Method, device and equipment for generating prediction result information based on multi-party secure computation

Abstract

Embodiments of the present disclosure disclose a prediction result information generation method, device and equipment based on multi-party secure computation. A specific implementation of the method comprises: in response to detecting early warning query sequence information, obtaining changed query sequence information according to a preset special word information set and a preset position; normalizing the split changed query sequence information to obtain target changed query sequence information; splitting a first parameter matrix information group to obtain each split parameter matrix information; generating encrypted matrix information according to the target changed query sequence information and each split parameter matrix information; generating encrypted item early warning information according to the encrypted matrix information and a second parameter matrix information group; decrypting the encrypted item early warning information to obtain item early warning information; and in response to determining that the item early warning information does not satisfy a preset condition, capturing and intercepting a target item. The implementation can shorten the time consumption for generating item early warning information and reduce the calculation resources consumed.

Description

TECHNICAL FIELD

[0001] Embodiments of the present disclosure relate to the cross-technical field of information security and computer application technology, in particular to a prediction result information generation method, device and equipment based on multi-party secure computation. BACKGROUND

[0002] The customs needs to quickly detect whether the holder of the article (for example, the holder can be a company holding the article) is an international or domestic prohibited holder of export or import, and if it is a prohibited holder of export or import of the article, the article is intercepted to maintain international security. At present, when detecting and intercepting the article, the commonly used way is to use a text matching model based on deep learning to determine whether the holder of the article is on a blacklist to generate early warning information of the article, wherein the blacklist can represent each holder prohibited to export or import the article, and the text matching model based on deep learning can realize the encryption processing of the query process based on the multi-party secure computation (MPC) protocol.

[0003] However, in practice, it is found that when the above method is used to generate the early warning information of the article, the following technical problems often exist:

[0004] When using the text matching model based on deep learning to realize the encryption processing of the query process, the multi-party data interaction using the multi-party secure computation (MPC) protocol is frequent, the data volume is large, and the protocol calculation complexity is high, resulting in a large communication overhead of the multi-party secure computation (MPC) protocol, and causing a long time and a large amount of computing resources to generate the early warning information of the article through the text matching model based on deep learning.

[0005] The above information disclosed in the background section of this document is only intended to enhance the understanding of the background of the present inventive concept and thus can include information that does not form the prior art known to those of ordinary skill in the art in the country. SUMMARY

[0006] The summary section of the present disclosure is used to introduce the concepts in a brief form, which will be described in detail in the specific embodiments section. The summary section of the present disclosure is not intended to identify key or essential features of the claimed technical solutions, nor is it intended to limit the scope of the claimed technical solutions.

[0007] Some embodiments of the present disclosure propose a prediction result information generation method, device, electronic equipment and computer readable medium to solve one or more of the technical problems mentioned in the background section.

[0008] In a first aspect, some embodiments of the present disclosure provide a method for generating prediction result information based on multi-party secure computation, the method comprising: in response to detecting pre-warning query sequence information corresponding to a target item, performing the following steps by using a pre-trained item pre-warning information generation model: performing modification processing on the pre-warning query sequence information according to a pre-set special token information set and a pre-set position to obtain modified query sequence information, wherein parameter information of the pre-trained item pre-warning information generation model comprises a first parameter matrix information group and a second parameter matrix information group; performing splitting processing on the modified query sequence information to obtain split processed modified query sequence information; performing normalization processing on the split processed modified query sequence information to obtain target modified query sequence information; performing splitting processing on each first parameter matrix information in the first parameter matrix information group to obtain each split parameter matrix information; generating encrypted matrix information corresponding to the target modified query sequence information according to the target modified query sequence information and the each split parameter matrix information; generating encrypted item pre-warning information corresponding to the target modified query sequence information according to the encrypted matrix information and the second parameter matrix information group; performing decryption processing on the encrypted item pre-warning information to obtain item pre-warning information as prediction result information; and in response to determining that the prediction result information does not satisfy a pre-set item qualification condition, controlling a robot arm to perform grabbing processing on the target item to intercept the target item.

[0009] In a second aspect, some embodiments of the present disclosure provide a prediction result information generation apparatus based on multi-party secure computation, comprising an execution unit configured to, in response to detecting a pre-warning query sequence information corresponding to a target item, perform the following steps by using a pre-trained item pre-warning information generation model: performing modification processing on the pre-warning query sequence information according to a preset special token information set and a preset position to obtain modified query sequence information, wherein parameter information of the pre-trained item pre-warning information generation model comprises a first parameter matrix information group and a second parameter matrix information group; performing splitting processing on the modified query sequence information to obtain split processed modified query sequence information; performing normalization processing on the split processed modified query sequence information to obtain target modified query sequence information; performing splitting processing on each first parameter matrix information in the first parameter matrix information group to obtain each split parameter matrix information; generating encrypted matrix information corresponding to the target modified query sequence information according to the target modified query sequence information and the each split parameter matrix information; generating encrypted item pre-warning information corresponding to the target modified query sequence information according to the encrypted matrix information and the second parameter matrix information group; performing decryption processing on the encrypted item pre-warning information to obtain item pre-warning information as prediction result information; and a control unit configured to, in response to determining that the prediction result information does not satisfy a preset item qualified condition, control a mechanical arm to perform grabbing processing on the target item to intercept the target item.

[0010] In a third aspect, some embodiments of the present disclosure provide an electronic device, comprising: one or more processors; a storage device having one or more programs stored thereon, when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in any of the implementations of the first aspect.

[0011] In a fourth aspect, some embodiments of the present disclosure provide a computer readable medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the method described in any of the implementations of the first aspect or the second aspect.

[0012] The above various embodiments of the present disclosure have the following beneficial effects: through the prediction result information generation method based on multi-party secure calculation of some embodiments of the present disclosure, the time consumption of generating the item early warning information can be shortened and the consumed computing resources can be reduced. The reason why the time consumption of generating the item early warning information is relatively long and the consumed computing resources are relatively large lies in that when the encryption processing of the query process is implemented by using the text matching model based on deep learning, the multi-party data interaction is frequent, the data volume is large and the protocol calculation complexity is high due to the encryption by using the multi-party secure calculation (MPC) protocol, which causes the communication overhead of using the multi-party secure calculation (MPC) protocol to be relatively large, resulting in that the time consumption of generating the item early warning information by using the text matching model based on deep learning is relatively long and the consumed computing resources are relatively large. Based on this, the prediction result information generation method based on multi-party secure calculation of some embodiments of the present disclosure first, in response to detecting the early warning query sequence information corresponding to the target item, performs the following steps by using the pre-trained item early warning information generation model: according to the preset special token information set and the preset position, the early warning query sequence information is changed to obtain the changed query sequence information. Thus, the early warning query sequence information of the target user can be changed to obtain the changed early warning query sequence information. Then, the changed query sequence information is split to obtain the split processed changed query sequence information. Thus, the changed early warning query sequence information can be split to obtain the split processed early warning query sequence information. Secondly, the split processed changed query sequence information is normalized to obtain the target changed query sequence information. Thus, the split processed changed query sequence information can be normalized to obtain the target changed query sequence information. Then, each first parameter matrix information in the first parameter matrix information group is split to obtain each split parameter matrix information. Thus, each split parameter matrix can be obtained. Secondly, according to the target changed query sequence information and the each split parameter matrix information, the encryption matrix information corresponding to the target changed query sequence information is generated. Thus, the encryption matrix corresponding to the target changed query sequence information can be obtained. Then, according to the encryption matrix information and the second parameter matrix information group, the encrypted item early warning information corresponding to the target changed query sequence information is generated. Thus, the encrypted item early warning data corresponding to the target changed query sequence information can be obtained. Secondly, the encrypted item early warning information is decrypted to obtain the item early warning information as the prediction result information. Thus, the decrypted item early warning data can be obtained. Finally, in response to determining that the prediction result information does not meet the preset item qualified condition, the mechanical arm is controlled to grasp the target item to intercept the target item. Thus, the item can be intercepted.Also, instead of directly using a multi-party secure computation (MPC) protocol to perform multi-party secure computation (MPC) to achieve encryption, the pre-warning query sequence information is first changed according to a preset special token information set, and the preset special token information in the preset special token information set is placed in a preset position of the pre-warning query sequence information, so as to change the pre-warning query sequence information. The preset special token information set can represent each random data whose activation value is greater than a preset threshold value selected from a random data set. Thus, the preset special token information with a large activation value can be fixed to the preset position of the pre-warning query sequence information (the preset position represents a position before the beginning of the pre-warning query sequence information). Thus, the number of comparisons for finding the maximum value of each row in the matrix corresponding to the pre-warning query sequence information can be reduced, and the number of multi-party data interactions and the number of multi-party secure computation (MPC) protocols can be reduced, so as to shorten the time consumed and reduce the computing resources consumed. Also, the pre-warning query sequence information is normalized according to the changed pre-warning query sequence information. Since the data range of the changed pre-warning query sequence information is small, the time consumed for normalizing the changed pre-warning query sequence information is short, and the computing resources consumed are less. Thus, the time consumed for generating the item pre-warning information can be shortened, and the computing resources consumed can be reduced. BRIEF DESCRIPTION OF DRAWINGS

[0013] The above and other features, advantages, and aspects of embodiments of the present disclosure will become more apparent by describing in detail some embodiments thereof with reference to the attached drawings. The same or similar components have the same or similar reference labels. It should be understood that the drawings are schematic and elements and features are not necessarily drawn to scale.

