Intelligent Algorithm Testing Method, System, Device and Storage Medium

Abstract

The present invention discloses an intelligent algorithm testing method, system, device and storage medium, which are one-to-one corresponding solutions. In the solution: by integrating multi-granularity coverage indicators, not only the efficiency of intelligent algorithm testing is improved, but also the depth and breadth of testing are significantly expanded from multiple dimensions. In terms of test depth, by introducing a multi-level coverage indicator system, it is possible to deeply explore the internal behavior mechanism and complex decision-making logic of the model, revealing subtle problems and hidden defects that are difficult to capture by traditional methods; in terms of test breadth, it covers various types of network behaviors and scenarios, significantly enhancing the comprehensiveness of the test. In addition, based on multi-granularity coverage indicators, the complementarity of each coverage indicator can be fully utilized to generate test samples with both diversity and coverage. Through these test samples, the intelligent algorithm is helped to discover more potential erroneous behaviors or abnormal decisions, thereby optimizing the security and reliability of the model and providing stronger guarantees for practical applications.

Description

Technical Field

[0001] The present invention relates to the technical field of neural network testing, and in particular, to an intelligent algorithm testing method, system, device and storage medium. Background Art

[0002] As a core technology in the field of artificial intelligence, deep neural networks (DNNs) have been widely applied in many key applications such as image recognition, natural language processing, and autonomous driving. With their popularization in practical scenarios, the requirements for the reliability and security of models are increasing day by day. However, due to their highly non-linear and complex "black box" characteristics, it is difficult to fully analyze the internal decision-making mechanism of deep neural networks. This complexity not only increases the difficulty of understanding the behavior of the models, but also poses severe challenges to the model testing and verification work, further giving rise to specialized testing technologies for neural networks. The goal of neural network testing technology is to systematically evaluate the performance of models in complex input scenarios, timely discover potential defects, and improve the generalization ability and robustness of the models by optimizing the network structure and training strategies, so as to better meet the actual needs.

[0003] Neural network testing technology is committed to comprehensively evaluating the stability, accuracy, and interpretability of the functional performance of models. This technology can not only reveal the internal operating mechanism of neural networks under specific inputs, but also deeply analyze the response ability of the models to abnormal inputs or adversarial attacks. With the expansion of the complexity and application scenarios of deep learning models, traditional software testing methods are no longer able to meet their special needs, which has promoted researchers to develop new testing methods more suitable for intelligent algorithms. Among them, neuron coverage is one of the most widely used core testing means, and its concept is derived from code coverage in traditional software testing, focusing on evaluating the activation degree of neurons. By analyzing the activation states in the network, neuron coverage can not only guide the generation of test samples, but also help identify potential defective inputs. This technology plays a key role in improving the reliability of models and maintaining security-sensitive applications.

[0004] The current neural network coverage testing mainly focuses on two aspects: the design of coverage metrics and the generation of test samples based on coverage metrics. These two aspects complement each other and jointly promote the comprehensive testing and maintenance of deep learning models. In terms of the design of coverage metrics, researchers first proposed neuron coverage as a basic metric to measure whether test samples sufficiently explore the model behavior. With the development of technology, more refined coverage metrics have been proposed, such as path-based coverage, neuron combination activation coverage, and adequacy coverage for functional modules. These coverage metrics can not only comprehensively reflect the behavioral complexity inside the model but also be adapted to different types of neural network architectures. For example, DeepXplore, as the first white-box testing framework, first applied neuron coverage to neural network testing and automatically detected potential defects in neural network models by generating difference-induced inputs; DeepGauge (a network depth testing framework) further proposed multi-granularity coverage metrics, including k-value activation coverage and interval coverage, on this basis; and for the special needs of recurrent neural networks (RNNs), DeepCruiser (an automated testing method for neural network models, aiming to discover potential errors and vulnerabilities by exploring the decision boundaries of the model) developed more adapted coverage metrics, improving the testing ability for time-series models. These studies have significantly promoted the application of coverage testing in different architectures and laid a foundation for the reliability evaluation of neural networks.

[0005] In terms of test sample generation, coverage-guided fuzz testing technology has gradually become a research hotspot in the field of neural network testing. The core idea of ​​this type of technology is to generate new test samples while retaining the semantic consistency of the original samples by designing diverse sample mutation strategies to cover the behavioral areas that are not fully explored by the neural network. These generated samples can not only help reveal potential defects in the model, but also effectively expand the test space, providing important support for the robustness and reliability verification of the model. In specific implementation, fuzz testing frameworks usually use coverage indicators as feedback signals, and continuously improve the adaptability and coverage of test samples to the target model through iterative optimization of the generation process. Taking DeepHunter (a neural network testing framework based on fuzz testing, which improves the robustness and security of the model by generating diverse test cases) as an example, this framework generates new samples that meet the coverage requirements by introducing diverse mutation strategies, such as pixel perturbation, structural transformation, and semantic enhancement of input samples. At the same time, DeepHunter combines diversity and recency strategies to select test seeds to ensure that the generated samples not only have a wide coverage capability, but also maintain the representativeness and efficiency of the test sample set. Experimental results show that the test samples generated by DeepHunter are significantly better than traditional methods in terms of coverage improvement, defect detection and sample diversity. For example, in image classification tasks, DeepHunter can quickly locate the network's sensitivity to boundary samples, which are often difficult to find with traditional random generation methods. In addition, the coverage-based generation strategy can also help locate high-risk areas of adversarial samples in the model, thereby further guiding the improvement and optimization of the model.

[0006] There are still many deficiencies in the current work. Most of the existing coverage standards and their corresponding test sample generation techniques focus on traditional fully connected networks (FNNs) and convolutional networks (CNNs), mainly serving image-related tasks. Only a small number of coverage indicators have been applied to natural language processing tasks, and only preliminary tests have been conducted on recurrent neural networks (RNNs). Therefore, the applicability of these coverage indicators on more complex text tasks, especially large language models (LLMs), has not been verified, and related coverage testing methods are still an area that needs to be developed and studied.

