Fuzzy test system for computing framework

Abstract

The invention discloses a fuzzy test system for a computing framework, and relates to the technical field of data processing, the fuzzy test system comprises a computer, two storage devices are installed on the computer, a data computing framework is stored in one storage device, and a fuzzy test system is stored in the other storage device. The fuzzy test system comprises a fuzzy test module and a later repair simulation module, the two storage devices store the data calculation frame and the fuzzy test system separately, the situation that the data calculation frame and the fuzzy test system collapse at the same time when accidents occur in the fuzzy test process is avoided, the working burden of a computer is relieved, and the working efficiency is improved. And a fuzzy test model in the fuzzy test execution module performs fuzzy test on the neural network model by using some data randomly selected in the data set.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a fuzzy testing system for a computing framework. Background Art

[0002] A computing framework is a basic software development framework that aims to build the underlying library of machine learning models, thereby simplifying and accelerating the development and deployment of machine learning. Among them, intelligent computing frameworks are used to support the training and execution of machine learning models, such as the TensorFlow framework (a symbolic mathematical system) and the PyTorch framework (an open source Python machine learning library). They can provide computing resources and algorithm libraries to help users quickly build, train, and deploy machine learning models.

[0003] In order to detect vulnerabilities and anomalies in intelligent computing frameworks, fuzz testing is often required for intelligent computing frameworks. Fuzz testing is an automated software testing method that injects illegal, malformed, or unexpected inputs into software to reveal software defects and vulnerabilities.

[0004] However, when the fuzz tester performs fuzz testing on the computing framework, since the computing framework is processed by multiple neural network models of different types of data, the fuzz test model will process a large amount of invalid data of different types at the same time during the fuzz testing process, causing the fuzz test model to work at high power for a long time, which may easily cause the fuzz tester to overload or even crash, affecting the service life of the fuzz tester. Summary of the invention

[0005] The purpose of the present invention is to provide a fuzzy testing system for a computing framework to solve the problems raised in the prior art.

[0006] To achieve the above-mentioned purpose, the present invention provides the following technical solutions: a fuzzy testing system for a computing framework, wherein two storage devices are installed on the computer, one of the storage devices stores a data computing framework, and the other storage device stores a fuzzy testing system, wherein the fuzzy testing system includes a fuzzy testing module and a post-patch simulation module, wherein the fuzzy testing module includes a test model data acquisition module, a test target locking, a fuzzy testing execution module, a test data storage module, a status analysis module, a report generation module, a network data acquisition module, a work simulation module, a data proofreading module, a seed interception module, a hazard rating unit, a partition control module and a virus detection module, and the fuzzy testing execution module includes multiple fuzzy testing models and multiple status monitoring modules.

[0007] Preferably, the multiple state monitoring modules correspond one-to-one to the multiple fuzzy testing models.

[0008] Preferably, the computer includes a human-computer interaction structure and a communication structure, and the communication structure includes a line connection system and a network communication protocol.

[0009] Preferably, the test model data acquisition module obtains the neural network model used for data calculation and the data set matched by the neural network model from the data calculation framework, and the test target is locked to the analysis and judgment of the test instruction.

[0010] Preferably, the fuzzy test execution module is used to perform fuzzy testing of the neural network model, the test data storage module stores test parameters, the state analysis module collects and analyzes the state data collected by the state monitoring module, and the report generation module generates a summary report based on the results.

[0011] Preferably, the network data acquisition module acquires data with lower security from the network, and the seed interception module compresses the network data that may affect the neural network model during the fuzzy test model testing process.

[0012] Preferably, the post-repair simulation module includes a framework damage simulation module, a damage determination module, a database retrieval module, a model variation module and a comprehensive fuzzy testing model.

[0013] Preferably, the comprehensive fuzzy testing model performs fuzzy testing on the data computing framework as a whole, and the damage determination module determines the existence of vulnerabilities and vulnerability risk levels in the data computing framework according to the changing state of the data computing framework during the fuzzy test.