[0014] Figure 1 is a flowchart of some embodiments of a multi-party secure computation-based prediction result information generation method according to the present disclosure;

[0015] Figure 2 is a structural schematic diagram of some embodiments of a multi-party secure computation-based prediction result information generation apparatus according to the present disclosure;

[0016] Figure 3 is a structural schematic diagram of an electronic device suitable for implementing some embodiments of the present disclosure. DETAILED DESCRIPTION

[0017] Embodiments of the present disclosure will be described below in greater detail with reference to the accompanying drawings. While certain embodiments of the present disclosure are shown in the drawings, it is understood that the present disclosure can be embodied in various forms and should not be construed as being limited to the embodiments set forth herein. Rather, these embodiments are provided so that the present disclosure will be thoroughly and completely understood. It should be understood that the drawings and embodiments of the present disclosure are only for illustrative purposes and are not intended to limit the scope of protection of the present disclosure.

[0018] In addition, it should be further noted that only parts related to the present application are shown in the drawings for ease of description. The embodiments in the present disclosure and the features in the embodiments can be combined with each other without conflict.

[0019] It should be noted that the terms "first", "second", and the like mentioned in the present disclosure are only used to distinguish different devices, modules or units, and are not intended to limit the order or interdependence of the functions performed by these devices, modules or units.

[0020] It should be noted that the terms "one", "multiple" mentioned in the present disclosure are illustrative and not restrictive, and those skilled in the art should understand that unless otherwise explicitly stated in the context, it should be understood as "one or more".

[0021] The names of the messages or information exchanged between the devices in the embodiments of the present disclosure are only for illustrative purposes, and are not intended to limit the scope of the messages or information.

[0022] The present disclosure will be described in detail below with reference to the accompanying drawings and in conjunction with the embodiments.

[0023] Figure 1 Flow 100 of some embodiments of a method for generating prediction result information based on multi-party secure computation according to the present disclosure is shown. The method for generating prediction result information based on multi-party secure computation includes the following steps:

[0024] Step 101, in response to detecting the early warning query sequence information corresponding to the target item, performing the following steps through the pre-trained item early warning information generation model:

[0025] Step 1011, according to the pre-set special token information set and the pre-set position, the above-mentioned early warning query sequence information is processed to obtain the changed query sequence information.

[0026] In some embodiments, the execution subject (e.g., a computing device) of the multi-party secure computation-based prediction result information generation method can perform alteration processing on the above-mentioned early warning query sequence information according to a preset special token information set and a preset position to obtain altered query sequence information, where the above-mentioned early warning query sequence information can represent a query sentence input by a target user into the above-mentioned pre-trained item early warning information generation model. The above-mentioned query sentence can represent a sentence for querying item early warning information of the target item. The above-mentioned early warning query sequence information can include individual characters. Here, the specific type of the target item and the early warning query sequence information are not limited, for example, the target item can represent furniture. For example, the early warning query sequence information can be "Please query whether the target item held by the party can enter or exit the country." The above-mentioned preset position can represent a position before the beginning of the above-mentioned early warning query sequence information. The above-mentioned altered query sequence information can represent a matrix corresponding to the altered early warning query sequence information. The above-mentioned pre-trained item early warning information generation model can be inputted with early warning query sequence information and outputted with item early warning information. The above-mentioned item early warning information can represent the result of querying the above-mentioned early warning query sequence information outputted by the above-mentioned pre-trained item early warning information generation model. For example, the query result can be "Yes." Here, the specific content of the above-mentioned early warning query sequence information and the above-mentioned item early warning information is not limited and can be adjusted according to actual needs. The training method of the above-mentioned pre-trained item early warning information generation model can be batch training. The above-mentioned pre-trained item early warning information generation model can represent a large language model based on a Decoder-only Transformer architecture. Here, the specific type of the large language model is not limited, for example, the large language model can be Llama2 7B. The above-mentioned Llama2 7B can include 32 Transformer blocks. The above-mentioned Transformer block can include an Attention layer and an MLP layer. The above-mentioned Attention layer can include a query weight matrix, a key weight matrix, a value weight matrix, and an output weight matrix. The above-mentioned MLP layer can include a dimension increasing weight matrix, a gating weight matrix, and a dimension decreasing weight matrix. The parameter information of the above-mentioned pre-trained item early warning information generation model includes a first parameter matrix information group and a second parameter matrix information group. The first parameter matrix information in the above-mentioned first parameter matrix information group can represent query matrix information, key matrix information, value matrix information, or output matrix information. The second parameter matrix information in the above-mentioned second parameter matrix information group can represent a dimension increasing weight matrix, a gating weight matrix, or a dimension decreasing weight matrix. The above-mentioned query matrix information can represent a query weight matrix corresponding to the above-mentioned Attention layer. The above-mentioned key matrix information can represent a key weight matrix corresponding to the above-mentioned Attention layer.The value matrix information can represent a value weight matrix corresponding to the attention layer. The output matrix information can represent an output weight matrix corresponding to the attention layer. The dimension-increasing weight matrix can represent a dimension-increasing weight matrix corresponding to the MLP layer. The gating weight matrix can represent a gating weight matrix corresponding to the MLP layer. The dimension-reducing weight matrix can represent a dimension-reducing weight matrix corresponding to the MLP layer.

[0027] It should be noted that the application scenario of the prediction result information generation method based on multi-party secure computation of the present disclosure can be any scenario requiring a multi-party secure computation (MPC) protocol. In this regard, the specific type of the application scenario of the prediction result information generation method based on multi-party secure computation of the present disclosure is not limited, and can be adjusted according to actual needs. For example, the application scenario can be a cloud service scenario requiring a multi-party secure computation (MPC) protocol.

[0028] In some optional implementations of some embodiments, the execution subject can perform the following steps to modify the early warning query sequence information according to the preset special token information set and the preset position, and obtain modified query sequence information:

[0029] Firstly, the preset special token information set is added to the preset position of the early warning query sequence information to obtain target sequence information. The target sequence information can represent the early warning query sequence information after the preset special token information set is prefixed to the preset position of the early warning query sequence information. In practice, the execution subject can add the preset special token information set to the preset position of the early warning query sequence information by using the KV-cache method to obtain the target sequence information.

[0030] Secondly, the target sequence information is modified to obtain modified query sequence information. In practice, firstly, the execution subject can perform tokenization processing on the target sequence information by using a tokenizer to obtain each character corresponding to the target sequence information. Then, each character corresponding to the target sequence information is mapped by using the tokenizer to map each character to a corresponding numerical value to obtain a numerical sequence. Subsequently, the obtained numerical sequence can be converted into a matrix by using a direct reshaping method, and finally, the converted matrix is determined as the modified query sequence information.

[0031] Optionally, before step 101, the execution subject can further perform the following steps:

[0032] In a first step, based on preset dataset information and a pre-trained item early warning information generation model, a token information set and an activation value information set corresponding to the token information set are generated. The preset dataset information can represent a natural language sequence set input by a user. The natural language sequence in the natural language sequence set can represent a query sentence input to the pre-trained item early warning information generation model. The natural language sequence can include at least one character. In this regard, the specific content of the natural language sequence is not limited, for example, the natural language sequence can be “Tomorrow's price is 14?” or “xx holder's table can be out?” The token information set can represent a token corresponding to each character included in the preset dataset information. The token can represent a numerical value corresponding to at least one character in the preset dataset information. In this regard, the specific content of the token is not limited, for example, the token can be “52”. The token information in the token information set corresponds to the activation value information in the activation value information set. The activation value information in the activation value information set can represent the response intensity of a certain layer or neuron of the pre-trained item early warning information generation model to the token.

[0033] In a second step, according to the activation value information set, each token information in the token information set that satisfies a preset threshold condition is determined as a special token information set as a preset special token information set. The preset threshold condition can be that the scale result corresponding to the token information is greater than a preset threshold. The scale result can represent the ratio between the activation value of the token information and the mean value of the activation value information set. The preset threshold can represent a hyperparameter preset for the pre-trained item early warning information generation model. In this regard, the specific value of the hyperparameter is not limited, for example, the hyperparameter can be 80. The special token information in the special token information set can represent the token in the token information set that satisfies the preset threshold condition. In practice, first, the execution subject can determine the mean value of each activation value represented by the activation value information set. Then, the ratio between the activation value corresponding to each token information in the token information set and the mean value corresponding to the activation value information set is determined to obtain the scale result corresponding to each token information in the token information set. Finally, each token information in the token information set that satisfies the preset threshold condition is determined as a special token information set.

[0034] In some optional implementations of some embodiments, the execution subject can generate a token information set and an activation value information set corresponding to the token information set based on preset dataset information and a pre-trained item early warning information generation model by the following steps:

[0035] In a first step, the preset dataset information is sampled to obtain sampled preset dataset information as sample sequence information of the item early warning information generation model. The sample sequence information can represent each natural language sequence randomly selected from the preset dataset information. In practice, the execution subject can randomly select each natural language sequence from the preset dataset information as sample sequence information.