[0007] In addition, existing single coverage metrics generally have limitations. Each coverage metric can only reflect a specific dimension of the neural network behavior and is difficult to comprehensively characterize the complexity of the model. Therefore, it cannot be used independently as a complete test metric. For example, neuron coverage may focus on the activation of a single node, while path coverage focuses on the exploration of certain paths in the network. However, neither of them can comprehensively reflect the functional behavior and potential defects of the network. This limitation directly leads to limited quality of test samples generated based on single coverage metrics, and may not effectively cover the key weak areas of the model.

[0008] Therefore, how to organically combine multiple granularity coverage metrics to make them complementary during the testing process is an important research direction.

[0009] In view of this, the present invention is specifically proposed. Summary of the Invention

[0010] The object of the present invention is to provide an intelligent algorithm testing method, system, device and storage medium, which can comprehensively utilize multiple granularity coverage metrics for testing, not only make up for the deficiencies of single coverage metrics, but also generate higher-quality test samples, improve the diversity and pertinence of test samples, and provide more comprehensive support for the reliability and security verification of intelligent algorithms.

[0011] The object of the present invention is achieved by the following technical solutions:

[0012] An intelligent algorithm testing method includes:

[0013] Step 1: Select a target intelligent model that conforms to the actual application scenario as the intelligent algorithm to be tested, collect the corresponding test data set, screen out test samples from the test data set as seeds to form a seed bank, and define multi-granularity neuron coverage metrics during testing;

[0014] Step 2: Calculate the comprehensive score of the seeds based on the comprehensive distance of the seeds on the multi-granularity neuron coverage metrics and the scheduling influence factor based on diversity and proximity, select seeds from the seed bank based on the comprehensive score of the seeds, generate new test samples through the metamorphic strategy, then input them into the intelligent algorithm to be tested for testing, and calculate the multi-granularity neuron coverage metrics;

[0015] Step 3: Combine the calculated multi-granularity neuron coverage metrics to screen out a part of new test samples;

[0016] Step 4: Add the screened new test samples to the seed bank and transfer to Step 2; continuously iterate until the set test termination condition is met.

[0017] An intelligent algorithm testing system includes:

[0018] An intelligent algorithm, test data selection, and metric definition unit, which is used to select a target intelligent model that conforms to the actual application scenario as the intelligent algorithm to be tested, collect the corresponding test data set, screen out test samples from the test data set as seeds to form a seed bank, and define the multi-granularity neuron coverage metric during testing;

[0019] A multi-granularity neuron coverage metric calculation unit, which is used to calculate the comprehensive score of the seeds based on the comprehensive distance of the seeds on the multi-granularity neuron coverage metric and the scheduling impact factors based on diversity and proximity, select seeds from the seed bank based on the comprehensive score of the seeds, generate new test samples through the metamorphic strategy, then input them into the intelligent algorithm to be tested for testing, and calculate the multi-granularity neuron coverage metric;

[0020] A test sample screening unit, which is used to screen out a part of new test samples in combination with the calculated multi-granularity neuron coverage metric;

[0021] An iterative testing unit, which is used to add the screened new test samples to the seed bank, and perform iterative testing through the multi-granularity neuron coverage metric calculation unit and test sample screening until the set test termination condition is met.

[0022] A processing device includes: one or more processors; a memory for storing one or more programs;

[0023] Wherein, when the one or more programs are executed by the one or more processors, the one or more processors are caused to implement the foregoing method.

[0024] A readable storage medium stores a computer program, and when the computer program is executed by a processor, the foregoing method is implemented.

[0025] As can be seen from the technical solution provided by the present invention above, by integrating multi-granularity coverage metrics, not only the efficiency of intelligent algorithm testing is improved, but also the depth and breadth of testing are significantly extended from multiple dimensions, bringing new ideas and methods for model verification in complex tasks. In terms of efficiency improvement, the present invention can quickly analyze and evaluate the coverage of the model, reduce redundant testing by accurately locating test gaps and uncovered areas, and improve the utilization rate of test resources. In terms of testing depth, by introducing a multi-level (multi-granularity) coverage metric system, the present invention can deeply explore the internal behavior mechanism and complex decision logic of the model, revealing subtle problems and hidden defects that are difficult to capture by traditional methods. In terms of testing breadth, it covers various types of network behaviors and scenarios, significantly enhancing the comprehensiveness of testing. In addition, based on the multi-granularity coverage metrics, the complementarity of each coverage metric can be fully utilized to generate test samples with both diversity and coverage rate. Through these test samples, it helps the intelligent algorithm discover more potential error behaviors or abnormal decisions, and then optimize the security and reliability of the model, providing stronger guarantees for practical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0027] Figure 1 It is a flowchart of an intelligent algorithm testing method provided by an embodiment of the present invention;

[0028] Figure 2 It is a schematic diagram of the intelligent algorithm coverage rate testing process provided by an embodiment of the present invention;

[0029] Figure 3 It is a schematic diagram of the high-quality sample generation process provided by an embodiment of the present invention;

[0030] Figure 4 It is a schematic diagram of an intelligent algorithm testing system provided by an embodiment of the present invention;

[0031] Figure 5 It is a schematic diagram of a processing device provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0032] The following clearly and completely describes the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0033] First, the following explanations are given for the terms that may be used in this article:

[0034] Descriptions with semantic meanings such as "including", "comprising", "containing", "having" or other similar ones should be interpreted as non-exclusive inclusion. For example, including a certain technical feature element (such as raw materials, components, ingredients, carriers, dosage forms, materials, dimensions, parts, components, mechanisms, devices, steps, processes, methods, reaction conditions, processing conditions, parameters, algorithms, signals, data, products or articles, etc.) should be interpreted as not only including the clearly listed certain technical feature element, but also including other technical feature elements well-known in the art that are not clearly listed.

[0035] The term "consisting of..." means excluding any technical feature element that is not clearly listed. If this term is used in a claim, this term will make the claim a closed type, so that it does not include technical feature elements other than the clearly listed technical feature elements, except for the related conventional impurities. If this term only appears in a sub-clause of a claim, then it only limits the elements clearly listed in that sub-clause, and the elements recorded in other sub-clauses are not excluded from the overall claim.

