A decision flow model computing acceleration method based on fully homomorphic encryption

Abstract

The present invention proposes a decision flow model calculation acceleration method based on fully homomorphic encryption, including parsing the decision flow model, determining the calculation operations and memory access operations in the decision flow model; re-arranging the calculation operations, merging similar calculation operations to form a merged decision flow model, merging the operands of similar calculation operations obtained by the memory access operation to form a merged operand; performing non-homomorphic encryption on the merged decision flow model, performing fully homomorphic encryption on the merged operands, and then sending them to the cloud. Through this method, the speed of decision flow model calculation is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data privacy protection, and is a decision flow model calculation acceleration method based on fully homomorphic encryption. Background Art

[0002] In the practice of enterprise management, scientific decision-making is an important responsibility of modern managers, and letting data speak is the basis for scientific decision-making in enterprises. A common scenario is that several feasible plans are formulated, and the implementation of each plan may produce several results. There is a certain probability for the occurrence of various results. Enterprise decision-making has a certain chance of success, but also a certain risk.

[0003] Computer-aided decision-making technology has been widely used in corporate investment decisions. It has become a practical decision-making tool by expressing the different situations in each plan through mathematical modeling and then implementing them by computer, that is, the decision-making method is mathematized and modeled. For example, routine management decisions can be programmed using mathematical models and automated by computers to improve efficiency.

[0004] In recent years, the digital wave has been booming, and the status of data as a basic strategic resource and key production factor has become increasingly prominent. As the core element and key factor of digital transformation, the importance of data to corporate decision-making is self-evident. When computers make decisions, the amount of data they face is growing exponentially, and the calculation models are becoming more and more complex. With the introduction of technologies such as big data and artificial intelligence, the cost of decision-making calculations for enterprises is also increasing. With the rapid development of cloud computing technology, more and more users upload data to cloud servers to complete calculations, and cloud service providers also provide data processing services to users. Therefore, when making decision calculations, enterprises are more willing to send decision data and models to cloud service providers in the form of decision flow models, and perform calculations in the cloud to save costs and maximize the value of data.

[0005] However, when completing decision-making calculations through cloud computing, enterprises are most concerned about the increasing security issues such as user privacy leakage, business secret leakage, and network attacks. Enterprises do not want other enterprises and cloud service providers to obtain their business information in the cloud computing environment, so encryption and decryption are required during data transmission. Traditional decision flow computing uses a combination of symmetric and asymmetric encryption algorithms to ensure the security of the data transmission process. When ciphertext data is sent to the cloud computing platform for decision-making calculations, it needs to be decrypted first and calculated in plain text. After the calculation is completed, it is encrypted and transmitted back. Although the security of data transmission and storage is achieved, the security of the calculation process cannot be guaranteed.

[0006] A technology called fully homomorphic encryption can effectively solve this problem. Fully homomorphic encryption (FHE) is an encryption scheme that allows a third party (such as a cloud service provider) to perform certain computable functions on encrypted data while retaining the functional and format characteristics of the encrypted data. In fact, this fully homomorphic encryption corresponds to a mapping in abstract algebra. As an example of an additive HE scheme, for sample messages m1 and m2, f(m1+m2) can be obtained by using f(m1) and f(m2) without explicitly knowing m1 and m2, where f represents the fully homomorphic encryption function. Unfortunately, however, homomorphic encryption is relatively complex and has low encryption efficiency. How to improve the computational efficiency of the decision flow model of fully homomorphic encryption is an urgent problem to be solved. Summary of the invention

[0007] In response to the problem of low technical efficiency of the decision flow model with fully homomorphic encryption, the present invention provides a decision flow model calculation acceleration method based on fully homomorphic encryption, which solves the problems of data privacy protection and low technical calculation efficiency during cloud calculation of the decision flow model.

[0008] In a first aspect, the present invention provides a decision flow model calculation acceleration method based on fully homomorphic encryption, the method comprising:

[0009] Analyze the decision flow model to determine the calculation operations and memory access operations in the decision flow model;

[0010] Rearrange the computing operations, merge computing operations of the same type to form a merged decision flow model, and merge operands of computing operations of the same type obtained by the memory access operation to form a merged operand;

[0011] The merged decision flow model is non-homomorphically encrypted, the merged operands are fully homomorphically encrypted, and then sent to the cloud.