[0036] In a second step, the sample sequence information and the pre-trained item early warning information generation model are used to generate token information corresponding to the preset dataset information. In practice, the execution subject can first input the sample sequence information into the pre-trained item early warning information generation model, then convert each natural language sequence included in the sample sequence information into each token through the tokenizer of the pre-trained item early warning information generation model to obtain each token. Finally, each token obtained is determined as token information.

[0037] In a third step, the token information is used to generate activation value information corresponding to the token information. In practice, the execution subject can input each token information in the token information into the pre-trained item early warning information generation model to obtain each activation value information corresponding to the token information as activation value information.

[0038] In step 1012, the change query sequence information is split to obtain split change query sequence information.

[0039] In some embodiments, the execution subject can split the change query sequence information to obtain split change query sequence information. The split change query sequence information can include first share information and second share information. The first share information can represent a matrix held by a first party participating in a multi-party secure computation (MPC) protocol. The second share information can represent a matrix held by a second party participating in the multi-party secure computation (MPC) protocol. In this regard, the relationship between the first share information and the second share information is not limited. For example, the matrix represented by the first share information and the matrix represented by the second share information are the same, or the sum of the matrix represented by the first share information and the matrix represented by the second share information is the same as the matrix represented by the change query sequence information. In practice, the execution subject can split the change query sequence information according to the multi-party secure computation (MPC) protocol to obtain the split change query sequence information.

[0040] In step 1013, the split change query sequence information is normalized to obtain target change query sequence information.

[0041] In some embodiments, the execution subject can normalize the split change query sequence information to obtain target change query sequence information. The target change query sequence information can represent a matrix obtained by normalizing the split change query sequence information. The target change query sequence information can include a first change matrix and a second change matrix. The first change matrix can represent a share (matrix) of the target change query sequence information held by a first party participating in a multi-party secure computation (MPC) protocol. The second change matrix can represent a share (matrix) of the target change query sequence information held by a second party participating in the multi-party secure computation (MPC) protocol. The first change matrix and the second change matrix are both matrices. Here, the relationship between the first change matrix and the second change matrix is not limited. For example, the matrix represented by the first change matrix is the same as the matrix represented by the second change matrix, or the sum of the matrix represented by the first change matrix and the matrix represented by the second change matrix is the same as the matrix represented by the target change query sequence information.

[0042] In some optional implementations of some embodiments, the execution subject can normalize the split change query sequence information to obtain target change query sequence information by the following steps:

[0043] In a first step, in response to determining that the preset normalization processing mode is a first normalization processing mode, a square matrix information and dimension sum information corresponding to the split processed change query sequence information are generated according to the split processed change query sequence information. The preset normalization processing mode can include the first normalization processing mode and a second normalization processing mode. The first normalization processing mode can represent a step of normalizing the split processed change query sequence information based on an additive secret sharing protocol. The specific type of the additive secret sharing protocol is not limited, for example, the additive secret sharing protocol can be CrypTen. The second normalization processing mode can represent a step of normalizing the split processed change query sequence information based on a function secret sharing protocol. The specific type of the function secret sharing protocol is not limited, for example, the function secret sharing protocol can be Sigma. The square matrix information can represent a matrix obtained by squaring a fusion matrix. The fusion matrix can represent a sum of a matrix represented by the first share information included in the split processed change query sequence information and a matrix represented by the second share information included in the split processed change query sequence information. The dimension sum information can represent a vector of a sum of each first addition result in the square matrix information and a minimum positive number. The first addition result can represent a result of adding each row of data in the square matrix information. The specific value of the minimum positive number is not limited, for example, the minimum positive number can be 1e-5. In practice, first, the execution subject can determine a sum of the matrix represented by the first share information and the matrix represented by the second share information included in the split processed change query sequence information as the fusion matrix. Then, the square matrix information is obtained by squaring the fusion matrix. After that, the dimension sum information is obtained by adding each row of data in the square matrix information and the minimum positive number.

[0044] In a second step, interpolation result information corresponding to the split processed change query sequence information is generated according to the dimension sum information and a preset interpolation mode. The preset interpolation mode can represent a polynomial approximation method in a multi-party computation (MPC) protocol. The interpolation result information can represent a matrix obtained by processing the split processed change query sequence information through the polynomial approximation method and an iterative optimization processing method. The specific type of the iterative optimization processing method is not limited, for example, the iterative optimization processing method can be Newton's approximation method. In practice, first, the execution subject can process the split processed change query sequence information through the polynomial approximation method to obtain processed change query sequence information. Then, the processed change query sequence information is updated according to the dimension sum information through the iterative optimization processing method to obtain the interpolation result information.

[0045] In the third step, a product of the square matrix information and the interpolation result information is determined as the target change query sequence information.

[0046] In the fourth step, in response to determining that the preset normalization processing mode is the second normalization processing mode, the target change sequence information after the splitting processing is normalized to obtain the target change query sequence information. In practice, first, the execution subject can determine a result of adding the target change sequence information after the splitting processing and a first random matrix as a second addition result. The first random matrix can represent a randomly selected matrix. Then, an element-wise square operation is performed on the second addition result to obtain a square matrix corresponding to the second addition result. Thereafter, a sum of the square matrix corresponding to the second addition result and the minimum positive number is determined as a third addition result. Secondly, by using a lookup table protocol, reciprocal square roots of each value in the third addition result are determined to obtain the respective reciprocal square roots as a reciprocal square root matrix. Here, the specific type of the lookup table protocol is not limited, for example, the lookup table protocol can be a resource optimization type lookup table. Finally, a multiplication protocol processing is performed on the reciprocal square root matrix and the second addition result to obtain the target change query sequence information. Here, the specific type of the multiplication protocol is not limited, for example, the multiplication protocol can be a secure multi-party multiplication protocol.

[0047] In step 1014, each first parameter matrix information in the first parameter matrix information group is split to obtain respective split parameter matrix information.

[0048] In some embodiments, the execution subject can split each of the first parameter matrix information in the first parameter matrix information set to obtain each split parameter matrix information. The split parameter matrix information in each of the split parameter matrix information can be split query matrix information, split key matrix information, split value matrix information, or split output matrix information. The split parameter matrix information in each of the split parameter matrix information can include first split parameter matrix information and second split parameter matrix information. Here, the relationship between the first split parameter matrix information and the second split parameter matrix information is not limited. For example, the matrix represented by the first split parameter matrix information is the same as the matrix represented by the second split parameter matrix information, or the sum of the matrix represented by the first split parameter matrix information and the matrix represented by the second split parameter matrix information is the same as the matrix represented by the split parameter matrix information. The first split parameter matrix information can represent first query matrix information, first key matrix information, first value matrix information, or first output matrix information. The first query matrix information can represent a share (matrix) of the query matrix information held by a first party participating in a multi-party computation (MPC) protocol. The first key matrix information can represent a share (matrix) of the key matrix information held by the first party participating in the multi-party computation (MPC) protocol. The first value matrix information can represent a share (matrix) of the value matrix information held by the first party participating in the multi-party computation (MPC) protocol. The first output matrix information can represent a share (matrix) of the output matrix information held by the first party participating in the multi-party computation (MPC) protocol. The second split parameter matrix information can represent second query matrix information, second key matrix information, second value matrix information, or second output matrix information. The second query matrix information can represent a share (matrix) of the query matrix information held by a second party participating in the multi-party computation (MPC) protocol. The second key matrix information can represent a share (matrix) of the key matrix information held by the second party participating in the multi-party computation (MPC) protocol. The second value matrix information can represent a share (matrix) of the value matrix information held by the second party participating in the multi-party computation (MPC) protocol. The second output matrix information can represent a share (matrix) of the output matrix information held by the second party participating in the multi-party computation (MPC) protocol. In practice, the execution subject can split each of the first parameter matrix information in the first parameter matrix information set based on a multi-party computation (MPC) protocol to obtain each split parameter matrix information. Here, the specific type of the multi-party computation (MPC) protocol is not limited. For example, the type of the multi-party computation (MPC) protocol can be a secret sharing protocol.

[0049] In the process of adopting the technical solutions to solve the above technical problems, the following technical problem two often accompanies: when detecting the entry and exit of the goods, a large number of activation values are generated when generating the goods warning information using the multi-party secure computing (MPC) protocol, and there are maximum values in the generated large number of activation values, resulting in a large value range; when determining the index results of each activation value using the SoftMax function, due to the large threshold range of each activation value, it consumes more computing resources and takes longer time to determine the index results of each activation value in the full value range, resulting in longer time to generate the goods warning information, and further accumulating more goods to be detected. In view of the above technical problem two, the conventional solution is generally to fix the threshold range corresponding to each activation value. Combined with the advantages of the multi-party secure computing (MPC) protocol possessed by the company where the inventors work, we decided to adopt the following solution:

[0050] Step 1015, generating encrypted matrix information corresponding to the target change query sequence information according to the target change query sequence information and each split parameter matrix information.

[0051] In some embodiments, the above execution subject can generate encrypted matrix information corresponding to the above target change query sequence information according to the above target change query sequence information and the above each split parameter matrix information. Wherein, the above encrypted matrix information can represent the encrypted target change query sequence information output by the Attention layer included in each Transformer Block in the above pre-trained goods warning information generation model.