[0036] The following provides a detailed description of an intelligent algorithm testing method, system, device, and storage medium provided by the present invention. The content not described in detail in the embodiments of the present invention belongs to the prior art well-known to those of ordinary skill in the art. In the embodiments of the present invention, those not specified in specific conditions are carried out according to the conventional conditions in the art or the conditions recommended by the manufacturer. The reagents or instruments not specified in the production manufacturer in the embodiments of the present invention are all conventional products that can be obtained through commercial purchase.

[0037] Embodiment 1

[0038] The embodiment of the present invention provides an intelligent algorithm testing method, as Figure 1 shown, mainly including the following steps:

[0039] Step 1: Select the intelligent algorithm and test data, and define the multi-granularity neuron coverage index.

[0040] Before starting the test, it is necessary to prepare the model to be tested, as well as the test data set for testing, and initialize the seed library accordingly.

[0041] In an embodiment of the present invention, a target intelligent model that conforms to the actual application scenario is selected as the intelligent algorithm to be tested, and the corresponding test data set is collected. Test samples are screened from the test data set as seeds to form a seed bank.

[0042] Exemplarily, the target intelligent model may include: a convolutional neural network for image processing, a Transformer (transform neural network) model for natural language processing, etc. At the same time, a test data set covering the current application direction is prepared to ensure the diversity and representativeness of the data, providing a basis for subsequent coverage metric calculation and seed bank initialization.

[0043] In an embodiment of the present invention, the multi-granularity neuron coverage metric is defined to include: single neuron coverage metric, horizontal neuron combination coverage metric, and vertical neuron combination coverage metric. The specific calculation method will be introduced later.

[0044] Step 2: Calculate the comprehensive score of the seed based on the comprehensive distance of the seed on the multi-granularity neuron coverage metric and the scheduling influence factor based on diversity and proximity. Select seeds from the seed bank based on the comprehensive score of the seeds, generate new test samples through the metamorphosis strategy, and then input them into the intelligent algorithm to be tested for testing, and calculate the multi-granularity neuron coverage metric.

[0045] In an embodiment of the present invention, a seed selection strategy is provided, including:

[0046] Given a seed bank, where a single seed is denoted as s, which corresponds to a test sample, and given the target coverage area Z during the test process; let represent the set of semantic paths covered by the seed s, and any one of the paths is denoted as ; let represent the set of abstract semantic paths of the target coverage area Z, and any one of the paths is denoted as ; according to the similarity between the path and the path , calculate the scheduling influence factor of the seed s based on path similarity, and then combine the coverage rate of the seed s on each multi-granularity neuron coverage metric to calculate the comprehensive distance of the seed s on the multi-granularity neuron coverage metric;

[0047] Combine the pre-allocated minimum scheduling probability of the seed s and the scheduling times of the seed s to calculate the scheduling influence factor of the seed s based on diversity and proximity;

[0048] By calculating , obtain the comprehensive score of the seed s; the remaining seeds are calculated in the same way to obtain the comprehensive score;

[0049] Select a part of the seeds from the seed bank based on the comprehensive score of the seeds.

[0050] After that, for the selected seeds, use the metamorphosis strategy to generate new test samples and input them into the intelligent algorithm to be tested. Calculate the multi-granularity neuron coverage index according to the neuron activation situation in the intelligent algorithm. Among them, the multi-granularity neuron coverage index includes: single neuron coverage index (neuron coverage index), horizontal neuron combination coverage index (path coverage index), and vertical neuron combination coverage index (cross-layer coverage index). The single neuron coverage index is the coverage rate of the activated neurons. The horizontal neuron combination coverage index is the coverage rate of the combined activation of the triggered neurons in the same layer. The vertical neuron combination coverage index is the coverage rate of the combined activation of the triggered neurons in different layers.

[0051] (1) Single neuron coverage index.

[0052] The single neuron coverage index mainly analyzes the activation situation of each neuron in the intelligent algorithm under specific input data, and evaluates the fine-grained response ability of the intelligent algorithm to the input data. By counting the number of activated neurons, measure the coverage range of the test samples on the model behavior exploration, and reveal the basic behavior characteristics of the model. Specifically, it can be obtained by calculating the coverage rate of the activated neurons. The calculation method includes: calculating the ratio of the activated neurons to the total number of neurons, expressed as:

[0053] ;

[0054] Among them, is the ratio of the activated neurons to the total number of neurons, is the set composed of all neurons; is the set The number of elements of, that is, the total number of neurons; is the neuron The activation value of, is the threshold; is the indicator function. When the condition is satisfied, the output is 1, otherwise the output is 0.

[0055] (2) Horizontal neuron combination coverage index.

[0056] The horizontal neuron combination coverage index mainly examines the activation combination pattern between neurons in the same layer, and evaluates the coverage ability of the test data on the interaction behavior of single-layer neurons. By comparing the triggered and untriggered combinations, analyze the behavior breadth of the intelligent algorithm, and provide an optimization direction for the subsequent test sample generation. It is obtained by calculating the coverage rate of the combined activation of the triggered neurons in the same layer. The calculation method includes:

[0057] For each layer, count the number of activated neuron combinations separately, and after summarization, obtain the number of activated neuron combinations of all layers. Also, count the number of triggered activated neuron combinations to obtain the number of all triggered activated neuron combinations. Calculate the ratio of the number of all triggered activated neuron combinations to the number of activation combinations of all layers , which is used as the coverage rate of triggered activated neuron combinations in the same layer, and is expressed as:

[0058] ;

[0059] where, is the set of all triggered activated neuron combinations, is the set of activated neuron combinations of all layers, and the symbol represents the number of elements in the set.

[0060] (3) Vertical neuron combination coverage metric.

