CN115526266A - Model training method and device, and business prediction method and device - Google Patents
Model training method and device, and business prediction method and deviceDownload PDFInfo
- Publication number
- CN115526266A CN115526266ACN202211272252.0ACN202211272252ACN115526266ACN 115526266 ACN115526266 ACN 115526266ACN 202211272252 ACN202211272252 ACN 202211272252ACN 115526266 ACN115526266 ACN 115526266A
- Authority
- CN
- China
- Prior art keywords
- parameter
- training
- parameter value
- neural network
- current
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
Description
Translated fromChinese技术领域technical field
本说明书一个或多个实施例涉及计算机技术，尤其涉及模型训练方法和装置、业务预测方法和装置。One or more embodiments of this specification relate to computer technology, especially to a model training method and device, and a service prediction method and device.
背景技术Background technique
人工神经网络是一种应用类似于大脑神经突触联接的结构进行信息处理的模型。神经网络模型是一种运算模型，由大量的节点(或称神经元)相互联接构成。每个节点代表一种特定的输出函数，称为激励函数(activation function)。每两个节点间的连接都代表一个对于通过该连接信号的加权值，称之为权重，这相当于人工神经网络的记忆。网络的输出则依网络的连接方式，权重值和激励函数的不同而不同。神经网络模型可实现函数逼近、数据聚类、模式分类、优化计算等功能。因此，神经网络模型广泛应用于人工智能、自动控制、机器人、统计学等领域的信息处理中。比如应用神经网络模型进行支付业务中的风险控制等。An artificial neural network is a model of information processing that uses a structure similar to the brain's synaptic connections. The neural network model is an operational model, which is composed of a large number of nodes (or neurons) connected to each other. Each node represents a specific output function called an activation function. Each connection between two nodes represents a weighted value for the signal passing through the connection, called weight, which is equivalent to the memory of the artificial neural network. The output of the network varies according to the way the network is connected, the weight value and the activation function. The neural network model can realize functions such as function approximation, data clustering, pattern classification, and optimization calculation. Therefore, the neural network model is widely used in information processing in the fields of artificial intelligence, automatic control, robotics, and statistics. For example, apply the neural network model for risk control in the payment business, etc.
神经网络模型中包括多种参数。通过训练神经网络模型可以确定参数的取值。参数取值的不同，将会大大影响神经网络模型的性能。Various parameters are included in the neural network model. The value of the parameter can be determined by training the neural network model. Different parameter values will greatly affect the performance of the neural network model.
因此，如何更好地利用神经网络模型来进行业务预测，从而得到更为准确的业务预测结果，是一个亟待解决的问题。Therefore, how to make better use of the neural network model for business prediction, so as to obtain more accurate business prediction results, is an urgent problem to be solved.
发明内容Contents of the invention
本说明书一个或多个实施例描述了模型训练方法和装置、业务预测方法和装置，能够更好地利用神经网络模型来进行业务预测。One or more embodiments of this specification describe a model training method and device, and a business forecasting method and device, which can better utilize a neural network model for business forecasting.
根据第一方面，提供了一种神经网络模型的训练方法，其中，包括：According to the first aspect, a training method of a neural network model is provided, which includes:
根据历史业务数据，获取训练样本数据；Obtain training sample data based on historical business data;
在每一轮训练中均执行：In each round of training execute:
将训练样本数据输入所述神经网络模型中，以对所述神经网络模型中每一个参数的参数值进行调整；以及inputting training sample data into the neural network model to adjust the parameter value of each parameter in the neural network model; and
检测本轮训练是否满足参数获取条件，如果是，则记录本轮训练得到的神经网络模型中每一个参数的当前参数值；Detect whether the current round of training meets the parameter acquisition conditions, and if so, record the current parameter value of each parameter in the neural network model obtained in the current round of training;
在各轮训练结束后，针对神经网络模型的每一个参数，根据记录的该参数的至少一个当前参数值，得到该参数对应的最终参数值；After each round of training is over, for each parameter of the neural network model, according to at least one current parameter value of the recorded parameter, the final parameter value corresponding to the parameter is obtained;
将神经网络模型中每一个参数的参数值设置为该参数对应的最终参数值。The parameter value of each parameter in the neural network model is set as the final parameter value corresponding to the parameter.