[0052] In some optional implementation manners of some embodiments, the above execution subject can generate encrypted matrix information corresponding to the above target change query sequence information according to the above target change query sequence information and the above each split parameter matrix information by the following steps:

[0053] First, for the split parameter matrix information in the above each split parameter matrix information, the following split steps are performed:

[0054] First sub-step, determining the product of the first split parameter matrix information included in the above split parameter matrix information and the above first change matrix as the first split matrix.

[0055] Second sub-step, determining the product of the second split parameter matrix information included in the above split parameter matrix information and the above second change matrix as the second split matrix.

[0056] In the second step, a split fusion matrix is generated according to the obtained first split matrix and the second split matrix. The first split matrix can represent the first query product information, the first key product information or the first value product information. The first query product information can represent a matrix obtained by performing matrix multiplication on the first query matrix information and the first change matrix. The first key product information can represent a matrix obtained by performing matrix multiplication on the first key matrix information and the first change matrix. The first value product information can represent a matrix obtained by performing matrix multiplication on the first value matrix information and the first change matrix. The second split matrix can represent the second query product information, the second key product information or the second value product information. The second query product information can represent a matrix obtained by performing matrix multiplication on the second query matrix information and the second change matrix. The second key product information can represent a matrix obtained by performing matrix multiplication on the second key matrix information and the second change matrix. The second value product information can represent a matrix obtained by performing matrix multiplication on the second value matrix information and the second change matrix. In practice, first, the execution subject can perform matrix multiplication on the transpose of the first query product information and the transpose of the first key product information to obtain a first target matrix. Then, the transpose of the second query product information and the transpose of the second key product information are multiplied to obtain a second target matrix. Finally, the first target matrix and the second target matrix are fused by a secret sharing protocol to obtain a split fusion matrix. The split fusion matrix can represent a matrix obtained by fusing the first target matrix and the second target matrix by a secret sharing protocol.

[0057] In the third step, in response to determining that the target change query sequence information satisfies a first preset normalization condition, an encrypted matrix information corresponding to the target change query sequence information is generated according to the split fusion matrix, the value parameter matrix information, the first exponential function information and the output parameter matrix information. The first preset normalization condition can represent that the target change query sequence information is normalized based on an additive secret sharing protocol. The first exponential function information can represent an exponential function protocol based on an additive secret sharing protocol. In practice, in the first execution step, the execution subject can execute the following steps according to the first exponential function information:

[0058] First, the split fusion matrix is split to obtain a target split fusion matrix. In practice, the execution subject can split the split fusion matrix by the MPC protocol to obtain the target split fusion matrix.

[0059] Then, the target split fusion matrix is converted by a change of base formula to obtain a first change of base matrix.

[0060] Then, the first change of base matrix is truncated to obtain a matrix corresponding to a decimal part of the first change of base matrix as a decimal part matrix. In practice, the first change of base matrix can be truncated by using a modf() function.

[0061] Then, a correction term matrix corresponding to the decimal part matrix is determined by using a lookup table protocol.

[0062] Then, a sum of the first change of base matrix, the correction term matrix, and a preset value is determined as a sum matrix. Here, the specific content of the preset value is not limited, for example, the preset value can be 121.274.

[0063] Then, the sum matrix is truncated to obtain a matrix corresponding to an exponent part of the sum matrix as an exponent part matrix, and a matrix corresponding to a mantissa part of the sum matrix as a mantissa part matrix. It should be noted that the truncation of the sum matrix is the same as the truncation of the first change of base matrix, which will not be described here.

[0064] Then, a real value matrix corresponding to the exponent matrix is obtained by determining a real value corresponding to each value in the exponent matrix by using a lookup table protocol. Here, the real value corresponding to the value can represent a value obtained by querying the value by using the lookup table protocol.

[0065] Then, the mantissa matrix and the real value matrix are multiplied by using a multiplication protocol to obtain an exponent function matrix as a result matrix of the exponent function calculation. Here, the target split fusion matrix can include a first target split fusion matrix and a second target split fusion matrix. The first target split fusion matrix can represent a share (matrix) of the target split fusion matrix held by a first party participating in a multi-party computation (MPC) protocol. The second target split fusion matrix can represent a share (matrix) of the target split fusion matrix held by a second party participating in the multi-party computation (MPC) protocol. The correction term matrix can represent a matrix for correcting the bias of the decimal part matrix.

[0066] A second execution step, matrix multiplication is performed between the exponential function matrix and the value parameter matrix information to obtain an exponential product matrix. In practice, first, the exponential function matrix is split by a secret sharing protocol to obtain a split exponential function matrix. The split exponential function matrix can include a first exponential function matrix and a second exponential function matrix. Here, the relationship between the first exponential function matrix and the second exponential function matrix is not limited. For example, the matrix represented by the first exponential function matrix is the same as the matrix represented by the second exponential function matrix, or the sum of the matrix represented by the first exponential function matrix and the matrix represented by the second exponential function matrix is the same as the matrix represented by the exponential function matrix. The first exponential function matrix can represent a share (matrix) of the exponential function matrix held by a first party participating in a multi-party secure computation (MPC) protocol. The second exponential function matrix can represent a share (matrix) of the exponential function matrix held by a second party participating in the multi-party secure computation (MPC) protocol. Then, the result of multiplying the first exponential function matrix in the split exponential function matrix by the first value parameter matrix information included in the value parameter matrix information is taken as a first product. The result of multiplying the second exponential function matrix in the split exponential function matrix by the second value parameter matrix information included in the value parameter matrix information is taken as a second product. Finally, the first product and the second product are fused by a secret sharing protocol to obtain the exponential product matrix.

[0067] In a third execution step, a matrix multiplication operation is performed on the index product matrix and the output parameter matrix information to obtain encrypted matrix information corresponding to the target change query sequence information. In practice, first, the index product matrix is split by a secret sharing protocol to obtain a split index product matrix. The split index product matrix can include a first index product matrix and a second index product matrix. In this regard, the relationship between the first index product matrix and the second index product matrix is not limited. For example, the matrix represented by the first index product matrix is the same as the matrix represented by the second index product matrix, or the sum of the matrix represented by the first index product matrix and the matrix represented by the second index product matrix is the same as the matrix represented by the index product matrix. The first index product matrix can represent a share (matrix) of the index product matrix held by a first party participating in a multi-party secure computation (MPC) protocol. The second index product matrix can represent a share (matrix) of the index product matrix held by a second party participating in the multi-party secure computation (MPC) protocol. Then, the result of multiplying the first index product matrix in the split index product matrix by the first output parameter matrix information included in the output parameter matrix information is taken as a third product. The result of multiplying the second index product matrix in the split index product matrix by the second output parameter matrix information included in the output parameter matrix information is taken as a fourth product. Finally, the third product and the fourth product are fused by a secret sharing protocol to obtain encrypted matrix information.

[0068] In a fourth step, in response to determining that the target change query sequence information satisfies a second preset normalization condition, encrypted matrix information corresponding to the target change query sequence information is generated based on the split fusion matrix, the value parameter matrix information, second exponential function information, and the output parameter matrix information. The second preset normalization condition can represent that the target change query sequence information is normalized based on a function secret sharing protocol. The second exponential function information can represent an exponential function protocol based on a function secret sharing protocol. In practice, in a first step, the execution subject can perform the following offline processing steps in an offline stage based on the second exponential function information:

[0069] First, a base conversion is performed on a preset random matrix to obtain a second base conversion matrix corresponding to the preset random matrix. The preset random matrix can represent a randomly selected matrix. The offline stage can represent that the pre-trained item early warning information generation model is in an offline state. It should be noted that the offline processing steps based on the second exponential function information do not need to be performed every time the pre-trained item early warning information generation model is in the offline stage. The offline processing steps can be set according to requirements.

[0070] Then, the second base conversion matrix is truncated to obtain a first decimal matrix corresponding to a decimal part of the second base conversion matrix and a first integer matrix corresponding to an integer part of the second base conversion matrix. In practice, the execution subject can use a modf() function to truncate the second base conversion matrix. The decimal part can represent decimals of each value in the matrix. The integer part can represent integers of each value in the matrix.

[0071] Secondly, an index calculation is performed on the first decimal matrix to obtain a first index matrix corresponding to the first decimal matrix.

[0072] Then, based on a function secret sharing protocol, a first mask matrix and a second mask matrix are generated. The first mask matrix and the second mask matrix can both represent randomly generated matrices. In practice, the execution subject can randomly generate the first mask matrix and the second mask matrix through the function secret sharing protocol.

[0073] Secondly, a first differential matrix and a second differential matrix are generated according to the first mask matrix and the first index matrix. The first differential matrix can represent a difference between the first index matrix and the first mask matrix. The second differential matrix can be the same as the first mask matrix. In practice, the execution subject can determine the difference between the first index matrix and the first mask matrix as the first mask matrix. The first differential matrix and the second differential matrix can both represent matrices.

[0074] Then, the second mask matrix is split through a function secret sharing protocol to obtain a first mask split matrix corresponding to the second mask matrix and a second mask split matrix corresponding to the second mask matrix. The sum of the first mask split matrix and the second mask split matrix is the second mask matrix.

[0075] Secondly, a lookup table protocol calculation is performed on the first integer matrix to obtain an index result matrix corresponding to the first integer matrix. In practice, the execution subject can perform a lookup table protocol calculation on the first integer matrix through a lookup table protocol to obtain an index result matrix corresponding to the first integer matrix. For example, the lookup table protocol can be an 8-bit LUT.