[0061] The vertical neuron combination coverage metric mainly analyzes the activation paths of cross-layer neurons, reveals the synergistic effects and dependencies between different layers. By evaluating the coverage of cross-layer paths, it discovers areas that have not been fully explored by the intelligent algorithm, providing a reference for optimizing the test strategy. It can be obtained from the coverage rate of triggered activated neuron combinations in different layers, and the calculation methods include:

[0062] Count the number of activated cross-layer neuron combinations, and after summarization, obtain the number of all activated cross-layer neuron combinations. Also, count the number of all triggered activated cross-layer neuron combinations, and calculate the ratio of the number of all triggered activated cross-layer neuron combinations to the number of all activated cross-layer neuron combinations which is used as the coverage rate of triggered activated neuron combinations in different layers, and is expressed as:

[0063] ;

[0064] where, is the set of all triggered activated cross-layer neuron combinations, is the set of all activated cross-layer neuron combinations, and the symbol represents the number of elements in the set.

[0065] Step 3: Combine the calculated multi-granularity neuron coverage metrics to select a part of new test samples.

[0066] In the embodiments of the present invention, combine the calculated multi-granularity neuron coverage metrics to evaluate the coverage of the currently selected seeds corresponding to the tests; according to the evaluated coverage, add perturbations or perform mutation operations on the seeds to generate new test samples.

[0067] Step 4: Add the newly selected test samples to the seed bank and transfer to Step 2; iterate continuously until the set test termination condition is met.

[0068] The above solution provided by the embodiments of the present invention mainly obtains the following beneficial effects:

[0069] (1) Organic combination of multi-granularity coverage metrics. By integrating coverage metrics such as single-neuron granularity, horizontal neuron combination granularity, and vertical neuron combination granularity, the present invention achieves a comprehensive capture of the behavioral characteristics of intelligent algorithms. The synergistic effect of these multi-granularity coverage metrics not only effectively makes up for the deficiencies of a single coverage rate but also expands the test coverage range, enabling accurate positioning of the weak points and untested areas of intelligent algorithms. Compared with traditional single coverage metrics, this method provides a more comprehensive description of the behavior of intelligent algorithms, laying a solid foundation for the verification of intelligent algorithms for complex tasks.

[0070] (2) Generation of high-quality test samples. By comprehensively utilizing multiple metrics, the present invention proposes a method for generating high-quality test samples guided by multi-granularity coverage rates, which can generate new test samples targeted according to the feedback of different granularity coverage metrics. This generation strategy significantly improves the diversity and pertinence of test samples, ensures coverage of the key behavioral areas of intelligent algorithms, and helps discover more potential error examples, thereby further enhancing the depth and effectiveness of testing.

[0071] (3) Complementary application of coverage metrics. The present invention emphasizes the complementary role between multiple coverage metrics, enabling testing to take into account both the structural characteristics, behavioral patterns, and semantic outputs of intelligent algorithms. Thus, it can not only evaluate the internal integrity of intelligent algorithms but also detect anomalies at the input-output level. This multi-level analysis method effectively avoids potential problems that may be overlooked by a single metric, providing strong technical support for improving the comprehensiveness of testing.

[0072] (4) Improvement in the security and reliability of intelligent algorithms. By generating test samples with rich diversity and high coverage rates, the present invention can quickly reveal potential defects of the model in complex scenarios, providing precise guidance for the improvement and optimization of intelligent algorithms. This multi-metric-driven testing method not only improves the testing efficiency but also significantly enhances the security and reliability of intelligent algorithms, especially suitable for high-risk application scenarios such as autonomous driving, medical diagnosis, and task verification of large language intelligent algorithms.

[0073] Generally speaking, by organically combining coverage metrics at multiple granularities, the present invention not only innovatively solves the limitations of insufficient single coverage rate but also realizes a more efficient and comprehensive intelligent algorithm testing method, providing a new solution for the security verification and optimization of intelligent algorithms, and having important research value and broad application prospects.

[0074] To more clearly demonstrate the technical solutions provided by the present invention and the resulting technical effects, the following uses specific embodiments to describe in detail the method provided by the embodiments of the present invention.

[0075] Considering that the existing single coverage rate index and its test sample generation technology have significant limitations, these indexes are usually difficult to comprehensively reflect the complex behavior of neural networks, resulting in limited quality of test samples, especially insufficient performance in discovering the key weaknesses of intelligent algorithms and optimizing security. Therefore, there is an urgent need to develop a test method based on multi-granularity coverage indexes to make up for the deficiencies of single indexes and improve test efficiency.

[0076] To this end, the embodiments of the present invention provide an intelligent algorithm test method based on multi-granularity neural network coverage rate, aiming to generate high-quality test samples through the synergistic effect of multi-granularity coverage indexes and achieve more efficient and comprehensive testing. For the sake of easy understanding, the following Figure 1 details two core parts in the shown process. One part is the coverage rate evaluation of multi-granularity coverage indexes, which is used to verify the effectiveness of different granularity coverage indexes on intelligent algorithms and clarify the ability of the data set to capture the behavioral characteristics of intelligent algorithms; the other part is high-quality sample generation, which uses these coverage indexes to generate test samples with both diversity and pertinence, helping intelligent algorithms discover more potential error behaviors or abnormal decisions and improving the security and reliability of intelligent algorithms. The present invention fills the deficiencies of the existing coverage index test method and provides a theoretical basis and technical support for intelligent algorithm testing.

[0077] I. Coverage rate evaluation of multi-granularity coverage indexes.

[0078] As Figure 2 shown, it shows the evaluation process of the multi-granularity coverage indexes of intelligent algorithms. First, prepare an appropriate data set and perform preprocessing to ensure that it can be used for testing; then, construct coverage indexes of different granularities, including single neuron granularity, horizontal neuron combination granularity, and vertical neuron combination granularity, to comprehensively evaluate the different levels of performance of intelligent algorithms; then, through inference, collect behavioral characteristics and calculate the coverage rate, and analyze the performance of the data set under different coverage indexes; subsequently, evaluate the effectiveness of the test data by analyzing the ability of different data sets to capture the behavioral characteristics of intelligent algorithms and clarify the test blind spots; finally, according to the analysis results, put forward improvement suggestions, optimize the intelligent algorithm and re-verify its performance. This method not only lays a theoretical foundation for the testing of intelligent algorithms, but also provides a clear direction for the optimization of intelligent algorithms.

[0079] 1. Determination of intelligent algorithms and data sets.