其中，所述检测本轮训练是否满足参数获取条件包括：Wherein, the detection of whether the current round of training meets the parameter acquisition conditions includes:
根据预先设置的采样周期，检测本轮是否达到了该采样周期，如果是，则确定本轮训练满足参数获取条件；其中，所述采样周期表征每N轮训练获取一次参数值；N为不小于1的正整数；According to the preset sampling period, detect whether the current round has reached the sampling period, and if so, determine that the current round of training satisfies the parameter acquisition condition; wherein, the sampling period represents that every N rounds of training obtain a parameter value; N is not less than positive integer of 1;
或者，or,
判断本轮得到的各参数的当前参数值是否优于上一次记录的各参数的当前参数值，如果是，则确定本轮训练满足参数获取条件。It is judged whether the current parameter value of each parameter obtained in this round is better than the current parameter value of each parameter recorded last time, and if so, it is determined that the current round of training satisfies the parameter acquisition condition.
其中，在预先设置的第L轮及其之后的每一轮中，均执行所述检测本轮训练是否满足参数获取条件的步骤，在预先设置的第L轮之前的每一轮中，不执行所述检测本轮训练是否满足参数获取条件的步骤；其中，L为大于1的正整数。Wherein, in the pre-set Lth round and each subsequent round, the step of detecting whether the current round of training meets the parameter acquisition conditions is performed, and in each round before the pre-set Lth round, the step of not performing The step of detecting whether the current round of training satisfies the parameter acquisition condition; wherein, L is a positive integer greater than 1.
其中，所述根据记录的该参数的至少一个当前参数值得到该参数对应的最终参数值，包括：Wherein, said obtaining the final parameter value corresponding to the parameter according to at least one current parameter value of the recorded parameter includes:
从记录的该参数的至少一个当前参数值中，选取最大的参数值，将该最大的参数值作为该参数对应的最终参数值。From the recorded at least one current parameter value of the parameter, select the largest parameter value, and use the largest parameter value as the final parameter value corresponding to the parameter.
其中，所述根据记录的该参数的至少一个当前参数值得到该参数对应的最终参数值，包括：Wherein, said obtaining the final parameter value corresponding to the parameter according to at least one current parameter value of the recorded parameter includes:
确定所记录的该参数的每一个当前参数值对应的权重值；Determine the weight value corresponding to each current parameter value of the recorded parameter;
将该参数的各个当前参数值进行加权平均，以得到该参数对应的所述最终参数值。Each current parameter value of the parameter is weighted and averaged to obtain the final parameter value corresponding to the parameter.
其中，所述确定所记录的该参数的每一个当前参数值对应的权重值，包括：Wherein, the determination of the weight value corresponding to each current parameter value of the recorded parameter includes:
将每一个当前参数值对应的权重值均设置为
或者，or,
根据得到每一个当前参数值时所处的轮数，设置每一个当前参数值对应的权重值，其中，得到一个当前参数值时所处的轮数在前时，该当前参数值对应的权重值较小，得到一个当前参数值时所处的轮数在后时，该当前参数值对应的权重值较大。According to the number of rounds in which each current parameter value is obtained, set the weight value corresponding to each current parameter value, wherein, when the number of rounds in which a current parameter value is obtained is in the first place, the weight value corresponding to the current parameter value Smaller, when the number of rounds in which a current parameter value is obtained is later, the weight value corresponding to the current parameter value is larger.
