Secure computing device, secure computing method, and program

The secure computing device simplifies secure computations by allowing data analysts to write plaintext scripts, embedding hyperparameters and training data into pre-defined templates, thus overcoming the complexity and limited functionality of existing frameworks, achieving high-speed and flexible secure computations.

WO2024241367A1PCT designated stage expired Publication Date: 2024-11-28NT T INC
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
PCT/JP2023/018743
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Filing Date
2023-05-19
Publication Date
2024-11-28

AI Technical Summary

Technical Problem

Existing secure computation frameworks require a high learning curve and limited functionality due to their distinct language interfaces and dependency on specific libraries, making them cumbersome for general data analysts to perform complex machine learning tasks.

Method used

A secure computing device with a machine learning method determination unit, template reading unit, hyperparameter embedding unit, and secure computing unit that allows data analysts to write plaintext scripts, determining the machine learning method, selecting and embedding hyperparameters and training data into pre-defined templates to generate secure calculation scripts, enabling high-speed and high-precision secure computations.

Benefits of technology

Enables data analysts to perform machine learning on secure computations with a simple plaintext script, providing high processing performance, an easy-to-use interface, and high extensibility without being tied to specific libraries or frameworks, reducing the complexity of writing extensive secure computation scripts.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure JP2023018743_28112024_PF_FP_ABST
    Figure JP2023018743_28112024_PF_FP_ABST
Patent Text Reader

Abstract

This secure computing device includes: a machine learning method discrimination unit that reads a plaintext script to discriminate a machine learning method described in the plaintext script; a template reading unit that selects and reads a template of a script for secure computing in which processing corresponding to the discriminated machine learning method is described; a hyperparameter embedding unit that generates a script for secure computing by acquiring a hyperparameter or data for learning described in the plaintext script and embedding the hyperparameter or the data for learning in a prescribed location of the template that has been read; and a secure computing unit that executes secure computing on the basis of the generated script for secure computing.
Need to check novelty before this filing date? Find Prior Art

Description

Secure computation device, secure computation method, and program

[0001] The present disclosure relates to a secure computing device, a secure computing method, and a program that reads a plaintext script and executes machine learning in a secure manner.

[0002] When writing machine learning using plaintext scripts, Python libraries such as scikit-learn and keras can be used. When writing machine learning using operations on secure computing, several frameworks using dedicated scripts have been proposed (Non-Patent Document 1).

[0003] In recent years, several libraries capable of handling machine learning over secure computation have also been proposed. Among them, Crypten (Non-Patent Document 2) is a secure computation AI library written using Pytorch, a plaintext machine learning library. Crypten can be written using an interface similar to Pytorch, so data analysts who have used Pytorch have a low barrier to mastery. A similar prior art, tf-encrypted (Non-Patent Document 3), is a secure computation AI library written using Tensorflow, a plaintext machine learning library, and is also thought to be relatively easy to master for data analysts who have used Tensorflow.

[0004] Keller, Marcel. "MP-SPDZ: A versatile framework for multi-party computation." Proceedings of the 2020 ACM SIGSAC conference on computer and communications security. 2020. Knott, Brian, et al. "Crypten: Secure multi-party computation meets machine learning." Advances in Neural Information Processing Systems 34 (2021): 4961-4973. Dahl, Morten, et al. "Private machine learning in tensorflow using secure computation." arXiv preprint arXiv:1810.08130 (2018).

[0005] Frameworks optimized for secure computation can provide high-speed, high-precision secure computation, but their language interfaces differ from the plaintext language interfaces used in general data analysis, making it difficult for general data analysts who are not familiar with secure computation to master them.In addition, secure computation frameworks often only provide basic operations, so when performing complex calculations such as machine learning, it is necessary to write the process from scratch by combining basic operations, which is significantly more time-consuming than plaintext machine learning libraries.

[0006] As mentioned above, secure computation machine learning libraries such as Crypten and tf-encrypted are likely to be easy to use if you have used Pytorch or Tensorflow libraries. However, because Crypten and tf-encrypted depend on specific libraries, they can only use machine learning methods within the scope of those libraries, which poses challenges in functionality and scalability. For example, because Pytorch and Tensorflow are deep learning libraries, only deep learning methods can be used with Crypten and tf-encrypted.