[0080] The selection of intelligent algorithms is the starting point of the testing work, and a suitable target model should be determined according to the actual application scenarios and task requirements. Common intelligent algorithms include convolutional neural networks for image tasks, recurrent neural networks for sequence processing, and the Transformer (transformer neural network) model based on the self-attention mechanism, which has been widely applied to multi-domain tasks in recent years. The design and training of these models are usually based on the characteristics of different tasks, such as classification, prediction, translation, etc. Therefore, it is necessary to ensure that the tested target model can cover the main application directions of the current research. When selecting intelligent algorithms, the complexity, application scope, and task scenarios of the models should be considered. For example, for image recognition tasks, structures such as ResNet (residual network), VGG (visual geometry group network), or EfficientNet (efficient convolutional network) can be selected as the test objects; for natural language processing tasks, language models such as BERT (bidirectional encoder representations from transformers), GPT (generative pre-trained model), or T5 (text-to-text model) can be selected. In multi-modal tasks such as autonomous driving, deep learning models that integrate multiple perception capabilities may need to be selected. The characteristics of these intelligent algorithms determine the key points and difficulties of the testing work. The selection of the dataset to be tested is equally crucial. The test dataset should be able to represent the diverse data distribution in the real application scenario and be challenging at the same time. The dataset usually contains a large number of input-output pairs, and each sample can trigger a specific behavior pattern of the model.

[0081] 2. Calculation of single-neuron coverage metric.

[0082] The single-neuron coverage metric is the most basic coverage measurement method, used to evaluate the activation of each single neuron in an intelligent algorithm under a specific input dataset. This coverage metric reflects the fine-grained response ability of the intelligent algorithm to the input data, thus revealing the basic behavioral characteristics of the intelligent algorithm. By counting the activation status of a single neuron on the dataset, its coverage can be obtained, and further analysis of the generality and robustness of the intelligent algorithm can be carried out. In a neural network model, a neuron is the smallest computing unit, and the activation value of each neuron is the result of a linear combination of its input weights and biases passing through a non-linear activation function. The single-neuron coverage metric aims to quantify how many neurons in the intelligent algorithm are activated under a specific input, that is, to measure whether the activation status of the neuron exceeds a preset threshold.

[0083] ;

[0084] Among them, represents the total number of neurons, is the activation value of neuron ; The threshold indicating whether it is activated. The specific process is as follows. First, traverse the database, load and input test samples into the intelligent algorithm in sequence. For each input sample, the processing process of the intelligent algorithm will activate the neurons in the network layer by layer, and the activation behaviors of these neurons need to be recorded in detail. Specifically, in each layer of the network, record the activation values of all neurons in real time, and set a predefined activation threshold to determine whether a neuron is activated. For neurons whose activation values exceed this threshold, count their numbers and mark them as effectively activated neurons. After completing the input and statistics of all samples, calculate the coverage rate of the network based on the recorded data. The definition of the coverage rate is: the ratio of the number of activated neurons to the total number of neurons in the network, which is used to evaluate the exploration degree of the test samples on the structure of the intelligent algorithm. Through this process, not only can the coverage performance of the samples be intuitively understood, but also data support can be provided for optimizing the generation of test samples and verifying the intelligent algorithm.

[0085] 3. Calculation of the coverage index for horizontal neuron combinations.

[0086] The coverage index for horizontal neuron combinations aims to analyze the activation patterns among neurons in the same layer. This coverage index reveals the behavioral characteristics of the intelligent algorithm in the interaction of neurons in the same layer by statistically calculating the coverage rates of different combined activation patterns, and is applicable to analyzing the synergistic effects and complex activation patterns of a single layer.

[0087] ;

[0088] Among them, is the set of all triggered neuron combination activations, is the set of all layer neuron combination activations, and the symbol represents the number of elements in the set. This index measures the coverage degree of the test data set on the neuron combination activation patterns in a specific layer. Further, regard the activation state vector of each input sample as an independent combined pattern, and record and count the combinations of all samples. These activation state combinations can not only reflect the activation of a single sample on the network, but also calculate the coverage rate of the combination by horizontally comparing the activation patterns of different samples, that is, record the proportion of the number of triggered activation combinations to the total number of possible combinations. Through this index, the breadth and depth of the exploration of neuron combination interactions by the test samples in the target network layer can be evaluated.

[0089] In addition, by statistically analyzing the distribution characteristics of these activation state combinations, untriggered combined patterns in the data set can be further discovered, and specific combined areas that are not fully activated can be located. These untriggered areas usually represent the unexplored space of the network in this layer, providing a clear optimization direction for the generation of subsequent test samples, such as activating these uncovered combinations by designing more diverse input samples.

[0090] The calculation method and applicable scope of combinatorial coverage are relatively wide, and it can be flexibly applied to different types of neural network structures, including fully connected networks, convolutional networks, and networks based on attention mechanisms. For fully connected networks, it can capture the simple linear combinations between nodes; for convolutional networks, it can explore the complex associations between convolutional kernels and feature maps; for networks based on attention mechanisms, it can reveal the dynamic change patterns of attention weights under different inputs. This analysis method is of great significance in the testing and optimization of complex intelligent algorithms, and can effectively improve the robustness and coverage rate of intelligent algorithms.

[0091] 4. Calculation of the longitudinal neuron combinatorial coverage metric.

[0092] The longitudinal neuron combinatorial coverage metric is used to analyze the cross-layer neuron activation paths in a neural network, aiming to capture the cooperative behavior of neurons between multiple layers. Different from the lateral neuron coverage, the longitudinal coverage metric focuses on the combination of activation states across layers, and can more comprehensively depict the global behavior characteristics of the network, especially the complex patterns in deep networks.

[0093] ;

[0094] Among them, is the set of all triggered cross-layer neuron combinatorial activations, is the set of all cross-layer neuron combinatorial activations, and the symbol represents the number of elements in the set.