根据第二方面，提供了业务预测方法，其中，包括：According to a second aspect, a business forecasting method is provided, comprising:
获取待预测的业务数据；Obtain business data to be predicted;
将待预测的业务数据输入神经网络模型中；所述神经网络模型是利用本说明书实施例的方法训练出的；Input the business data to be predicted into the neural network model; the neural network model is trained by using the method of the embodiment of this specification;
得到该神经网络模型输出的对该待预测业务数据的识别结果；Obtaining the identification result of the business data to be predicted output by the neural network model;
根据第三方面，提供了模型训练装置，该装置包括：According to a third aspect, a model training device is provided, the device comprising:
样本获取模块，配置为根据历史业务数据，获取训练样本数据；The sample acquisition module is configured to acquire training sample data according to historical business data;
训练执行模块，配置为在每一轮训练中均执行：将训练样本数据输入所述神经网络模型中，以对所述神经网络模型中每一个参数的参数值进行调整；以及检测本轮训练是否满足参数获取条件，如果是，则记录本轮训练得到的神经网络模型中每一个参数的当前参数值；The training execution module is configured to execute in each round of training: input training sample data into the neural network model to adjust the parameter value of each parameter in the neural network model; and detect whether the current round of training is Satisfy the parameter acquisition condition, if yes, record the current parameter value of each parameter in the neural network model obtained in this round of training;
参数值确定模块，配置为在各轮训练结束后，针对神经网络模型的每一个参数，根据记录的该参数的至少一个当前参数值，得到该参数对应的最终参数值；The parameter value determination module is configured to obtain the final parameter value corresponding to the parameter according to at least one current parameter value of the recorded parameter for each parameter of the neural network model after each round of training ends;
模型生成模块，配置为将神经网络模型中每一个参数的参数值设置为该参数对应的最终参数值。The model generation module is configured to set the parameter value of each parameter in the neural network model as the final parameter value corresponding to the parameter.
根据第四方面，提供了业务预测装置，其中，包括：According to the fourth aspect, a service prediction device is provided, which includes:
业务数据获取模块，配置为获取待预测的业务数据；A business data acquisition module configured to acquire business data to be predicted;
输入模块，配置为将待预测的业务数据输入神经网络模型中；所述神经网络模型是本说明书实施例的模型训练装置而训练出的；The input module is configured to input the business data to be predicted into the neural network model; the neural network model is trained by the model training device in the embodiment of this specification;
输出模块，配置为得到该神经网络模型输出的对该待预测业务数据的识别结果。The output module is configured to obtain the recognition result of the service data to be predicted output by the neural network model.
根据第五方面，提供了一种计算设备，包括存储器和处理器，所述存储器中存储有可执行代码，所述处理器执行所述可执行代码时，实现本说明书任一实施例所述的方法。According to a fifth aspect, there is provided a computing device, including a memory and a processor, wherein executable code is stored in the memory, and when the processor executes the executable code, the method described in any embodiment of this specification is implemented. method.
本说明书实施例提供的模型训练方法及装置、业务预测方法及装置，至少具有如下有益效果：The model training method and device, service prediction method and device provided in the embodiments of this specification have at least the following beneficial effects:
1、只需要训练一个神经网络模型，而不需要如现有技术中训练多个神经网络模型，这样，在后续部署及进行业务预测时，只需要使用训练出的该一个神经网络模型即可，因此，大大节省了系统资源。1. Only one neural network model needs to be trained, instead of multiple neural network models as in the prior art. In this way, only the trained neural network model needs to be used during subsequent deployment and business forecasting. Therefore, system resources are greatly saved.