[0007] Therefore, an object of the present disclosure is to provide a highly scalable secure computing device that can execute machine learning in secure computation simply by writing a plaintext script.

[0008] The secure computation device of the present disclosure includes a machine learning method determination unit, a template reading unit, a hyperparameter embedding unit, and a secure computation unit.

[0009] The machine learning method determination unit reads the plaintext script and determines the machine learning method described in the plaintext script. The template reading unit selects and reads a template of a secure computation script in which processing corresponding to the determined machine learning method is described. The hyperparameter embedding unit acquires hyperparameters or learning data described in the plaintext script and embeds the hyperparameters or learning data in predetermined locations in the read template to generate a secure computation script. The secure computation unit performs secure computation based on the generated secure computation script.

[0010] The secure computing device of the present disclosure is highly scalable, as it can perform machine learning in a secure manner simply by writing a plaintext script.

[0011] Fig. 1 is a diagram showing an example in which a library summarizing mevalai processing is imported into a Python program. Fig. 2 is a block diagram showing the functional configuration of a secure computing device of Example 1. Fig. 3 is a flowchart showing the operation of the secure computing device of Example 1. Fig. 4 is a diagram showing an example of the functional configuration of a computer.

[0012] Hereinafter, embodiments of the present disclosure will be described in detail. Note that components having the same functions are assigned the same numbers, and redundant explanations will be omitted.

[0013] In general, scripts written using secure computing frameworks are longer than those written in plain text for similar processes. When writing complex processes such as machine learning, the scripts can reach thousands of lines.

[0014] However, when data analysts actually perform machine learning, they do not need to write thousands of lines of code each time; it is sufficient to specify the training data to be used for learning and prediction, and some hyperparameters required for learning, such as the number of times of learning.

[0015] In other words, most of the machine learning scripts in secure computing can be fixed processes, and if an interface is provided that allows learning data and hyperparameters to be freely specified externally, data analysts will not need to write large-scale scripts directly.

[0016] As such, when performing machine learning in secure computation, it is only necessary to rewrite a portion of a large program, so converting a program from one language to another does not require the implementation of something like an advanced transcompiler. A template describing the machine learning process is created in advance using a secure computation framework, and the parts related to the training data and hyperparameters are made replaceable. Then, by embedding the necessary information in the template based on a plaintext script written by a data analyst, a secure computation script that can easily execute the desired process can be generated. The initial template implementation must be performed by someone with some familiarity with the secure computation framework, but once implemented, it can be reused any number of times.

[0017] The secure computing device 1 of Example 1 is a device devised from the above-mentioned perspective, and performs secure machine learning based on a script written in plain text (plain text script) in the same way as when a data analyst uses a plain text machine learning library.

[0018] For example, in the example shown in Figure 1, a data analyst can import mevalai, a library that compiles machine learning processes in secure computation, into a Python program and use machine learning in secure computation. This usage is the same as using a general plaintext machine learning library.

[0019] The functional configuration of the secure computing device 1 of this embodiment will be described below with reference to Fig. 2. As shown in Fig. 2, the secure computing device 1 of this embodiment includes a machine learning method determination unit 11, a template reading unit 12, a hyperparameter embedding unit 13, and a secure computing unit 14. The operation of each component will be described below with reference to Fig. 3.

[0020] <Machine Learning Method Determining Unit 11> The machine learning method determining unit 11 reads the plain text script and determines the machine learning method described in the plain text script (S11).

[0021] <Template reading unit 12> The template reading unit 12 selects and reads a template of a secure computation script in which a process corresponding to the determined machine learning method is described (S12). Note that it is assumed that predetermined character strings (initial values) are input so that they can be rewritten later in the locations where specific values ​​of hyperparameters and learning data, which will be described later, should be input in the template.

[0022] <Hyperparameter Embedding Unit 13> The hyperparameter embedding unit 13 acquires the hyperparameters or learning data written in the plaintext script, and embeds the hyperparameters or the learning data in predetermined locations of the template into which the hyperparameters or the learning data have been read, thereby generating a secure computation script (S13).