[0095] Similar to the calculation method of lateral combinatorial coverage, the longitudinal coverage is also based on the statistics of combinations and the calculation of coverage rate, but its focus extends from the combination patterns of a single layer to the paths spanning multiple layers. In a deep neural network, each cross-layer combination can be regarded as a path, which is composed of the activation states of neurons in different layers and can reflect the layer-by-layer information transmission and interaction patterns from input to output. The core of the longitudinal coverage lies in capturing the cooperative activation behavior between layers in a deep network. By counting the number of cross-layer paths triggered by all test samples and comparing it with the total number of theoretically possible paths, the longitudinal coverage can be calculated. This metric can not only reveal the distribution of inter-layer dependencies in a deep neural network, but also analyze the cooperative activation efficiency of intelligent algorithms between different layers, revealing potential redundant connections or feature paths that are not fully utilized in the network. By analyzing these cross-layer paths, it is possible to evaluate whether there are test blind spots in the dataset. If most of the potential paths are not triggered, it indicates that the coverage range of test samples is limited and the global behavior of the intelligent algorithm has not been fully explored. This finding is of great significance for optimizing the generation of test samples, which can guide the design of more diverse and challenging input data to trigger more potential cross-layer paths, thereby improving the comprehensiveness of testing.

[0096] In the above process, the test samples input to the intelligent algorithm are the seeds selected from the seed bank. At the same time, diverse test samples are continuously generated based on the evaluation results to update the seed bank, and this process will be introduced in the next part.

[0097] II. High-quality sample generation.

[0098] As Figure 3 shown, it demonstrates the high-quality sample generation process guided by coverage metrics. By analyzing the coverage rate results to locate the test blind spots and combining with the behavioral characteristics of the intelligent algorithm, new samples that can fill the coverage gaps are generated. First, define the coverage metrics and evaluate the coverage of the existing dataset; then design an extensible seed selection strategy according to the coverage blind spots, and then generate targeted samples through the metamorphic strategy; finally, add the samples to the test set to verify their improvement effects on the coverage rate and the performance of the intelligent algorithm, and iteratively optimize the generation strategy. This process effectively improves the test coverage rate and the robustness of the intelligent algorithm, providing reliable support for the evaluation of the intelligent algorithm.

[0099] In this part, through designing a reasonable seed selection strategy, coverage analysis is carried out. This process aims to evaluate the quality of the generated samples and optimize the generation mechanism based on this, promoting the generation of new samples, thereby improving the comprehensiveness and effectiveness of the test.

[0100] 1. Extensible seed selection strategy.

[0101] To improve the efficiency and coverage of fuzz testing, an extensible seed selection strategy is designed. This strategy can dynamically adjust the scheduling method of test seeds to adapt to different types of neural networks and their complex application scenarios. By intelligently optimizing the selection and scheduling of seeds, this strategy can effectively improve the comprehensiveness and effectiveness of the test. The core of this strategy lies in three key technologies: progress awareness, random sorting, and frequency awareness: (1) Progress awareness: Dynamically adjust the seed selection strategy according to the test progress, and preferentially select seeds that can trigger new semantic features or cover unexplored areas in the current stage to ensure the balanced and efficient test process and avoid resource waste; (2) Random sorting: Introduce random perturbation sorting in the candidate seed set to break the fixed selection pattern, prevent local convergence of test paths, improve the diversity of test inputs, and thus enhance the exploration ability of potential defects in the neural network; (3) Frequency awareness: Dynamically adjust the priority of seeds based on the historical frequency information of test execution, avoid repeated use of inefficient seeds, and reasonably control the selection of high-frequency seeds to maximize the test coverage rate and the ability to discover potential vulnerabilities. The specific selection and scheduling strategy is as follows:

[0102] First, define , and respectively represent the coverage metrics of the horizontal neuron combination, the vertical neuron combination, and the single neuron (i.e., the aforementioned , , ). Given a seed bank, where a single seed is denoted as s, which corresponds to a test sample, and Z is the target coverage area. For the three granularity coverage metrics, let denote the set of semantic paths covered by the seed s, and any path in it is denoted as ; let denote the set of abstract semantic paths of the target coverage area Z, and any path in it is denoted as ; for the path and the path , their similarity can be calculated as:

[0103] ;

[0104] where, and are constants and ; for any and , three sets , and can be generated, which are expressed as: , and ; the scheduling impact factor of each seed s based on path similarity can be defined as:

[0105] ;

[0106] where, represents the number of elements in the corresponding set. For example, represents the number of elements in the set .

[0107] Taking the horizontal neuron combination coverage metric as an example, the maximum and minimum values of the historical coverage rate in the target coverage area are respectively denoted as and , the current coverage rate of the seed s in the target coverage area is denoted as , then the distance from the seed file s to the target coverage area Z can be calculated as:

[0108] ;

[0109] The maximum and minimum historical coverage rates in the above-mentioned target coverage area consider the seeds in the entire seed bank (the seeds in the initial seed bank will all undergo a calculation of the multi-granularity neuron coverage metric), while the current coverage rate mainly refers to the coverage of the horizontal neuron combination coverage metric of seed s in the target coverage area.

[0110] According to the above calculation steps, the distances from seed s to the target coverage area Z are calculated respectively for the vertical neuron combination and the single neuron coverage metric and ; finally, the comprehensive distance on the three-granularity neuron coverage metrics is:

[0111] ;

[0112] Among them, the coefficients 、 、 and the accuracy and effectiveness of each level of coverage metric need to be judged and regulated through the operation results of the fuzz testing design layer. This topic will further study how to learn the optimal parameter configuration for the multi-level distance function.

[0113] In the embodiments of the present invention, the set of semantic paths covered by seed s is a general term in the field of intelligent algorithm testing. It mainly describes a set composed of a series of processing steps or decision paths triggered by seed s when testing an intelligent algorithm. These steps or paths correspond to a part of all possible situations or state spaces that the intelligent algorithm needs to cover; taking a single semantic path as an example, it includes but is not limited to: which nodes (or neurons) can be activated by seed s, and which paths or channels seed s passes through in different layers of the network, etc. Similarly, the aforementioned target coverage area is a general term in the field of intelligent algorithm testing, which can be understood as a part of all possible situations or state spaces that need to be covered when solving a certain problem. Simply put, it is all different types of input data and their corresponding output result ranges that the intelligent algorithm should be able to reach when processing data. The set of abstract semantic paths it involves has a similar meaning, so it will not be elaborated here.