2、不是采用最后阶段得到的某一轮神经网络模型的参数的参数值，而是每当在一轮训练中满足参数获取条件时，就记录一次参数的当前参数值，最终得到的神经网络模型的参数的参数值不是在一轮训练中得到的，而是根据多轮训练中得到的多个当前参数值而综合得到的。也就是说，本说明书实施例中最终训练出的神经网络模型中，每一个参数的参数值体现了训练过程的不同阶段中神经网络模型的特点，就好像对于一个学生成绩的考量不是只看期末考试一次的成绩，而是综合考量该学生在本学期多次考试中的多次成绩。因此，训练出的神经网络模型更能符合训练要求，学习到的内容更加丰富，后续则能更为准确地进行业务预测，比如进行风险评估等。2. Instead of using the parameter value of a certain round of neural network model parameters obtained in the final stage, the current parameter value of the parameter is recorded every time the parameter acquisition condition is met in a round of training, and the final neural network model obtained The parameter value of the parameter is not obtained in one round of training, but is obtained comprehensively according to multiple current parameter values obtained in multiple rounds of training. That is to say, in the neural network model finally trained in the embodiment of this specification, the parameter value of each parameter reflects the characteristics of the neural network model in different stages of the training process, just as the consideration of a student's grade is not only at the end of the semester The result of one test is not used, but the student's multiple scores in multiple tests in this semester are considered comprehensively. Therefore, the trained neural network model can better meet the training requirements, the learned content is richer, and subsequent business predictions, such as risk assessment, can be performed more accurately.
附图说明Description of drawings
为了更清楚地说明本说明书实施例或现有技术中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图是本说明书的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。In order to more clearly illustrate the technical solutions in the embodiments of this specification or the prior art, the following will briefly introduce the drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are For some embodiments of this specification, those skilled in the art can also obtain other drawings based on these drawings without creative work.
图1是本说明书一个实施例中神经网络模型的训练方法的流程图。FIG. 1 is a flowchart of a training method for a neural network model in an embodiment of the present specification.
图2是本说明书一个实施例中业务预测方法的流程图。Fig. 2 is a flow chart of a service forecasting method in an embodiment of this specification.
图3是本说明书一个实施例中神经网络模型的训练装置的结构示意图。Fig. 3 is a schematic structural diagram of a training device for a neural network model in an embodiment of the present specification.
图4是本说明书一个实施例中业务预测装置的结构示意图。Fig. 4 is a schematic structural diagram of a service prediction device in an embodiment of the present specification.
具体实施方式detailed description
如前所述，通过训练神经网络模型可以确定参数的取值。参数取值的不同，将会大大影响神经网络模型的性能。在现有技术中，为了更好地进行业务预测，得到准确的业务预测结果，通常都会部署多个神经网络模型，该多个神经网络模型具有相同的结构，但是其参数却具有不同的参数取值。利用该多个神经网络模型分别得到各自的业务预测结果，然后利用得到的多个业务预测结果综合得到(比如利用投票法或者平均法)最终的业务预测结果。因为使用了参数值不同的多个神经网络模型来对一个业务进行综合预测，因此，能够更为充分地考虑不同参数值对神经网络模型的影响，从而得到更优的业务预测结果。As mentioned above, the values of the parameters can be determined by training the neural network model. Different parameter values will greatly affect the performance of the neural network model. In the prior art, in order to better perform business prediction and obtain accurate business prediction results, multiple neural network models are usually deployed. The multiple neural network models have the same structure, but their parameters have different parameter values. value. The multiple neural network models are used to obtain respective service forecast results, and then the multiple service forecast results are used to synthesize (for example, use a voting method or an average method) to obtain a final service forecast result. Because multiple neural network models with different parameter values are used to comprehensively predict a service, the impact of different parameter values on the neural network models can be more fully considered, thereby obtaining better service forecast results.
然而，神经网络模型在部署、运行中都会占用较多的资源。为了进行业务预测，将多个神经网络模型同步部署到线上会产生极大的消耗，有时甚至会导致系统无法正常运行。However, the neural network model will occupy more resources during deployment and operation. For business prediction, synchronously deploying multiple neural network models online will cause great consumption, and sometimes even cause the system to fail to operate normally.
下面结合附图，对本说明书提供的方案进行描述。The solutions provided in this specification will be described below in conjunction with the accompanying drawings.