[0014] Preferably, the database retrieval module obtains the vulnerability resolution data model from the database, and the model mutation module performs layer replacement, layer addition and layer deletion processing on the data calculation framework.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. In the present application, two storage devices store the data computing framework and the fuzz testing system separately to avoid the simultaneous crash of the data computing framework and the fuzz testing system when an accident occurs during the fuzz testing process, thereby reducing the workload of the computer. The fuzz testing model in the fuzz testing execution module performs fuzz testing on the neural network model with some data randomly selected from the data set. Multiple state monitoring modules are used to monitor the state changes of the neural network model when the fuzz testing model tests the neural network model. The multiple state monitoring modules correspond one-to-one to the multiple fuzz testing models. The multiple neural network models are fuzz tested separately to reduce the amount of data processed by the fuzz testing model at one time. The fuzz testing model can be specifically constructed and trained according to the type of parameter model, thereby improving the fuzz testing speed of the data computing framework, avoiding the crash of the traditional fuzz testing model when processing a large amount of data of different types at one time, and reducing the workload of the computer.

[0016] 2. In the present application, the fuzzy testing model performs fuzzy testing on the neural network model with test parameters, and the test parameters that have not been processed by the neural network model but are within the processing range can be involved in the fuzzy testing, and the neural network model can be comprehensively fuzz tested. The network data with lower security can be involved in the fuzzy testing of the neural network model by the fuzzy testing model, and the ability of the neural network model to process the lower security data obtained on the network can be tested, thereby improving the comprehensiveness of the fuzzy testing of the neural network model. The environmental simulation data can be involved in the fuzzy testing of the neural network model by the fuzzy testing model, and the state of the neural network model processing data from different sources can be fuzzily tested, so as to make the parameters used in the fuzzy testing process of the neural network model as close as possible to the state encountered in the actual work of the data computing framework.

[0017] 3. In the present application, the fuzz testing module causes damage to the neural network model during the fuzz testing process of the neural network model, that is, the test parameters and network data that detect the vulnerabilities in the neural network model are transmitted to the later patching simulation module, and the test parameters and network data participate in the fuzz testing process of the data computing framework by the comprehensive fuzz testing model, and the data computing framework is fuzzified in a targeted manner, which reduces the test time investment and directly allows the data that effectively detects the vulnerabilities to participate in the fuzz testing of the computing framework, which can reduce the processing of invalid data and the amount of data processed by the comprehensive fuzz testing model. It can intuitively test the neural network model with vulnerabilities in the fuzz testing of the data computing framework and the impact of the vulnerabilities in the neural network model on the overall data computing framework, as well as whether it is part of a large vulnerability in the data computing framework, and conduct a comprehensive test of the data computing framework. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 A schematic diagram of the structure of a fuzzy testing system for a computing framework according to the present invention; Figure 2 A schematic diagram of a communication structure of a fuzzy testing system for a computing framework according to the present invention; Figure 3 It is a structural schematic diagram of a fuzzy testing module of a fuzzy testing system for a computing framework according to the present invention; Figure 4 A schematic diagram of the structure of a fuzzy test execution module of a fuzzy test system for a computing framework according to the present invention; Figure 5 The present invention is a structural schematic diagram of a post-patching simulation module of a fuzzy testing system for a computing framework.

[0019] Numbers in the figure: 1. Computer; 2. Storage device; 3. Data calculation framework; 4. Human-computer interaction structure; 5. Communication structure; 501. Line connection system; 502. Network communication protocol; 6. Fuzzy test system; 7. Fuzzy test module; 701. Test model data acquisition module; 702. Test target locking; 703. Fuzzy test execution module; 7031. Fuzzy test model; 7032. Status monitoring module; 704. Test data storage module; 705. Status analysis module; 706. Report generation module; 707. Network data acquisition module; 708. Work simulation module; 709. Data proofreading module; 710. Seed interception module; 711. Hazard rating unit; 712. Partition control module; 713. Virus detection module; 8. C; 801. Framework damage simulation module; 802. Damage judgment module; 803. Database retrieval module; 804. Model mutation module; 805. Comprehensive fuzzy test model. DETAILED DESCRIPTION

