
The patent uses encrypted data as an intermediary form that enables server-side machine learning while protecting user privacy. The encryption acts as a mediator that allows computational processing and model generation without exposing sensitive user behavior information to the server or potential attackers.
Data Source
AI summary
A control method of a terminal apparatus is disclosed. A control method of a terminal apparatus comprises the steps of: determining the number of different ciphertexts, into which behavior data generated according to operation of the terminal apparatus by a user is to be encrypted, for each data type of the behavior data; generating ciphertexts by encrypting behavior data of an identical type in a unit of the determined number of the different ciphertexts so that the behavior data can be encrypted into different ciphertexts; transmitting the generated ciphertexts to an external server; when a model of the behavior data having been used to acquire learning on the basis of the transmitted ciphertexts is received, monitoring an operation of the terminal apparatus on the basis of the received model. Here, the model may be obtained through acquisition of learning according to at least one of a machine learning algorithm, a neural network algorithm, and a deep learning algorithm.