图1是本说明书一个实施例中神经网络模型的训练方法的流程图。该方法的执行主体为神经网络模型的训练装置。可以理解，该方法也可以通过任何具有计算、处理能力的装置、设备、平台、设备集群来执行。参见图1，该方法包括：FIG. 1 is a flowchart of a training method for a neural network model in an embodiment of the present specification. The subject of execution of the method is the training device of the neural network model. It can be understood that the method can also be executed by any device, device, platform, or device cluster that has computing and processing capabilities. Referring to Figure 1, the method includes:
步骤101：根据历史业务数据，获取训练样本数据。Step 101: Obtain training sample data according to historical service data.
在每一轮训练中均执行步骤103、步骤105、步骤107及步骤109：
步骤103：将训练样本数据输入神经网络模型中，以对神经网络模型中每一个参数的参数值进行调整；Step 103: Input training sample data into the neural network model to adjust the parameter value of each parameter in the neural network model;
步骤105：检测本轮训练是否满足参数获取条件，如果是，执行步骤107，否则，执行步骤109；Step 105: Detect whether the current round of training satisfies the parameter acquisition condition, if yes, execute
步骤107：记录本轮训练得到的神经网络模型中每一个参数的当前参数值。Step 107: Record the current parameter value of each parameter in the neural network model obtained in this round of training.
步骤109：判断所有轮训练是否结束，如果是，执行步骤111，否则，返回步骤103。Step 109: Determine whether all rounds of training are over, if yes, execute
步骤111：在所有轮训练结束后，针对神经网络模型的每一个参数，根据记录的该参数的至少一个当前参数值，得到该参数对应的最终参数值。Step 111: After all rounds of training are finished, for each parameter of the neural network model, according to at least one current parameter value recorded for the parameter, obtain the final parameter value corresponding to the parameter.
步骤113：将神经网络模型中每一个参数的参数值设置为该参数对应的最终参数值。Step 113: Set the parameter value of each parameter in the neural network model as the final parameter value corresponding to the parameter.
根据上述图1所示的流程可以看出，在本说明书实施例中，只需要训练一个神经网络模型，而不需要如现有技术中训练多个神经网络模型，这样，在后续部署及进行业务预测时，只需要使用训练出的该一个神经网络模型即可，因此，大大节省了系统资源。According to the process shown in Figure 1 above, it can be seen that in the embodiment of this specification, only one neural network model needs to be trained, instead of multiple neural network models as in the prior art. In this way, in the subsequent deployment and business When predicting, only the trained neural network model needs to be used, thus greatly saving system resources.
同时，在图1所示的训练神经网络模型的过程中，不是采用最后阶段得到的某一轮神经网络模型的参数的参数值，而是每当在一轮训练中满足参数获取条件时，就记录一次参数的当前参数值，最终得到的神经网络模型的参数的参数值不是在一轮训练中得到的，而是根据多轮训练中得到的多个当前参数值而综合得到的。也就是说，本说明书实施例中最终训练出的神经网络模型中，每一个参数的参数值体现了训练过程的不同阶段中神经网络模型的特点，就好像对于一个学生成绩的考量不是只看期末考试一次的成绩，而是综合考量该学生在本学期多次考试中的多次成绩。因此，训练出的神经网络模型更能符合训练要求，学习到的内容更加丰富，后续则能更为准确地进行业务预测，比如进行风险评估等。At the same time, in the process of training the neural network model shown in Figure 1, instead of using the parameter values of the parameters of a certain round of neural network model obtained in the last stage, whenever the parameter acquisition conditions are met in a round of training, the The current parameter value of the parameter is recorded once, and the final parameter value of the neural network model is not obtained in one round of training, but is obtained comprehensively based on multiple current parameter values obtained in multiple rounds of training. That is to say, in the neural network model finally trained in the embodiment of this specification, the parameter value of each parameter reflects the characteristics of the neural network model in different stages of the training process, just as the consideration of a student's grade is not only at the end of the semester The result of one test is not used, but the student's multiple scores in multiple tests in this semester are considered comprehensively. Therefore, the trained neural network model can better meet the training requirements, the learned content is richer, and subsequent business predictions, such as risk assessment, can be performed more accurately.