[0076] Finally, a matrix multiplication operation is performed on the index result matrix and the second mask matrix to obtain a noise mask matrix.

[0077] Secondly, the following online processing steps can be performed in the online phase:

[0078] Firstly, a sum of the preset random matrix and the split fusion matrix is determined as a target addition matrix corresponding to the split fusion matrix. The online stage can represent that the pre-trained item early warning information generation model is in an online state. The exponential function protocol based on the function secret sharing protocol can represent the offline processing step and the online processing step performed by the execution subject.

[0079] Then, the target addition matrix is subjected to a base conversion process to obtain a target base conversion matrix corresponding to the target addition matrix.

[0080] Secondly, the target base conversion matrix is subjected to a truncation process, and a matrix corresponding to a decimal part of the target base conversion matrix is a second decimal matrix, and a matrix corresponding to an integer part of the target base conversion matrix is a second integer matrix. It should be noted that the target base conversion matrix is subjected to the truncation process in the same way as the first base conversion matrix, which will not be described here.

[0081] Then, the second decimal matrix is subjected to an exponential calculation according to the first decimal matrix to obtain a second exponential matrix corresponding to the second decimal matrix.

[0082] Secondly, the first target difference matrix and the second target difference matrix are generated according to the first difference matrix, the second difference matrix and the second exponential matrix. The first target difference matrix can represent the product between the first difference matrix and the second exponential matrix. The second target difference matrix can represent the product between the second difference matrix and the second exponential matrix. In practice, firstly, the execution subject can determine the product between the first difference matrix and the second exponential matrix as the first target difference matrix. Then, the product between the second difference matrix and the second exponential matrix can be determined as the second target difference matrix.

[0083] Then, a sum of the first target difference matrix and the first mask split matrix is determined as a first intermediate result matrix corresponding to the first target difference matrix.

[0084] Secondly, a sum of the second target difference matrix and the second mask split matrix is determined as a second intermediate result matrix corresponding to the second target difference matrix.

[0085] Then, a sum of the first intermediate result matrix and the second intermediate result matrix is determined as a target fusion result matrix corresponding to the second decimal matrix.

[0086] Secondly, a table lookup protocol calculation is performed on the second integer matrix to obtain a target table lookup result matrix corresponding to the second integer matrix. In practice, the execution subject can perform a table lookup protocol calculation on the second integer matrix through a table lookup protocol to obtain a target table lookup result matrix corresponding to the second integer matrix. Here, the specific type of the table lookup protocol is not limited.

[0087] Then, matrix multiplication is performed on the target fusion result matrix and the target table lookup result matrix to obtain a target multiplication matrix.

[0088] Finally, the sum of the target multiplication matrix and the noise mask matrix is determined as a target exponential function matrix. It should be noted that the target exponential function matrix obtained is obtained through the second exponential function information.

[0089] Thirdly, matrix multiplication is performed on the target exponential function matrix and the value parameter matrix information to obtain a target exponential product matrix. In practice, firstly, the target exponential function matrix is split through a secret sharing protocol to obtain a split target exponential function matrix. The split target exponential function matrix can include a first target exponential function matrix and a second target exponential function matrix. Here, the relationship between the first target exponential function matrix and the second target exponential function matrix is not limited. For example, the matrix represented by the first target exponential function matrix is the same as the matrix represented by the second target exponential function matrix, or the sum of the matrix represented by the first target exponential function matrix and the matrix represented by the second target exponential function matrix is the same as the matrix represented by the target exponential function matrix. The first target exponential function matrix can represent the share (matrix) of the target exponential function matrix held by the first party participating in the multi-party secure computation (MPC) protocol. The second target exponential function matrix can represent the share (matrix) of the target exponential function matrix held by the second party participating in the multi-party secure computation (MPC) protocol. Then, the result of multiplying the first target exponential function matrix included in the split target exponential function matrix and the first value parameter matrix information included in the value parameter matrix information is taken as a fifth product. The result of multiplying the second exponential function matrix included in the split value parameter matrix information and the second value parameter matrix information included in the value parameter matrix information is taken as a sixth product. Finally, the fifth product and the sixth product are fused through a secret sharing protocol to obtain a target exponential product matrix.

[0090] In the fourth step, the target exponent product matrix and the output parameter matrix information are subjected to matrix multiplication to obtain the encrypted matrix information corresponding to the split fusion matrix. In practice, the target exponent product matrix is subjected to splitting processing through a secret sharing protocol to obtain a split target exponent product matrix. The split target exponent product matrix can include a first target exponent product matrix and a second target exponent product matrix. In this regard, the relationship between the first target exponent product matrix and the second target exponent product matrix is not limited. For example, the matrix represented by the first target exponent product matrix is the same as the matrix represented by the second target exponent product matrix, or the sum of the matrix represented by the first target exponent product matrix and the matrix represented by the second target exponent product matrix is the same as the matrix represented by the target exponent product matrix. The first target exponent product matrix can represent the share (matrix) of the target exponent product matrix held by a first party participating in a multi-party secure computation (MPC) protocol. The second target exponent product matrix can represent the share (matrix) of the target exponent product matrix held by a second party participating in the multi-party secure computation (MPC) protocol. Then, the result of multiplying the first target exponent product matrix in the split target exponent product matrix by the first output parameter matrix information included in the output parameter matrix information is taken as a seventh product. The result of multiplying the second target exponent product matrix in the split target exponent product matrix by the second output parameter matrix information included in the output parameter matrix information is taken as an eighth product. Finally, the seventh product and the eighth product are fused through a secret sharing protocol to obtain the encrypted matrix information.

[0091] As an inventive point of the embodiment of the present disclosure, the above technical solution solves the second technical problem that the determination of the index results of the activation values in the full value domain range consumes more computing resources and takes more time, resulting in a longer time for generating the item warning information, and further causing more to-be-detected items to be accumulated. The reasons for the determination of the index results of the activation values in the full value domain range consuming more computing resources and taking more time, resulting in a longer time for generating the item warning information, and further causing more to-be-detected items to be accumulated are as follows: when detecting the entry and exit items, using the multi-party secure computation (MPC) protocol, a large number of activation values are generated when generating the item warning information, and there are maximum values in the large number of generated activation values, resulting in a large value domain range; when using the SoftMax function to determine the index results of the activation values, because the value threshold range of the activation values is large, the determination of the index results of the activation values in the full value domain range consumes more computing resources and takes more time, resulting in a longer time for generating the item warning information, and further causing more to-be-detected items to be accumulated. If the above factors are solved, the effect of reducing the computing resources consumed for determining the index results of the activation values and shortening the time, shortening the time for generating the item warning information, and further reducing the situation of to-be-detected items being accumulated can be achieved. In order to achieve this effect, the present disclosure performs base conversion on the above target split fusion matrix or the above preset random matrix to replace each value in the above target split fusion matrix or the above preset random matrix with an exponential calculation with a base of 2, instead of determining the results of the exponent with a base of e through the SoftMax function, thereby reducing the computing resources consumed for determining the index results of the activation values and shortening the time. Moreover, the value threshold range of the above target changed query sequence information after the pre-processing is determined, and because the above changed query sequence information is obtained by pre-processing the above warning query sequence information using the special token information set, the maximum activation value at the specified position can be determined. The maximum activation value at the specified position can represent the activation value at the beginning of the sentence and the activation value corresponding to the last two positions of each row of the split fusion matrix. The activation value at the beginning of the sentence can represent the beginning of the sentence corresponding to the above warning query sequence information. Thus, the value threshold range of each activation value is reduced, and the number of comparisons for determining the maximum value is reduced, the comparison time is shortened, and the computing resources consumed are reduced. Thus, the effect of reducing the computing resources consumed for determining the index results of the activation values and shortening the time, shortening the time for generating the item warning information, and further reducing the situation of to-be-detected items being accumulated can be achieved.

[0092] In step 1016, the encrypted item warning information corresponding to the target changed query sequence information is generated according to the encryption matrix information and the second parameter matrix information group.

[0093] In some embodiments, the execution subject can generate the encrypted item early warning information corresponding to the target change query sequence information according to the encrypted matrix information and the second parameter matrix information set. The encrypted item early warning information can represent the prediction result of the pre-trained item early warning information generation model on the early warning query sequence information. The state of the prediction result is in an encrypted state.

[0094] In some optional implementations of some embodiments, the execution subject can generate the encrypted item early warning information corresponding to the target change query sequence information according to the encrypted matrix information and the second parameter matrix information set by the following steps:

[0095] Firstly, the encrypted matrix information is normalized to obtain target encrypted matrix information. The target encrypted matrix information can represent the normalized encrypted matrix information. In practice, firstly, the execution subject can split the encrypted matrix information by a secret sharing protocol to obtain a first split encrypted matrix corresponding to the encrypted matrix information and a second split encrypted matrix corresponding to the encrypted matrix information. The first split encrypted matrix can represent the share (matrix) of the encrypted matrix information held by the first party participating in the multi-party secure computation (MPC) protocol. The second split encrypted matrix can represent the share (matrix) of the encrypted matrix information held by the second party participating in the multi-party secure computation (MPC) protocol. Then, the first split encrypted matrix and the second split encrypted matrix are normalized by a preset normalization processing method to obtain the target encrypted matrix information.