[0020] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0021] Example: Figure 1-Figure 5 As shown, the present invention provides a technical solution for a fuzzy testing system for a computing framework. Two storage devices 2 are installed on a computer 1, one of which stores a data computing framework 3, and the other storage device 2 stores a fuzzy testing system 6. The fuzzy testing system 6 includes a fuzzy testing module 7 and a post-patching simulation module 8. The fuzzy testing module 7 is composed of a test model data acquisition module 701, a test target locking 702, a fuzzy testing execution module 703, a test data storage module 704, a state analysis module 705, a report generation module 706, a network data acquisition module 707, a work simulation module 708, a data proofreading module 709, a seed interception module 710, a hazard rating unit 711, a partition control module 712 and a virus killing module 713. The fuzzy testing execution module 703 includes multiple fuzzy testing models 7031 and multiple state monitoring modules 7032. Use two storage devices 2 to store the data computing framework 3 and the fuzzy testing system 6 separately, so as to avoid the situation that the data computing framework 3 and the fuzzy testing system 6 crash at the same time when an accident occurs during the fuzzy testing process, thereby reducing the workload of the computer 1; A human-computer interaction structure 4 and a communication structure 5 are installed on the computer 1. The human-computer interaction structure 4 meets the needs of the staff for inputting instructions, inputting data, plugging in storage, viewing data, etc.; the communication structure 5 is composed of two major parts: a line connection system 501 and a network communication protocol 502. The communication structure 5 meets the needs of data transmission, command reception, wireless communication, hardware control, etc. between the two storage devices 2, the human-computer interaction structure 4 and the storage device 2, the computer 1 and the network, etc.; In the data computing framework 3, multiple neural network models are used to compute multiple types of data; The fuzzy testing module 7 in the fuzzy testing system 6 performs fuzzy testing on the data computing framework 3 to detect whether there is a vulnerability; The test model data acquisition module 701 in the fuzzy test module 7 acquires the neural network model used for data calculation and the data set matched by the neural network model from the data calculation framework 3 through the communication structure 5; In the fuzzy test module 7, the test target lock 702 analyzes and judges the test instruction, and locks a certain neural network model that needs to be tested; The fuzzy test execution module 703 in the fuzzy test module 7 is used to execute the fuzzy test work of the neural network model; The fuzzy test execution module 703 is composed of multiple fuzzy test models 7031 and multiple state monitoring modules 7032. The multiple fuzzy test models 7031 satisfy the simultaneous testing of multiple neural network models. The multiple state monitoring modules 7032 correspond to the multiple fuzzy test models 7031 one by one. The multiple state monitoring modules 7032 are used to monitor the state changes of the neural network model during the fuzzy test. The test data storage module 704 in the fuzzy test module 7 is used to store the test parameters required by the fuzzy test execution module 703 during the test process; The state analysis module 705 in the fuzzy testing module 7 collects and analyzes the state data collected by the state monitoring module 7032; The report generation module 706 in the fuzzy testing module 7 generates a summary report based on the analysis results of the state analysis module 705; The network data acquisition module 707 in the fuzzy test module 7 acquires multiple data of the same type from the network according to the data type that the test neural network model is responsible for calculating; The working simulation module 708 in the fuzzy test module 7 generates environmental simulation data according to the network data acquired by the network data acquisition module 707 and the test parameters stored in the test data storage module 704; The data proofreading module 709 in the fuzzy test module 7 performs comparative analysis on the data processed by the neural network model in the fuzzy test; The seed interception module 710 in the fuzzy test module 7 compresses the network data that affects the neural network model during the test of the fuzzy test model 7031 to form test parameters; The hazard rating unit 711 in the fuzzy testing module 7 is used to judge the level of damage caused by the test parameters and network data to the neural network model, that is, the vulnerability hazard level; The isolation control module 712 in the fuzzy test module 7 isolates the fuzzy test model 7031 for testing the remaining neural network models when the neural network model is destroyed; The virus detection module 713 in the fuzzy test module 7 performs a comprehensive virus and Trojan detection after the fuzzy test is completed.