下面结合具体的例子对上述图1中每一个步骤分别进行说明。In the following, each step in the above-mentioned FIG. 1 will be described separately in combination with specific examples.
首先对于步骤101：根据历史业务数据，获取训练样本数据。Firstly, for step 101: according to the historical business data, obtain training sample data.
本说明书实施例的方法可以应用于多种业务场景中，比如人脸识别业务场景，也就是说，训练出一个用于人脸识别的神经网络模型；再如风控场景，也就是说训练出一个用于进行诸如交易等的风险识别的神经网络模型。The method in the embodiment of this specification can be applied to various business scenarios, such as the face recognition business scenario, that is, training a neural network model for face recognition; another example is the risk control scenario, that is to say, training a neural network model A neural network model for risk identification such as trading.
相应地，训练样本数据则是对应业务场景中的样本数据。比如为带有标签(真脸或者假脸)的人脸图像，或者为带有标签(有风险或者无风险)的交易数据等。Correspondingly, the training sample data is the sample data in the corresponding business scenario. For example, it is a face image with a label (real face or fake face), or a transaction data with a label (risky or risk-free).
当需要训练一个神经网络模型时，可以首先设置该模型的结构，比如输入层、中间层、输出层的结构。When you need to train a neural network model, you can first set the structure of the model, such as the structure of the input layer, intermediate layer, and output layer.
之后，则需要通过多轮比如1万轮的训练，来确定该神经网络模型中每一个参数的参数值。After that, multiple rounds of training, such as 10,000 rounds, are required to determine the parameter value of each parameter in the neural network model.
接下来对于步骤103：在每一轮训练中，将训练样本数据输入当前训练的神经网络模型中，以对神经网络模型中每一个参数的参数值进行调整。Next to step 103: in each round of training, input the training sample data into the neural network model currently being trained, so as to adjust the parameter value of each parameter in the neural network model.
训练的过程包括对神经网络模型中参数的参数值进行调整，以期望将参数值调整为最优，提高神经网络模型的性能。The training process includes adjusting the parameter values of the parameters in the neural network model, so as to adjust the parameter values to be optimal and improve the performance of the neural network model.
接下来对于步骤105：在每一轮训练中，检测本轮训练是否满足参数获取条件。Next to step 105: in each round of training, check whether the current round of training satisfies the parameter acquisition condition.
如前所述，在本说明书实施例中，最终得到的神经网络模型的参数的参数值不是在一轮训练中得到的，而是根据在多轮训练中得到的多个当前参数值而综合得到的。因此，在本步骤105中，检测在本轮训练中是否满足参数获取条件，只要满足，就通过步骤107记录一次本轮训练出的当前神经网络模型中每一个参数的当前参数值，从而体现在本轮训练这一阶段中神经网络模型的特点。As mentioned above, in the embodiment of this specification, the parameter values of the parameters of the finally obtained neural network model are not obtained in one round of training, but are obtained comprehensively based on multiple current parameter values obtained in multiple rounds of training of. Therefore, in this
在本步骤105中，检测本轮训练是否满足参数获取条件的实现方式至少包括：In this
方式一、参数获取条件对应一个周期，记为采样周期。每训练N轮，就认为满足了一次参数获取条件。Method 1. The parameter acquisition condition corresponds to a period, which is recorded as the sampling period. Every N rounds of training, it is considered that the parameter acquisition condition is met once.
在该方式一中，检测本轮训练是否满足参数获取条件的步骤可以包括：In the first way, the step of detecting whether the current round of training satisfies the parameter acquisition conditions may include:
步骤1051：根据预先设置的采样周期，检测本轮是否达到了该采样周期，如果是，则确定本轮训练满足参数获取条件；其中，N为不小于1的正整数，所述采样周期表征每隔N轮获取一次参数值；如果否，则认为本轮训练不满足参数获取条件。Step 1051: According to the preset sampling period, check whether the current round has reached the sampling period, and if so, determine that the current round of training meets the parameter acquisition conditions; where N is a positive integer not less than 1, and the sampling period represents each The parameter value is obtained every N rounds; if not, it is considered that the current round of training does not meet the parameter acquisition conditions.