[0096] In a second step, the second parameter matrix information set is split to obtain each target split matrix information. The target split matrix information in each target split matrix information can represent a split dimension-increasing weight matrix, a split gate weight matrix, or a split dimension-reducing weight matrix. The target split matrix information in each target split matrix information can include first target split parameter matrix information and second target split parameter matrix information. The first target split parameter matrix information can represent a share (matrix) of the target split matrix information held by a first party participating in a multi-party computation (MPC) protocol. The second target split parameter matrix information can represent a share (matrix) of the target split matrix information held by a second party participating in the multi-party computation (MPC) protocol. Here, the relationship between the first target split parameter matrix information and the second target split parameter matrix information is not limited. For example, the matrix represented by the first target split parameter matrix information is the same as the matrix represented by the second target split parameter matrix information, or the sum of the matrix represented by the first target split parameter matrix information and the matrix represented by the second target split parameter matrix information is the same as the matrix represented by the target split parameter matrix information. The first target split parameter matrix can represent a first dimension-increasing weight matrix, a first gate weight matrix, or a first dimension-reducing weight matrix. The first dimension-increasing weight matrix can represent a share (matrix) of the dimension-increasing weight matrix held by the first party participating in the multi-party computation (MPC) protocol. The first gate weight matrix can represent a share (matrix) of the gate weight matrix held by the first party participating in the multi-party computation (MPC) protocol. The first dimension-reducing weight matrix can represent a share (matrix) of the dimension-reducing weight matrix held by the first party participating in the multi-party computation (MPC) protocol. The second target split parameter matrix can represent a second dimension-increasing weight matrix, a second gate weight matrix, or a second dimension-reducing weight matrix. The second dimension-increasing weight matrix can represent a share (matrix) of the dimension-increasing weight matrix held by the second party participating in the multi-party computation (MPC) protocol. The second gate weight matrix can represent a share (matrix) of the gate weight matrix held by the second party participating in the multi-party computation (MPC) protocol. The second dimension-reducing weight matrix can represent a share (matrix) of the dimension-reducing weight matrix held by the second party participating in the multi-party computation (MPC) protocol. In practice, the execution subject can split each parameter matrix information in the second parameter matrix information set based on the multi-party computation (MPC) protocol to obtain each target split matrix information.

[0097] In the third step, in response to determining that the preset normalization processing mode is the first normalization processing mode, the encrypted item early warning information corresponding to the target change query sequence information is generated according to the above-mentioned target split matrix information, the above-mentioned target encrypted matrix information and the first activation function information. The first activation function information can represent an activation function protocol based on an additive secret sharing protocol. In practice, the execution subject can execute the following first activation function processing steps through the first activation function information:

[0098] Firstly, the execution subject can perform matrix multiplication between the first split encrypted matrix corresponding to the target encrypted matrix information and the first dimension-increasing weight matrix to obtain first dimension-increasing product information. The activation function protocol based on the additive secret sharing protocol can represent the first activation function processing steps executed by the execution subject.

[0099] Then, the second split encrypted matrix corresponding to the target encrypted matrix information and the second dimension-increasing weight matrix are subjected to matrix multiplication to obtain second dimension-increasing product information.

[0100] After that, the first dimension-increasing product information and the second dimension-increasing product information are fused through a secret sharing protocol to obtain a first fusion matrix.

[0101] Secondly, the first split encrypted matrix corresponding to the target encrypted matrix information and the first gating weight matrix can be subjected to matrix multiplication to obtain first gating product information.

[0102] Then, the second split encrypted matrix corresponding to the target encrypted matrix information and the second gating weight matrix are subjected to matrix multiplication to obtain second gating product information.

[0103] After that, the first gating product information and the second gating product information are fused through a secret sharing protocol to obtain a second fusion matrix.

[0104] Secondly, for each numerical value in the second fusion matrix, the following steps are executed:

[0105] Firstly, the numerical value after the base conversion processing is determined as a first numerical value.

[0106] Secondly, the first numerical value is subjected to truncation processing to obtain a first decimal value corresponding to the decimal part of the first numerical value and a first integer value corresponding to the integer part of the first numerical value. It should be noted that the truncation processing of the first numerical value is the same as the truncation processing of the first base conversion matrix, which will not be described here.

[0107] Thirdly, the first decimal value is updated to obtain an updated decimal value corresponding to the first decimal value. In practice, the execution subject can update the first decimal value to obtain the updated decimal value by using the polynomial approximation method.

[0108] Fourthly, the first integer value is used to generate a first lookup table value and a second lookup table value corresponding to the first integer value. In practice, in response to determining that the first integer value is less than a preset value, firstly, the execution subject can determine a target preset value as the first lookup table value by using a lookup table protocol. Then, a value obtained by performing an exponential operation on the first integer value is determined as the second lookup table value. The preset value can be 0. The target preset value can be 1. Finally, in response to determining that the first integer value is greater than the preset value, a value obtained by performing an exponential operation on the negative of the first integer value is determined as the first lookup table value. Then, the target preset value is determined as the second lookup table value.

[0109] Fifthly, a product of the second lookup table value and the updated decimal value is determined as a second value.

[0110] Sixthly, a sum of the second value and the first lookup table value is determined as a third value.

[0111] Seventhly, the third value is updated to obtain a fourth value corresponding to the third value. In practice, the execution subject can update the third decimal value to obtain the fourth value by using the polynomial approximation method.

[0112] Eighthly, the fourth value is updated to obtain an updated fourth value. In practice, the execution subject can update the fourth value to obtain the updated fourth value by using an iterative optimization processing method. In this regard, the specific type of the iterative optimization processing method is not limited, for example, the iterative optimization processing method can be the Newton approximation method.

[0113] Ninthly, a product of the updated fourth value and the value is determined as a fifth value.

[0114] Subsequently, each obtained fifth value is determined as a changed fusion matrix corresponding to the second fusion matrix.

[0115] Secondly, a matrix obtained by performing element-wise multiplication on the changed fusion matrix and the first fusion matrix is determined as an element-wise multiplication matrix.

[0116] Afterwards, the element-wise multiplication matrix is split by a secret sharing protocol to obtain a split element-wise multiplication matrix, wherein the split element-wise multiplication matrix can include a first element-wise matrix and a second element-wise matrix. Here, the relationship between the first element-wise matrix and the second element-wise matrix is not limited. For example, the first element-wise matrix and the second element-wise matrix are the same, or the sum of the first element-wise matrix and the second element-wise matrix is the same as the element-wise multiplication matrix. The first element-wise matrix can represent the share (matrix) of the element-wise matrix held by the first party participating in the multi-party secure computation (MPC) protocol. The second element-wise matrix can represent the share (matrix) of the element-wise matrix held by the second party participating in the multi-party secure computation (MPC) protocol.

[0117] Afterwards, the product of the first element-wise matrix and the first dimension reduction weight matrix is determined as a first multiplication matrix.

[0118] Secondly, the product of the second element-wise matrix and the second dimension reduction weight matrix is determined as a second multiplication matrix.

[0119] Finally, the first multiplication matrix and the second multiplication matrix are fused by a secret sharing protocol to obtain encrypted product early warning information corresponding to the target change query sequence information.

[0120] Fourthly, in response to determining that the preset normalization processing mode is the second normalization processing mode, the encrypted product early warning information corresponding to the target change query sequence information is generated according to the target split matrix information, the target encrypted matrix information, and the second activation function information. The second activation function information can represent an activation function protocol based on a function secret sharing protocol.

[0121] In practice, the execution subject can perform the following steps:

[0122] First execution step, the execution subject can determine a second fusion matrix corresponding to the target encrypted matrix information. It should be noted that when the preset normalization processing mode is the second normalization processing mode, the way of determining the second fusion matrix corresponding to the target encrypted matrix information is the same as when the preset normalization processing mode is the first normalization processing mode. Here, when the preset normalization processing mode is the second normalization processing mode, the step of determining the second fusion matrix corresponding to the target encrypted matrix information is not repeated.

[0123] Second execution step, for each value in the second fusion matrix, the following second activation function processing step can be performed:

[0124] In a first step, a sum of the value and a preset random value is determined as a sixth value. The preset random value can be a randomly selected value. The activation function protocol based on the function secret sharing protocol can represent the second activation function processing step performed by the execution subject. In a second step, the sixth value is subjected to an intercepting process to obtain an integer part corresponding to the sixth value as a second integer value. In practice, the execution subject can use the modf() function to perform the intercepting process on the sixth value.

[0125] In a third step, a third lookup table value and a fourth lookup table value corresponding to the second integer value are generated according to the second integer value. In practice, in response to determining that the second integer value is greater than the preset value, the execution subject can determine a target preset value as the third lookup table value through a lookup table protocol. In response to determining that the second integer value is less than or equal to the preset value, the execution subject can determine the preset value as the third lookup table value through the lookup table protocol. Then, in response to determining that the second integer value is greater than or equal to a first preset value and less than or equal to a second preset value, the target preset value is determined as the fourth lookup table value. Finally, in response to determining that the second integer value is greater than or equal to the second preset value or less than or equal to the first preset value, the preset value is determined as the fourth lookup table value. The second preset value can be 4. The first preset value can be -4.