[0022] The data calculation framework 3 to be tested is stored in the computer 1 through the human-computer interaction structure 4. When one or more of the multiple neural network models need to be fuzzy tested, the test instructions are input through the human-computer interaction structure 4, and the fuzzy test system 6 stored in another storage device 2 works according to the test instructions; After inputting the keyword of the type data that the neural network model is responsible for calculating on the fuzzy test module 7, the human-computer interaction structure 4 generates a test instruction and transmits it to the fuzzy test system 6. The test target lock 702 analyzes and judges the test instruction. After locking the neural network model to be tested, the test model data acquisition module 701 obtains the neural network model and the data set matched by the neural network model from the data calculation framework 3 through the communication structure 5. Then, the fuzzy test model 7031 in the fuzzy test execution module 703 performs a fuzzy test on the neural network model with some data randomly selected from the data set. Multiple state monitoring modules 7032 are used to monitor the state changes of the neural network model when the fuzzy test model 7031 tests the neural network model, and the multiple state monitoring modules 7032 correspond to the multiple fuzzy test models 7031 one by one, and the multiple neural network models are subjected to separate fuzzy tests, thereby reducing the amount of data processed by the fuzzy test model 7031 at one time, and the fuzzy test model 7031 can be targetedly constructed and trained according to the type of parameter model, thereby improving the fuzzy test speed of the data calculation framework 3, avoiding the situation where the traditional fuzzy test model 7031 crashes when processing a large amount of data of different types at one time, and reducing the workload of the computer 1; When some data randomly selected from the data set obtained from the data calculation framework 3 are used up, a part of the test parameters of the same type as the data calculated by the neural network model under test stored in the test data storage module 704 is transmitted to the corresponding fuzzy test model 7031 in the fuzzy test execution module 703. The fuzzy test model 7031 performs fuzzy testing on the neural network model with the test parameters, so that the test parameters that have not been processed by the neural network model but are within the processing range can participate in the fuzzy test, and a comprehensive fuzzy test is performed on the neural network model. After the fuzzy test work involving the test parameters stored in the test data storage module 704 is completed, the network data acquisition module 707 acquires multiple data of the same type from the network according to the data type that the test neural network model is responsible for calculating, and the network data is transmitted to the fuzzy test execution module 703, so that the network data with lower security participates in the fuzzy test of the neural network model by the fuzzy test model 7031, and the ability of the neural network model to process the data with lower security acquired on the network is tested, thereby improving the comprehensiveness of the fuzzy test of the neural network model; At the same time, the working simulation module 708 generates environmental simulation data according to the network data acquired by the network data acquisition module 707 and the test parameters stored in the test data storage module 704, and the environmental simulation data is transmitted to the fuzzy test execution module 703. The environmental simulation data participates in the fuzzy test of the neural network model by the fuzzy test model 7031, and can fuzzily test the state of the neural network model processing data from different sources, and try to make the parameters used in the fuzzy test of the neural network model close to the state encountered in the actual work of the data calculation framework 3; when the state analysis module 705 determines that the neural network model is damaged during the fuzzy test and shows virus invasion, the partition control module 712 will isolate all fuzzy test models 7031 except the fuzzy test model 7031 that tests the neural network model, so as to ensure the safety of the fuzzy test module 7 and reduce the impact on the fuzzy test of the remaining fuzzy test models 7031.

[0023] In addition, the data proofreading module 709 in the fuzzy testing module 7 compares and analyzes the processed data of the neural network model in the fuzzy test, and provides parameters for determining whether the neural network model is working normally. The state analysis module 705 collects and analyzes the state data collected by the state monitoring module 7032. The hazard rating unit 711 is used to judge the level of damage caused to the neural network model by the test parameters and network data, and to determine the level of vulnerabilities in the neural network model. The report generation module 706 generates a summary report based on the analysis results of the state analysis module 705, the test parameters involved in the fuzzy test, the parameters provided by the data proofreading module 709, the level of damage caused, and other data. The summary report is displayed on the human-computer interaction structure 4.

[0024] In addition, the seed interception module 710 will compress the network data that affects the neural network model during the fuzzy test to form new test parameters. The new test parameters are stored in the test data storage module 704 to enrich the data stored in the fuzzy test module 7 to ensure the pertinence and effectiveness of the fuzzy test. The virus detection module 713 in the fuzzy test module 7 will perform a comprehensive virus and Trojan detection after the fuzzy test is completed to avoid the situation where the network data causes the virus to hide inside the fuzzy test module 7.