[0023] <Secure Computation Unit 14> The secure computation unit 14 executes secure computation based on the generated secure computation script (S14).

[0024] With a secure computing device that includes the above functional configuration, data analysts can use a high-speed, high-precision secure computing framework simply by writing plaintext scripts, without having to write complex secure computing scripts themselves.

[0025] <Specific Operation Example> A specific operation example of the above four processing steps (S11, S12, S13, S14) will be described below.

[0026] For example, if the plain text script that has been read contains the description model.fit(x_test, y_test, batch_size=128, epoch=3), the machine learning method determination unit 11 determines from the type of model that the machine learning method is a serial Neural Network (NN) model (S11).

[0027] The template reading unit 12 selects and reads a process corresponding to the determined machine learning method (serial NN model), that is, a template of a serial NN script for secure computation prepared in advance (S12).

[0028] The hyperparameter embedding unit 13 obtains the mini-batch size and the number of learning epochs from batch_size=128 and epoch=3 in the plaintext script description model.fit(x_test, y_test, batch_size=128, epoch=3), and embeds these hyperparameters into character strings (initial values) corresponding to the mini-batch size and the number of epochs on the template, thereby generating a secure computation script (S13).

[0029] The secure computation unit 14 executes secure computation based on the generated secure computation script (S14).

[0030] <Example of Plain Text Script> For example, the plain text script can be a script written using Python.

[0031] <Example of a Template for a Script for Secure Computation> For example, a template for a script for secure computation can be a template in which machine learning processing is described in advance using MEVAL (Reference Non-Patent Document 1).

[0032] (Reference Non-Patent Document 1: Naoto Kiribuchi, Dai Igarashi, Hiroki Hamada, Ryo Kikuchi, "Programmable Secure Computation Library MEVAL3," SCIS, 2018.) The above examples can also be combined. For example, the plaintext script can be a script written using Python, and the secure computation script template can be a template in which machine learning processing is written in advance using MEVAL.

[0033] <Effects> The secure computing device 1 of this embodiment can convert a plaintext script into a script for secure computation, thereby achieving both an easy-to-use language interface such as a plaintext library and the high processing performance of a framework optimized for secure computation. Furthermore, since the secure computing device 1 of this embodiment does not depend on a specific plaintext library or secure computation framework, any library or framework can be used, and the functions can be easily expanded.

[0034] The secure computing device 1 of this embodiment has the following advantages by creating a template that describes machine learning processing in advance using a secure computing framework, obtaining information such as learning data and hyperparameters from a program written in plain text by a data analyst, and embedding the information in the template.

[0035] - High processing performance of a framework optimized for secure computation - An easy-to-use language interface such as a general plaintext library - High scalability that is not limited to specific libraries or frameworks <Additional Notes> The device disclosed herein may, for example, be a single hardware entity, including an input unit to which a keyboard or the like can be connected, an output unit to which an LCD display or the like can be connected, a communication unit to which a communication device (e.g., a communication cable) capable of communicating with an external device of the hardware entity can be connected, a CPU (which may also include a central processing unit, cache memory, registers, etc.), memory such as RAM and ROM, an external storage device such as a hard disk, and buses connecting these input unit, output unit, communication unit, CPU, RAM, ROM, and external storage device so as to enable data exchange. If necessary, the hardware entity may also be provided with a device (drive) capable of reading and writing to a recording medium such as a CD-ROM. A physical entity equipped with such hardware resources includes a general-purpose computer.

[0036] The external storage device of the hardware entity stores the programs required to realize the above-mentioned functions and the data required for processing these programs (the programs may be stored in a ROM, which is a read-only storage device, for example, instead of an external storage device). Data obtained by processing these programs is stored in RAM, the external storage device, etc. as appropriate.

[0037] In a hardware entity, each program stored in an external storage device (or ROM, etc.) and the data required to process each program are loaded into memory as needed, and interpreted, executed, and processed by the CPU as appropriate, resulting in the CPU realizing a predetermined function (each component represented as a unit, means, etc.).