[0126] In a fourth step, in response to determining that the third lookup table value is the preset value, the value is changed to the target preset value to obtain a first changed value.

[0127] In a fifth step, in response to determining that the fourth lookup table value is the preset value and the value is less than the first preset value, the value is changed to the first preset value to obtain a second changed value.

[0128] In a sixth step, in response to determining that the fourth lookup table value is the preset value and the value is greater than the second preset value, the value is changed to the second preset value to obtain a third changed value.

[0129] In a seventh step, the value is subjected to an updating process to obtain an updated value. In practice, the execution subject can input the value, the second changed value, or the third changed value into a ReLU activation function to determine a first result. Then, in response to determining that the third lookup table value is the target preset value, a sum of the first result and the value is determined as the updated value. Finally, in response to determining that the third lookup table value is the target preset value, a sum of the first result and the first changed value is determined as the updated value.

[0130] A third execution step determines the obtained updated values as a target change fusion matrix corresponding to the second fusion matrix.

[0131] A fourth execution step generates encrypted item warning information corresponding to the target change query sequence information according to the target change fusion matrix. It should be noted that the step of generating encrypted item warning information corresponding to the target change query sequence information according to the target change fusion matrix is the same as the step of generating encrypted item warning information corresponding to the target change query sequence information according to the change fusion matrix. Therefore, it will not be repeated here.

[0132] Step 1017, decrypting the encrypted item warning information to obtain item warning information as prediction result information.

[0133] In some embodiments, the execution subject can decrypt the encrypted item warning information to obtain item warning information as prediction result information. The item warning information can represent the decrypted prediction result finally output by the pre-trained item warning information generation model. In practice, the execution subject can decrypt the encrypted item warning information through a secret sharing protocol to obtain item warning information as prediction result information.

[0134] Step 102, in response to determining that the prediction result information does not satisfy the preset item qualification condition, controlling the mechanical arm to perform grabbing processing on the target item to intercept the target item.

[0135] In some embodiments, in response to determining that the prediction result information does not satisfy the preset item qualification condition, the execution subject can control the mechanical arm to perform grabbing processing on the target item to intercept the target item. The preset item qualification condition can be that the item warning information represents that the target item can enter or exit the country. Here, the type, number of joints, and rotatable angle of the mechanical arm are not limited and can be adjusted according to actual needs. For example, the mechanical arm can be a SCARA mechanical arm (planar joint mechanical arm). In practice, the execution subject can control the mechanical arm to perform grabbing processing on the target item to intercept the target item through a trajectory planning method. For example, the trajectory planning method can be a point-to-point (PTP) motion control method.

[0136] The above various embodiments of the present disclosure have the following beneficial effects: the prediction result information generation method based on multi-party secure calculation of some embodiments of the present disclosure can shorten the time consumption and reduce the calculation resources consumed in generating the item early warning information. The reason why the time consumption in generating the item early warning information is relatively long and the calculation resources consumed are relatively large is that when the encryption processing of the query process is implemented using the text matching model based on deep learning, the multi-party data interaction is frequent, the data volume is large, and the protocol calculation complexity is high due to the encryption using the multi-party secure calculation (MPC) protocol, resulting in a large communication overhead of the multi-party secure calculation (MPC) protocol, and causing the time consumption in generating the item early warning information based on the text matching model based on deep learning to be relatively long and the calculation resources consumed to be relatively large. Based on this, the prediction result information generation method based on multi-party secure calculation of some embodiments of the present disclosure first, in response to detecting the early warning query sequence information corresponding to the target item, performs the following steps by using the pre-trained item early warning information generation model: according to the pre-set special token information set and the pre-set position, the above-mentioned early warning query sequence information is changed to obtain the changed query sequence information. Thus, the early warning query sequence information of the target user can be changed to obtain the changed early warning query sequence information. Then, the changed query sequence information is split to obtain the split processed changed query sequence information. Thus, the changed early warning query sequence information can be split to obtain the split processed early warning query sequence information. Secondly, the split processed changed query sequence information is normalized to obtain the target changed query sequence information. Thus, the split processed changed query sequence information can be normalized to obtain the target changed query sequence information. Then, each first parameter matrix information in the first parameter matrix information group is split to obtain each split parameter matrix information. Thus, each split parameter matrix can be obtained. Secondly, according to the target changed query sequence information and the each split parameter matrix information, the encryption matrix information corresponding to the target changed query sequence information is generated. Thus, the encryption matrix corresponding to the target changed query sequence information can be obtained. Then, according to the encryption matrix information and the second parameter matrix information group, the encrypted item early warning information corresponding to the target changed query sequence information is generated. Thus, the encrypted item early warning data corresponding to the target changed query sequence information can be obtained. Then, the encrypted item early warning information is decrypted to obtain the item early warning information as the early warning detection result information. Thus, the decrypted item early warning data can be obtained. Finally, in response to determining that the prediction result information does not meet the pre-set item qualified condition, the mechanical arm is controlled to grasp the target item to intercept the target item. Thus, the item can be intercepted.Also because the multi-party secure computation (MPC) protocol is not directly used for multi-party secure computation (MPC) protocol to achieve encryption, but first, according to the preset special word information set, the above early warning query sequence information is changed to change the preset special word information in the preset special word information set to the preset position of the above early warning query sequence information, thereby realizing the change of the above early warning query sequence information. Among them, the above-mentioned preset special word information set can represent each random data whose activation value is greater than the preset threshold value selected from the random data set, thereby the preset special word information with a larger activation value can be fixed to the preset position of the above early warning query sequence information (the preset position represents the position before the beginning of the above early warning query sequence information). Thus, the comparison times of finding the maximum value of each row in the matrix corresponding to the above early warning query sequence information can be reduced, and the number of multi-party data interaction and the number of participating multi-party secure computation (MPC) protocol can be reduced, thereby shortening the time consumed and reducing the calculation resources consumed. Because the normalized processing is performed according to the changed early warning query sequence information, the data range of the changed early warning query sequence information is smaller, so the time consumed for normalizing the changed early warning query sequence information is shorter and the calculation resources consumed are less. Thus, the time consumed for generating the item early warning information can be shortened and the calculation resources consumed can be reduced.

[0137] Further reference Figure 2 , as an implementation of the method shown in the above figures, the present disclosure provides some embodiments of a multi-party secure computation based prediction result information generation method, which corresponds to the method embodiments shown in Figure 1 , the device can be applied to various electronic devices.

[0138] As Figure 2As shown, the multi-party secure computation based prediction result information generation apparatus 200 of some embodiments includes an execution unit 201 and a control unit 202. The execution unit 201 is configured to, in response to detecting the pre-warning query sequence information corresponding to the target item, perform the following steps by using the pre-trained item pre-warning information generation model: performing modification processing on the pre-warning query sequence information according to a preset special token information set and a preset position to obtain modified query sequence information, wherein the parameter information of the pre-trained item pre-warning information generation model includes a first parameter matrix information group and a second parameter matrix information group; performing splitting processing on the modified query sequence information to obtain split processed modified query sequence information; performing normalization processing on the split processed modified query sequence information to obtain target modified query sequence information; performing splitting processing on each first parameter matrix information in the first parameter matrix information group to obtain each split parameter matrix information; generating encrypted matrix information corresponding to the target modified query sequence information according to the target modified query sequence information and the each split parameter matrix information; generating encrypted item pre-warning information corresponding to the target modified query sequence information according to the encrypted matrix information and the second parameter matrix information group; and performing decryption processing on the encrypted item pre-warning information to obtain item pre-warning information as prediction result information. The control unit 202 is configured to, in response to determining that the prediction result information does not satisfy a preset item qualified condition, control a mechanical arm to perform grabbing processing on the target item to intercept the target item.

[0139] It can be understood that the units described in the item pre-warning information generation apparatus 200 correspond to the respective steps in the method described above. Figure 1 Therefore, the operations, features and advantages described above for the method also apply to the item pre-warning information generation apparatus 200 and the units included therein, which will not be described here again.

[0140] Reference is made below to Figure 3 which shows a structural schematic diagram of an electronic device 300 (e.g., a computing device) suitable for implementing some embodiments of the present disclosure. Figure 3 The electronic device shown is merely an example and should not impose any limitation on the functions and use range of the embodiments of the present disclosure.

[0141] As Figure 3As shown, the electronic device 300 can include a processing device 301 (e.g., a central processing unit, a graphics processing unit, etc.) that can perform various appropriate actions and processes according to programs stored in a read-only memory (ROM) 302 or loaded into a random access memory (RAM) 303 from a storage device 308. Various programs and data required for the operation of the electronic device 300 are also stored in the RAM 303. The processing device 301, the ROM 302, and the RAM 303 are connected to each other through a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0142] In general, the following devices can be connected to the I / O interface 305: input devices 306 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; output devices 307 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 308 including, for example, a magnetic tape, a hard disk, etc.; and communication devices 309. The communication devices 309 can allow the electronic device 300 to communicate wirelessly or wired with other devices to exchange data. Although Figure 3 The electronic device 300 is shown with various devices, but it should be understood that not all of the shown devices are required to be implemented or present. More or fewer devices can alternatively be implemented or present. Figure 3 Each block shown in the middle can represent a device or, as desired, multiple devices.