[0114] The given seed bank and the target coverage area Z, where m is the number of seeds. In addition to the path-based seed scheduling influence factor, the present invention will also consider seed diversity and proximity. During the scheduling process, a minimum scheduling probability is assigned to each seed. According to the scheduling times of seed s, the scheduling influence factor based on diversity and proximity is calculated, expressed as:

[0115] ;

[0116] Among them, the parameter is used to limit the rate of decrease of the scheduling probability.

[0117] For all seeds in the seed bank, the comprehensive distance and the scheduling influence factor based on diversity and proximity are calculated in the above manner. By calculating the product of the two (i.e., ), the comprehensive score of the seed is obtained. Seeds are selected based on the comprehensive score of the seeds. Generally, a specified number of seeds with the top comprehensive scores are selected.

[0118] The above seed selection strategy can achieve two key objectives: one is to improve the overall value of the multi-level coverage index, and the other is to locate potential defective modules of more neural networks.

[0119] 2. Metamorphic strategy.

[0120] After the seed selection is completed, the target sample is determined. Then, the metamorphic strategy is used to perform mutation operations on this sample. By making small modifications (i.e., mutations) to the selected seeds, a new set of mutant samples is generated. These mutant samples can simulate potential program errors or defects and help the test system discover untapped vulnerabilities. Mutation operations usually include modifying specific parts of the input data, such as numerical changes, character replacements, logical adjustments, etc., so as to generate a series of test cases with different characteristics. These metamorphic relationships can generate diverse new test samples based on the existing seed samples by adding perturbations or performing mutation operations, thereby exploring more model behavior spaces.

[0121] 3. Multi-granularity coverage index analysis.

[0122] After generating new test samples, coverage analysis is performed on these samples, and the multi-granularity neuron coverage index is calculated. That is, the new test samples are input into the intelligent algorithm and evaluated according to the solution in the first part mentioned above. By comparing the multi-granularity neuron coverage index, the test samples that can significantly improve the coverage range are selected and added to the seed bank. This incremental update mechanism can ensure the dynamic optimization of the seed bank, enabling each round of sample generation to further expand the model test space.

[0123] Based on the above two parts of the introduction, continuous cyclic testing is carried out until a preset termination condition is reached (for example, the test timeouts or the coverage index converges). This cyclic process can not only continuously improve the comprehensiveness of the test but also efficiently locate the potential defective areas of the model. Through this dynamic iterative method, the test process realizes the continuous optimization of the seed bank and the efficient generation of test samples, providing strong support for the robustness and defect discovery of deep neural networks.

[0124] Through the description of the above embodiments, those skilled in the art can clearly understand that the above embodiments can be implemented by software or by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solutions of the above embodiments can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.), including several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in various embodiments of the present invention.

[0125] Embodiment 2

[0126] The present invention also provides an intelligent algorithm testing system, which is mainly used to implement the methods provided in the foregoing embodiments, such as Figure 4 shown, the system mainly includes:

[0127] An intelligent algorithm and test data selection and index definition unit, which is used to select a target intelligent model that conforms to the actual application scenario as the intelligent algorithm to be tested, collect the corresponding test data set, screen out test samples from the test data set as seeds to form a seed library, and define the multi-granularity neuron coverage index during testing;

[0128] A multi-granularity neuron coverage index calculation unit, which is used to calculate the comprehensive score of the seeds based on the comprehensive distance of the seeds on the multi-granularity neuron coverage index and the scheduling influence factors based on diversity and proximity, select seeds from the seed library based on the comprehensive score of the seeds, generate new test samples through the metamorphic strategy, and then input them into the intelligent algorithm to be tested for testing, and calculate the multi-granularity neuron coverage index;

[0129] A test sample screening unit, which is used to screen out a part of new test samples in combination with the calculated multi-granularity neuron coverage index;

[0130] An iterative testing unit, which is used to add the screened new test samples to the seed library, and perform iterative testing through the multi-granularity neuron coverage index calculation unit and the test sample screening until the set test termination condition is met.

[0131] Considering that the specific technical details involved in the above system have been introduced in detail in the previous embodiments, they will not be elaborated here.

[0132] Those skilled in the art can clearly understand that for the convenience and conciseness of description, only the above division of each functional module is used as an example for illustration. In actual applications, the above functions can be allocated to different functional modules according to needs, that is, the internal structure of the system is divided into different functional modules to complete all or part of the functions described above.

[0133] Embodiment III

[0134] The present invention further provides a processing device, such as Figure 5 shown, which mainly includes: one or more processors; a memory for storing one or more programs; wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the method provided in the foregoing embodiment.

[0135] Furthermore, the processing device further includes at least one input device and at least one output device; in the processing device, the processor, the memory, the input device, and the output device are connected through a bus.

[0136] In the embodiments of the present invention, the specific types of the memory, the input device, and the output device are not limited; for example:

[0137] The input device may be a touch screen, an image acquisition device, a physical button, or a mouse, etc.;

[0138] The output device may be a display terminal;

[0139] The memory may be a Random Access Memory (RAM), or a non-volatile memory, such as a disk memory.

[0140] Embodiment IV

[0141] The present invention further provides a readable storage medium storing a computer program, which implements the method provided in the foregoing embodiment when the computer program is executed by a processor.

[0142] In the embodiments of the present invention, the readable storage medium as a computer-readable storage medium may be disposed in the foregoing processing device, for example, as the memory in the processing device. In addition, the readable storage medium may also be various media that can store program codes, such as a USB flash drive, a mobile hard disk, a Read-Only Memory (ROM), a magnetic disk, or an optical disc.

[0143] As described above, the above are only the preferred specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention should be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims. The information disclosed in the background art part of this article is only intended to deepen the understanding of the overall background technology of the present invention, and should not be regarded as an admission or any form of implication that this information constitutes the prior art already known to those skilled in the art.