[0038] The present disclosure is not limited to the above-described embodiments, and modifications can be made as appropriate without departing from the spirit of the present disclosure. Furthermore, the processes described in the above embodiments may not only be executed in chronological order according to the order described, but may also be executed in parallel or individually depending on the processing capabilities of the devices that execute the processes or as needed.

[0039] As described above, when the processing functions of the hardware entities (apparatuses of the present disclosure) described in the above embodiments are realized by a computer, the processing contents of the functions that the hardware entities should have are described by a program, and the processing functions of the hardware entities are realized on the computer by executing this program.

[0040] The various processes described above can be implemented by loading a program that executes each step of the above method into the recording unit 10020 of the computer 10000 shown in Figure 4, and operating the control unit 10010, input unit 10030, output unit 10040, etc.

[0041] The program describing the processing contents can be recorded on a computer-readable recording medium. Examples of computer-readable recording media include magnetic recording devices, optical disks, magneto-optical recording media, and semiconductor memories. Specifically, examples of magnetic recording devices include hard disk drives, flexible disks, and magnetic tapes; optical disks include DVDs (Digital Versatile Discs), DVD-RAMs (Random Access Memory), CD-ROMs (Compact Disc Read Only Memory), and CD-Rs (Recordable / Rewritable) / RWs; magneto-optical recording media include MOs (Magneto-Optical Discs), and semiconductor memories include EEP-ROMs (Electrically Erasable and Programmable-Read Only Memory).

[0042] The program may be distributed by, for example, selling, transferring, lending, etc. portable recording media such as DVDs and CD-ROMs on which the program is recorded. Furthermore, the program may be stored in a storage device of a server computer, and then transferred from the server computer to other computers via a network, thereby distributing the program.

[0043] A computer that executes such a program may first temporarily store the program recorded on a portable recording medium or transferred from a server computer in its own storage device. Then, when executing a process, the computer reads the program stored on its own recording medium and executes the process in accordance with the read program. Alternatively, the computer may read the program directly from a portable recording medium and execute the process in accordance with the program. Furthermore, the computer may execute the process in accordance with each program transferred from the server computer. Alternatively, the server computer may not transfer the program to the computer, but may instead execute the process through a so-called ASP (Application Service Provider) service, which realizes the processing function by issuing an execution instruction and obtaining the results. In this embodiment, the program includes information used for processing by a computer that is equivalent to a program (e.g., data that is not a direct instruction to the computer but has properties that define computer processing).

[0044] In addition, in this embodiment, a hardware entity is configured by executing a predetermined program on a computer, but at least a part of the processing contents may be realized by hardware.

Claims

1. A secure computation device including: a machine learning method discrimination unit that reads a plaintext script and discriminates a machine learning method described in the plaintext script; a template reading unit that selects and reads a template of a secure computation script in which processing corresponding to the discriminated machine learning method is described; a hyperparameter embedding unit that acquires hyperparameters or learning data described in the plaintext script and embeds the hyperparameters or the learning data in a predetermined location of the read template to generate a secure computation script; and a secure computation unit that performs secure computation based on the generated secure computation script.

2. The secure computing device according to claim 1, wherein the plaintext script is a script written using Python.

3. The secure computing device according to claim 1, wherein the template of the secure computation script is a template in which machine learning processing is described in advance using MEVAL.

4. A secure computing device according to claim 1, wherein the plaintext script is a script written using Python, and the template of the secure computing script is a template in which machine learning processing is written in advance using MEVAL.

5. A secure computation method executed by a secure computing device, comprising: a step of reading a plaintext script and determining a machine learning method described in the plaintext script; a step of selecting and reading a template of a secure computation script in which processing corresponding to the determined machine learning method is described; a step of acquiring hyperparameters or learning data described in the plaintext script and embedding the hyperparameters or the learning data in a predetermined location of the template into which the hyperparameters or the learning data have been read to generate a secure computation script; and a step of performing secure computation based on the generated secure computation script.

6. A program for causing a computer to function as a secure computing device according to any one of claims 1 to 4.

Citation Information

Cited By

  • Calculation method for machine readable digital uncertainty

    CN120447963A