[0025] In addition, the fuzzy test system 6 is also provided with a post-repair simulation module 8, in which the framework damage simulation module 801 in the post-repair simulation module 8 stores a comprehensive fuzzy test model 805, and the comprehensive fuzzy test model 805 performs fuzzy testing on the data calculation framework 3 as a whole; The damage determination module 802 in the later repair simulation module 8 determines the existence of vulnerabilities and vulnerability risk levels of the data computing framework 3 according to the change state of the data computing framework 3 in the fuzzy test; The database retrieval module 803 in the later repair simulation module 8 obtains the vulnerability solution data model from the database according to the judgment result of the damage judgment module 802; The model variation module 804 in the post-repair simulation module 8 performs layer replacement, layer addition and layer deletion processing on the data calculation framework 3; When the fuzzy test module 7 completes the fuzzy test of the neural network model that needs to be tested in the data computing framework 3, the post-patching simulation module 8 starts working. The fuzzy test module 7 damages the neural network model during the fuzzy test of the neural network model, that is, the test parameters and network data that detect the vulnerabilities in the neural network model are transmitted to the post-patching simulation module 8. The test parameters and network data participate in the fuzzy test of the data computing framework 3 by the comprehensive fuzzy test model 805. The data computing framework 3 is fuzzy tested in a targeted manner, which reduces the test time investment. At the same time, the data that effectively detects the vulnerabilities directly participates in the fuzzy test of the data computing framework 3, which can reduce the processing of invalid data and the amount of data processed by the comprehensive fuzzy test model 805. It can intuitively test the neural network model with vulnerabilities in the fuzzy test of the data computing framework 3 and the impact of the vulnerabilities in the neural network model on the data computing framework 3 as a whole, and whether it is part of a large vulnerability in the data computing framework 3, and conduct a comprehensive test on the data computing framework 3. Subsequently, the database retrieval module 803 obtains the same vulnerability resolution data model from the database according to the judgment result of the damage judgment module 802. The model mutation module 804 performs mutation processing on the data computing framework 3 by layer replacement, layer addition and layer deletion with reference to the vulnerability resolution data model. After the model mutation module 804 records the mutation steps and effects of the data computing framework 3, it generates a guidance report through the report generation module 706 and displays it on the human-computer interaction structure 4, so as to provide assistance for the staff to repair the data computing framework 3.

[0026] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

Claims

1. A fuzzy testing system for a computing framework, comprising a computer (1), characterized in that: The computer (1) is installed with two storage devices (2), one of which stores a data computing framework (3) internally, and the other storage device (2) stores a fuzzy test system (6) internally, the fuzzy test system (6) comprising a fuzzy test module (7) and a post-patching simulation module (8); the fuzzy test module (7) comprises a test model data acquisition module (701), a test target locking (702), a fuzzy test execution module (703), a test data storage module (704), a state analysis module (705), a report generation module (706), a network data acquisition module (707), a work simulation module (708), a data proofreading module (709), a seed interception module (710), a hazard rating unit (711), a partition control module (712) and a virus detection module (713); the fuzzy test execution module (703) comprises a plurality of fuzzy test models (7031) and a plurality of state monitoring modules (7032).

2. A fuzzy testing system for a computing framework according to claim 1, characterized in that: The multiple state monitoring modules (7032) and the multiple fuzzy test models (7031) correspond one to one.

3. A fuzzy testing system for a computing framework according to claim 1, characterized in that: The computer (1) comprises a human-computer interaction structure (4) and a communication structure (5), wherein the communication structure (5) comprises a line connection system (501) and a network communication protocol (502).

4. A fuzzy testing system for a computing framework according to claim 1, characterized in that: The test model data acquisition module (701) acquires a neural network model used for data calculation and a data set matched with the neural network model from a data calculation framework (3), and the test target locking (702) parses and judges the test instructions.

5. The fuzzy testing system for a computing framework according to claim 1, characterized in that: The fuzzy test execution module (703) is used to execute the fuzzy test of the neural network model, the test data storage module (704) stores the test parameters, the state analysis module (705) collects and analyzes the state data collected by the state monitoring module (7032), and the report generation module (706) generates a summary report based on the results.

6. A fuzzy testing system for a computing framework according to claim 1, characterized in that: The network data acquisition module (707) acquires data with lower security from the network, and the seed interception module (710) compresses the network data that affects the neural network model during the testing process of the fuzzy test model (7031).

7. A fuzzy testing system for a computing framework according to claim 1, characterized in that: The post-repair simulation module (8) comprises a framework damage simulation module (801), a damage determination module (802), a database retrieval module (803), a model variation module (804) and a comprehensive fuzzy test model (805).

8. A fuzzy testing system for a computing framework according to claim 7, characterized in that: The comprehensive fuzzy test model (805) performs fuzzy testing on the data computing framework (3) as a whole, and the damage judgment module (802) judges whether the data computing framework (3) has vulnerabilities and the vulnerability risk level according to the change state of the data computing framework (3) during the fuzzy test.

9. A fuzzy testing system for a computing framework according to claim 7, characterized in that: The database retrieval module (803) obtains the vulnerability solution data model from the database, and the model variation module (804) performs layer replacement, layer addition and layer deletion processing on the data calculation framework (3).

Images