Claims

1. An intelligent algorithm testing method, characterized in that: include: Step 1: Select the target intelligent model that meets the actual application scenario as the intelligent algorithm to be tested, collect the corresponding test data set, select the test samples from the test data set as seeds, form a seed library, and define the multi-granularity neuron coverage indicators during the test; Step 2: Calculate the comprehensive score of the seed based on the comprehensive distance of the seed on the multi-granularity neuron coverage index and the scheduling impact factor based on diversity and proximity. Select the seed from the seed library based on the comprehensive score of the seed, generate a new test sample through the metamorphosis strategy, and then input it into the intelligent algorithm to be tested for testing, and calculate the multi-granularity neuron coverage index; Step 3: Combine the calculated multi-granularity neuron coverage index to select a part of new test samples; Step 4: Add the selected new test samples to the seed library and go to step 2; continue to iterate until the set test termination condition is met; The comprehensive score of the seed is calculated based on the comprehensive distance of the seed on the multi-granular neuron coverage index and the scheduling impact factor based on diversity and proximity, and the seed is selected from the seed library based on the comprehensive score of the seed, including: Given a seed library, where a single seed is denoted as s, corresponding to a test sample, and a target coverage area Z during the test; let represents the set of semantic paths covered by seed s, any of which is denoted as ;make Represents a set of abstract semantic paths covering the target area Z, where any path is denoted as ; According to the path With path The similarity of the seed s is used to calculate the scheduling impact factor of the seed s based on the path similarity, and then combined with the coverage rate of the seed s on each multi-granularity neuron coverage index, the comprehensive distance of the seed s on the multi-granularity neuron coverage index is calculated. ; Combine the minimum scheduling probability of the pre-assigned seed s and the number of scheduling of seed s to calculate the scheduling influence factor of seed s based on diversity and proximity ; By calculation , get the comprehensive score of seed s; the comprehensive scores of other seeds are calculated in the same way; Select a portion of seeds from the seed bank based on the comprehensive score of the seeds; The calculation of the comprehensive distance of the seed s on the multi-granularity neuron coverage index is include: Calculate Path With path Similarity , expressed as: ； in, and is a constant and ; For any and , which can produce three sets , and , expressed as: , and ; Then the scheduling impact factor of each seed s based on path similarity is Defined as: ； in, Indicates the number of elements in the corresponding set; The definition of multi-granular neuron coverage indicators includes: single neuron coverage indicator, horizontal neuron combination coverage indicator and vertical neuron combination coverage indicator; for the horizontal neuron combination coverage indicator, the maximum and minimum values ​​of the historical coverage rate in the target coverage area are recorded as and , the current coverage rate of seed s in the target coverage area is recorded as , then the distance from the seed file s to the target coverage area Z for: ； Based on the longitudinal neuron combination coverage index and single neuron coverage index, the distance from seed s to the target coverage area Z is calculated as and , the comprehensive distance of seed s on the multi-granularity neuron coverage index is calculated by the following formula : ； in, , , are three coefficients.

2. The intelligent algorithm testing method according to claim 1, characterized in that: The multi-granularity neuron coverage index includes: a single neuron granularity coverage index, a horizontal neuron combination coverage index and a vertical neuron combination coverage index; the single neuron granularity coverage index is the coverage rate of activated neurons; the horizontal neuron combination coverage index is the coverage rate of triggered neuron combination activations in the same layer; the vertical neuron combination coverage index is the coverage rate of triggered neuron combination activations in different layers.

3. The intelligent algorithm testing method according to claim 2, characterized in that: The calculation method of the coverage of activated neurons includes: Calculate the ratio of activated neurons to the total number of neurons, expressed as: ； in, is the ratio of activated neurons to the total number of neurons, is the set of all neurons; For collection The number of elements of , that is, the total number of neurons; For neurons The activation value of is the threshold value; is the indicator function, when the condition is met , the output is 1, otherwise the output is 0.

4. The intelligent algorithm testing method according to claim 2, characterized in that: The coverage of combined activations of triggered neurons in the same layer is calculated as follows: For each layer, count the number of activated neuron combinations separately, summarize the number of activated neuron combinations of all layers, and count the number of triggered neuron combinations to obtain the number of all triggered neuron combinations activated; calculate the ratio of the number of all triggered neuron combinations activated to the number of activated combinations of all layers , as the coverage of the combined activation of triggered neurons in the same layer, is expressed as: ； in, is the set of combined activations of all fired neurons, is the set of combined activations of all layer neurons, symbol Indicates the number of elements in a collection.

5. The intelligent algorithm testing method according to claim 2, characterized in that: The coverage of combined activations of triggered neurons in different layers is calculated as follows: Count the number of cross-layer neuron combination activations, summarize the number of all cross-layer neuron combination activations, count the number of all triggered cross-layer neuron combination activations, and calculate the ratio of the number of all triggered cross-layer neuron combination activations to the number of all cross-layer neuron combination activations. As the coverage of combined activations of triggered neurons in different layers, it is expressed as: ； in, is the set of combined activations of all fired neurons across layers, is the set of combined activations of all neurons across layers, symbol Indicates the number of elements in a collection.

6. An intelligent algorithm testing system, characterized in that: The method for implementing any one of claims 1 to 5 comprises: The intelligent algorithm and test data selection and indicator definition unit is used to select the target intelligent model that meets the actual application scenario as the intelligent algorithm to be tested, collect the corresponding test data set, screen out test samples from the test data set as seeds to form a seed library, and define the multi-granularity neuron coverage indicators during the test; A multi-granularity neuron coverage index calculation unit is used to calculate the comprehensive score of the seed based on the comprehensive distance of the seed on the multi-granularity neuron coverage index and the scheduling influence factor based on diversity and proximity, select the seed from the seed library based on the comprehensive score of the seed, and generate a new test sample through the metamorphosis strategy, and then input it into the intelligent algorithm to be tested for testing, and calculate the multi-granularity neuron coverage index; A test sample screening unit is used to screen out a part of new test samples by combining the calculated multi-granularity neuron coverage index; The iterative test unit is used to add the screened new test samples to the seed library, and perform iterative testing through the multi-granularity neuron coverage index calculation unit and test sample screening until the set test termination conditions are met.

7. A processing device, characterized in that: include: one or more processors; A memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 5.

8. A readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images