[0143] In particular, processes described above with reference to the flowcharts can be implemented as a computer software program according to some embodiments of the present disclosure. For example, some embodiments of the present disclosure include a computer program product comprising a computer program carried on a computer readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In some such embodiments, the computer program can be downloaded and installed from a network through the communication devices 309, or installed from the storage devices 308, or installed from the ROM 302. When the computer program is executed by the processing device 301, the above-described functions defined in the methods of some embodiments of the present disclosure are performed.

[0144] Note that the computer-readable medium in some embodiments of the present disclosure can be a computer-readable signal medium or a computer-readable storage medium or any combination thereof. The computer-readable storage medium may, for example and without limitation, be an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of the computer-readable storage medium can include, but are not limited to, an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In some embodiments of the present disclosure, the computer-readable storage medium can be any tangible medium that contains or stores a program used by or in connection with an instruction execution system, apparatus, or device. In some embodiments of the present disclosure, the computer-readable signal medium can include a computer-readable program code contained in a data signal communicated in a baseband or as part of a carrier wave. Such a propagated data signal can take any of a variety of forms, including but not limited to electro-magnetic, optical, or any suitable combination thereof. The computer-readable signal medium can also be any computer-readable medium that is not a storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF (radio frequency), and the like, or any suitable combination of the foregoing.

[0145] In some embodiments, the client, server, or both can communicate using any current known or future developed network protocol, such as HTTP (HyperText Transfer Protocol), and can be interconnected with any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include local area networks ("LAN"), wide area networks ("WAN"), the Internet, and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any current known or future developed networks.

[0146] The computer readable medium can be included in the electronic device; or can exist independently of the electronic device. The computer readable medium carries one or more programs, when the one or more programs are executed by the electronic device, the electronic device is caused to: in response to detecting early warning query sequence information corresponding to a target item, through a pre-trained item early warning information generation model, perform the following steps: according to a preset special token information set and a preset position, change the early warning query sequence information to obtain changed query sequence information, wherein parameter information of the pre-trained item early warning information generation model includes a first parameter matrix information group and a second parameter matrix information group; split the changed query sequence information to obtain split processed changed query sequence information; normalize the split processed changed query sequence information to obtain target changed query sequence information; split each first parameter matrix information in the first parameter matrix information group to obtain each split parameter matrix information; generate encrypted matrix information corresponding to the target changed query sequence information according to the target changed query sequence information and the each split parameter matrix information; generate encrypted item early warning information corresponding to the target changed query sequence information according to the encrypted matrix information and the second parameter matrix information group; decrypt the encrypted item early warning information to obtain item early warning information as prediction result information; in response to determining that the prediction result information does not satisfy a preset item qualified condition, control a mechanical arm to perform grabbing processing on the target item to intercept the target item.

[0147] Computer program code for carrying out operations of some embodiments of the present disclosure can be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C++ or the like, and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code can execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider).

[0148] The flow and block diagrams in the drawings represent possible architectural, functional, and operational scenarios of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block can represent a module, a segment, or a portion of code that comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that in some alternative implementations, the functions noted in the blocks can occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently or in the reverse order, depending on the functionality involved. It will also be noted that each block of the block diagrams and / or flowchart illustrations, and combinations thereof, can be implemented by a dedicated hardware-based system that performs the specified functions or operations, or combinations of hardware and software.

[0149] The units described in some embodiments of the present disclosure can be implemented by software or by hardware. The described units can also be arranged in a processor, for example, can be described as: an execution unit and a control unit. Among them, the name of these units does not constitute a limitation to the units themselves in some cases, for example, the control unit can also be described as: "a unit for controlling the robot arm to perform the grabbing processing on the target object to intercept the target object, in response to determining that the prediction result information does not meet the preset qualified condition of the object".

[0150] The functions described above in this document can be performed at least in part by one or more hardware logic components. For example, without limitation, example types of hardware logic components that can be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), system on a chip (SOCs), complex programmable logic devices (CPLDs), etc.

[0151] The above description is merely some preferred embodiments of the present disclosure and a description of the principles of the technology used. Those skilled in the art should understand that the scope of the application involved in the embodiments of the present disclosure is not limited to the technical solutions formed by the specific combinations of the above technical features, and should also cover other technical solutions formed by any combination of the above technical features or equivalent features without departing from the above inventive concept. For example, the above features are replaced with the technical features disclosed in the embodiments of the present disclosure (but not limited to) having similar functions to form technical solutions.

Claims

1. A method for generating prediction result information based on multi-party secure computation, comprising: In response to the detection of a warning query sequence information corresponding to the target item, a model is generated using pre-trained item warning information, and the following steps are performed: Based on a preset set of special word information and a preset position, the warning query sequence information is modified to obtain modified query sequence information. The parameter information of the pre-trained item warning information generation model includes a first parameter matrix information group and a second parameter matrix information group. The change query sequence information is split to obtain split change query sequence information; The split change query sequence information is normalized to obtain the target change query sequence information; Each first parameter matrix information in the first parameter matrix information group is split to obtain each split parameter matrix information; Based on the target change query sequence information and the various split parameter matrix information, generate encrypted matrix information corresponding to the target change query sequence information; Based on the encryption matrix information and the second parameter matrix information group, generate encrypted item warning information corresponding to the target change query sequence information; The encrypted item warning information is decrypted to obtain the item warning information as the prediction result information. In response to determining that the predicted result information does not meet the preset item qualification conditions, the robotic arm is controlled to grasp the target item in order to intercept the target item.

2. The method according to claim 1, wherein, The method further includes: Based on the preset dataset information and the pre-trained item warning information generation model, a set of word element information and a set of activation value information corresponding to the set of word element information are generated. Based on the activation value information set, each word information in the word information set that meets the preset threshold condition is determined as a special word information set as a preset special word information set.

3. The method according to claim 2, wherein, The item warning information generation model based on preset dataset information and pre-trained model generates a set of lexical information and a set of activation value information corresponding to the lexical information set, including: The preset dataset information is sampled and processed to obtain the sampled and processed preset dataset information as the sampling sequence information of the item warning information generation model; Based on the sampled sequence information and the pre-trained item warning information generation model, a set of lexical information corresponding to the preset dataset information is generated; Based on the set of lexical information, a set of activation value information corresponding to the set of lexical information is generated.

4. The method according to claim 1, wherein, The step of modifying the warning query sequence information based on a preset set of special word element information and a preset position to obtain modified query sequence information includes: The preset set of special word elements is added to the preset position of the warning query sequence information to obtain the target sequence information; The target sequence information is modified to obtain modified query sequence information.

5. The method according to claim 1, wherein, The normalization process performed on the split change query sequence information to obtain the target change query sequence information includes: In response to determining that the preset normalization processing method is the first normalization processing method, the square matrix information and dimension information corresponding to the split-processed change query sequence information are generated according to the split-processed change query sequence information; Based on the dimensions and information and the preset interpolation method, generate interpolation result information corresponding to the modified query sequence information after the splitting process; The product of the square matrix information and the interpolation result information is determined as the target change query sequence information; In response to determining that the preset normalization processing method is the second normalization processing method, the target change sequence information after splitting is normalized to obtain the target change query sequence information.

6. The method according to claim 1, wherein, The step of generating encrypted item warning information corresponding to the target change query sequence information based on the encryption matrix information and the second parameter matrix information group includes: The encryption matrix information is normalized to obtain the target encryption matrix information; The second parameter matrix information group is split to obtain each target split matrix information. The target split matrix information in each target split matrix information includes a first target split parameter matrix and a second target split parameter matrix. The target split matrix information in each target split matrix information is a split-up weight matrix, a split-gated weight matrix, or a split-down weight matrix. In response to determining that the preset normalization processing method is the first normalization processing method, an encrypted item warning information corresponding to the target change query sequence information is generated based on the target splitting matrix information, the target encryption matrix information, and the first activation function information. In response to determining that the preset normalization processing method is the second normalization processing method, encrypted item warning information corresponding to the target change query sequence information is generated based on the target splitting matrix information, the target encryption matrix information, and the second activation function information.

7. A device for generating prediction result information based on multi-party secure computation, comprising: An execution unit is configured to, in response to the detection of a warning query sequence information corresponding to a target item, perform the following steps using a pre-trained item warning information generation model: modifying the warning query sequence information according to a preset set of special terminology information and a preset position to obtain modified query sequence information, wherein the parameter information of the pre-trained item warning information generation model includes a first parameter matrix information group and a second parameter matrix information group; splitting the modified query sequence information to obtain split modified query sequence information; normalizing the split modified query sequence information to obtain target modified query sequence information; splitting each first parameter matrix information in the first parameter matrix information group to obtain various split parameter matrix information; generating encrypted matrix information corresponding to the target modified query sequence information based on the target modified query sequence information and the various split parameter matrix information; generating encrypted item warning information corresponding to the target modified query sequence information based on the encrypted matrix information and the second parameter matrix information group; and decrypting the encrypted item warning information to obtain item warning information as prediction result information. The control unit is configured to control the robotic arm to grab the target item in response to determining that the prediction result information does not meet the preset item qualification conditions, so as to intercept the target item.

8. An electronic device, comprising: One or more processors; A storage device on which one or more programs are stored; When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1 to 6.

9. A computer-readable medium having a computer program stored thereon, wherein, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 6.

Images