[0012] The second aspect of the present invention provides a decision flow model calculation acceleration device based on fully homomorphic encryption, including a processor and a readable storage medium, wherein the readable storage medium stores executable instructions that can be executed by the processor, and when the processor executes the executable instructions, it implements the decision flow model calculation acceleration method of the present invention.

[0013] Through the above method provided by the present invention, when using the decision flow model for cloud computing, the same or similar model is often used to calculate a batch of data. By re-arranging the calculation model and merging similar calculation operations, the number of fully homomorphic encryptions is greatly reduced. While achieving the privacy protection of the decision flow model cloud computing data, it also achieves the effect of improving the efficiency of fully homomorphic encryption calculations.

[0014] Compared with the prior art, the present invention has the following advantages:

[0015] (1) Compared with traditional decision flow model calculation, this method solves the risk of privacy leakage of original data in decision flow model calculation based on fully homomorphic encryption technology.

[0016] (2) High data security: the original data involved in arithmetic operations is in the form of homomorphic encryption ciphertext to ensure the privacy of the original data;

[0017] (3) High computational efficiency. It is used to adopt fully homomorphic encryption. There is no need to encrypt and decrypt data in the cloud, which improves processing efficiency. At the same time, on the user side, the efficiency of fully homomorphic encryption is improved because the operands are merged. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art are briefly introduced below.

[0019] Figure 1 A schematic diagram of an example of a decision flow model;

[0020] Figure 2 This is a schematic diagram of the decision flow model calculation acceleration method proposed in the present invention. DETAILED DESCRIPTION

[0021] In order to make the technical problems, technical solutions and advantages to be solved by the present invention more clear, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0022] Consider a classic decision flow model. A company has designed three feasible plans to increase the output of a certain product: one is to invest 800,000 yuan to build a new production workshop; the second is to invest 400,000 yuan to expand the old workshop; the third is to subcontract the production to other factories for a period of 5 years. The decision flow model can be used as follows: Figure 1 The decision tree model shown indicates that there is a 60% chance that the market will improve. In this case, the newly built production workshop can make a profit of 700,000 yuan per year, the expanded old workshop can make a profit of 500,000 yuan per year, and subcontracting the production to other factories can make a profit of 300,000 yuan per year. In addition, there is a 40% chance that the market will be bad. In this case, the newly built production workshop can lose 200,000 yuan per year, the expanded old workshop can make a profit of 200,000 yuan per year, and subcontracting the production to other factories can make a profit of 100,000 yuan per year.

[0023] Next, the expected profit is calculated according to each decision path. The decision calculation process is as follows:

[0024] E1 = (70 × 0.6-20 × 0.4) × 5-80 = 90 (ten thousand yuan)

[0025] E2 = (50 × 0.6 + 20 × 0.4) × 5-40 = 150 (ten thousand yuan)

[0026] E3 = (30 × 0.6 + 10 × 0.4) × 5 = 110 (ten thousand yuan)

[0027] It can be seen that the second decision path is the best.

[0028] For example, see Figure 2 ,When an enterprise performs decision calculations in the cloud, the following steps need to be completed.

[0029] Step 1: parse the decision flow model to determine the calculation operations and memory access operations in the decision flow model.

[0030] In the above example, there are three decision flow models, which are used to calculate E1, E2, and E3 respectively. This includes the memory access operations for various data in the three decision paths, as well as the determination of the calculation operations:

[0031] Expected E = (m1×p1+m2×p2))×YC,

[0032] Among them, m1 is the annualized profit and loss when the market is good, p1 is the probability of a good market, m2 is the annualized profit and loss when the market is bad, p2 is the probability of a bad market, Y is the number of years, and C is the input cost.

[0033] Step 2: Rearrange the computing operations, merge similar computing operations to form a merged decision flow model, and merge the operands of similar computing operations obtained by memory access operations to form a merged operand.

[0034] Since there are three paths, theoretically, three decision flow calculation models need to be sent to the cloud for calculating E1, E2, and E3 respectively. In this way, three decision flow models need to be sent:

[0035] E1=(m 11 ×p 11 +m 21 ×p 21 )×Y-C1

[0036] E2=(m 12 ×p 12 +m 22 ×p 22 )×Y-C2

[0037] E3=(m 13 ×p 13 +m 23 ×p 23 )×Y-C3

[0038] Among them, E1 is the expected benefit value of the first decision path, m 11is the annualized profit and loss of the first decision path when the market is good, p 11 is the probability of a favorable market in the first decision path, m 21 is the annualized profit and loss of the first decision path when the market is not good, p 21 is the probability of a bad market for the first decision path, C1 is the input cost of the first decision path; E2 is the expected value of the benefit of the second decision path, m 12 is the annualized profit and loss of the second decision path when the market is good, p 12 is the probability of a favorable market in the second decision path, m 22 is the annualized profit and loss of the second decision path when the market is bad, p 22 is the probability of a bad market for the second decision path, C2 is the input cost of the second decision path; E3 is the expected value of the benefit of the third decision path, m 13 is the annualized profit and loss of the third decision path when the market is good, p 13 is the probability of the third decision path market improving, m 23 is the annualized profit and loss of the third decision path when the market is bad, p 23 is the probability that the market is not good for the third decision path, and C3 is the investment cost of the third decision path.

[0039] It can be seen that it is necessary to transmit m 11 、m 12 、m 13 、p 11 、p 12 、p 13 、m 21 、m 22 、m 23 、p 21 、p 22 、p 23 , Y, C1, C2, C3, a total of 16 data, need to be fully homomorphic encrypted 16 times; the transmission decision flow model needs to be non-homomorphic encrypted 3 times. Therefore, when it is placed in the cloud for computing, the amount of encryption calculation required is relatively large.

[0040] The inventor of the present invention has found in long-term practical research that when the decision flow model is calculated in the cloud, it often uses the same or similar model to calculate a batch of data. Taking advantage of this feature, when a batch of data is fully homomorphically encrypted, similar calculation operations can be merged by re-arranging the calculation model. Taking the above decision calculation as an example, since E1, E2, and E3 use basically the same decision calculation model when performing decision calculation, they can be merged into E merge =([m 11 m 12 m 13 ]×[p 11 p 12 p13 ]+[m 21 m 22 m 23 ]×[p 21 p 22 p 23 ])×Y-[C1C2C3]. Among them, [m 11 m 12 m 13 ] is the plaintext m 11 、m 12 、m 13 The concatenated long plaintext, [p 11 p 12 p 13 ] is the plain text p 11 、p 12 、p 13 The concatenated long plaintext, [m 21 m 22 m 23 ] is the plaintext m 21 、m 22 、m 23 The concatenated long plaintext, [p 21 p 22 p 23 ] is the plain text p 21 、p 22 、p 23 The long plaintext obtained by splicing, [YYY] is the long plaintext obtained by splicing plaintext Y, Y, Y, [C1C2C3] is the long plaintext obtained by splicing plaintext C1, C2, C3. By splicing similar data, it can be found that the amount of data that needs full homomorphic encryption is reduced, only [m 11 m 12 m 13 ]、[p 11 p 12 p 13 ]、[m 21 m 22 m 23 ]、[p 21 p 22 p 23 ], [YYY], [C1C2C3], a total of 6 data, only need to be fully homomorphic encrypted 6 times; the transmission decision flow model only needs to be non-homomorphic encrypted once. Therefore, when it is placed in the cloud for computing, the amount of encryption calculation required is greatly reduced.

[0041] Step 3: Perform non-homomorphic encryption on the merged decision flow model, perform fully homomorphic encryption on the merged operands, and then send them to the cloud.

[0042] Taking the above decision calculation as an example, the merged decision flow model E merge =([m11 m 12 m 13 ]×[p 11 p 12 p 13 ]+[m 21 m 22 m 23 ]×[p 21 p 22 p 23 ])×Y-[C1C2C3] is non-homomorphically encrypted and then sent to the cloud. After decryption in the cloud, the decision calculation model can be obtained in the cloud. Then the combined operation number [m 11 m 12 m 13 ]、[p 11 p 12 p 13 ]、[m 21 m 22 m 23 ]、[p 21 p 22 p 23 ], [YYY], [C1C2C3] are fully homomorphically encrypted to obtain the fully homomorphically encrypted data f(m 11 m 12 m 13 )、f(p 11 p 12 p 13 )、f(m 21 m 22 m 23 )、f(p 21 p 22 p 23 ), f(YYY), f(C1C2C3), and send them to the cloud for cloud computing.

[0043] The decision flow model E obtained by decryption is used in the cloud merge =([m 11 m 12 m 13 ]×[p 11 p 12 p 13 ]+[m 21 m 22 m 23 ]×[p 21 p 22 p 23 ])×Y-[C1C2C3], calculate the above fully homomorphically encrypted data to obtain f(E merge ), then f(E merge ) is returned to the user end as a result, and the user end responds to f(E merge) Decrypt the fully homomorphic encryption to get the result E merge Then the user end passes the result E merge Split and obtain the calculation results E1, E2, and E3 of the three decision paths.

[0044] It should be noted that the specific method for merging operands of the same type is to directly connect multiple operands through n zeros, where n is a natural number. 11 =70, m 12 =50, m 13 =30, possible merging results [m 11 m 12 m 13 ]=70050030, or [m 11 m 12 m 13 ]=7000500030. This is because when the model is operated, a carry may occur, and 0 needs to be added during the merging to prevent the operation result from overflowing.

[0045] The specific value of n can be reserved according to the possible results of the operation. For example, for the addition operation m 11 +m 21 , m 12 +m 22 , the original operand m 11 =70, m 12 =50, m 21 =20, m 22 =,80, both are 2-digit numbers, so the addition operation may result in a carry, so the combined operand [m 11 m 12 ] needs to add 1 bit 0 as the result overflow protection bit, [m 11 m 12 ]=70050, similarly merge the operands [m 21 m 22 ]=20080, then the calculation result [m 11 m 12 ]+[m 21 m 22 ]=90130, which can be divided into three-digit numbers to obtain two addition results 90 and 130 respectively.

[0046] Similarly, for a multiplication operation, each operand is padded with 0s when the operands are merged, so that the number of digits of the operand after the 0-padded operation is the sum of the maximum number of digits of the operands before the merging operation and the maximum number of digits of the operands before the merging operation.

[0047] Through the above method provided by the present invention, the decision flow model is analyzed to determine the computing operations and memory access operations in the decision flow model for computing operations involved in the cloud computing process of the decision flow model. Then, the decision flow model is used to often use the same or similar models to calculate a batch of data in cloud computing. By rearranging the computing model, similar computing operations are merged, thereby greatly reducing the number of fully homomorphic encryption times. While solving the problem of data privacy protection in cloud computing of the decision flow model, the problem of low computing efficiency of fully homomorphic encryption is also solved.

[0048] It should be understood that each part of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system.

[0049] A person of ordinary skill in the art can understand that all or part of the steps of implementing the above-mentioned embodiment method can be completed by instructing related hardware (such as a processor) through a program, and the program can be stored in a computer-readable storage medium. When the program is executed, it implements one or a combination of the steps of the method embodiment.

[0050] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A decision flow model computing acceleration method based on fully homomorphic encryption, the method comprising: Parsing the decision flow model to determine the calculation operations and memory access operations in the decision flow model; Rearrange the computing operations, merge computing operations of the same type to form a merged decision flow model, and merge operands of computing operations of the same type obtained by the memory access operation to form a merged operand; The merged decision flow model is non-homomorphically encrypted, the merged operands are fully homomorphically encrypted, and then sent to the cloud; The merging of operands of similar computing operations obtained by the memory access operation to form a merged operand is specifically to directly connect multiple operand plaintexts through n zeros, where n is a natural number; When combining operands in addition operations, you need to add 1 bit of 0 as a protection bit for result overflow; When merging operands in a multiplication operation, add 0 to the front of each operand so that the number of digits of the operand after the 0-padded operand is the sum of the maximum number of digits of the operands before the merging operation and the maximum number of digits of the operands before the merging operation. The decision flow model obtained by decryption is used in the cloud to calculate the merged operands after the fully homomorphic encryption, obtain the fully homomorphic encryption calculation result, and return it to the user end.

2. The decision flow model computing acceleration method based on fully homomorphic encryption according to claim 1 is characterized in that When the decision flow model is calculated in the cloud, a batch of data is calculated using the same or similar model.

3. The decision flow model computing acceleration method based on fully homomorphic encryption according to claim 1 is characterized in that: n is the number of bits reserved for the possible results of the operation.

4. The decision flow model computing acceleration method based on fully homomorphic encryption according to claim 1 is characterized in that The method also includes the user end performing fully homomorphic decryption on the fully homomorphic encrypted calculation result to obtain a combined calculation result.

5. The decision flow model computing acceleration method based on fully homomorphic encryption according to claim 4 is characterized in that The method also includes the user end obtaining final calculation results of multiple decision flow models by dividing the combined calculation results.

6. A decision flow model computing acceleration device based on fully homomorphic encryption, characterized in that: The method comprises a processor and a readable storage medium, wherein the readable storage medium stores executable instructions that can be executed by the processor, and when the processor executes the executable instructions, the method according to any one of claims 1 to 5 is implemented.

Images