举例说明。比如，预先设置的采样周期表征每隔10轮获取一次参数值，那么，在第10、20、30、40轮……等每隔10轮则会满足一次参数获取条件，即每隔10轮需要记录一次本轮得到的参数的当前参数值。比如在步骤1051中，检测到本轮为第9轮，因此未达到每10轮的采样周期，则不满足参数获取条件。再如，在步骤1051中，检测到本轮为第30轮，因此达到了每10轮的采样周期，则满足参数获取条件。for example. For example, the pre-set sampling period means that the parameter value is obtained every 10 rounds. Then, in the 10th, 20th, 30th, 40th round... etc., the parameter acquisition condition will be satisfied every 10th round, that is, every 10th round requires Record the current parameter value of the parameter obtained in this round once. For example, in step 1051, it is detected that the current round is the ninth round, so the sampling period of every 10th round is not reached, and the parameter acquisition condition is not met. For another example, in step 1051, it is detected that the current round is the 30th round, so the sampling period of every 10 rounds is reached, and the parameter acquisition condition is met.
方式二、参数获取条件对应参数值变优，即，每当在一轮中得到的各个参数的参数值优于上一次记录的参数值，则认为满足了一次参数获取条件。Method 2: The corresponding parameter value of the parameter acquisition condition becomes better, that is, whenever the parameter value of each parameter obtained in one round is better than the parameter value recorded last time, it is considered that the parameter acquisition condition is met once.
在该方式二中，检测本轮训练是否满足参数获取条件的步骤可以包括：In the second method, the step of detecting whether the current round of training satisfies the parameter acquisition conditions may include:
步骤1053：在本轮训练结束后，判断本轮得到的神经网络模型中各参数的当前参数值是否优于上一次记录的该各参数的当前参数值，如果是，则确定本轮训练满足参数获取条件。Step 1053: After the current round of training is over, judge whether the current parameter value of each parameter in the neural network model obtained in this round is better than the current parameter value of each parameter recorded last time, if yes, then determine that the current round of training meets the parameters Get conditions.
接下来对于步骤107：记录本轮训练得到的神经网络模型中每一个参数的当前参数值。Next to step 107: record the current parameter value of each parameter in the neural network model obtained in this round of training.
举例说明上述步骤105及步骤107的处理。比如，针对神经网络模型，在第80轮训练中记录了该模型中各个参数的当前参数值。之后，以第80轮记录的各个参数的参数值为比较基准，如果第81轮训练中得到的各个参数的当前参数值不优于第80轮训练中记录的各个参数的当前参数值，则不会记录第81轮训练中得到的各个参数的当前参数值。依次执行，比如第82轮、第83轮以及第84轮训练中得到的各个参数的当前参数值均不优于第80轮训练中记录的各个参数的当前参数值，则也不会记录第82轮、第83轮以及第84轮训练中得到的各个参数的当前参数值。如果第85轮训练中得到的各个参数的当前参数值优于上一次记录(即在第80轮训练中记录)的各个参数的当前参数值，则会记录第85轮训练中得到的各个参数的当前参数值。之后，以第85轮记录的各个参数的参数值为比较基准，如果第86轮训练中得到的各个参数的当前参数值优于第85轮训练时记录的各个参数的当前参数值，则记录第86轮训练中得到的各个参数的当前参数值，否则不记录。The processing of the
在本说明书实施例中，针对训练样本数据，可以通过神经网络模型输出的识别结果与标签之间的相似程度来判断一个参数的当前参数值是否优于上一次记录的当前参数值。比如，在第80轮训练中，神经网络模型输出的训练样本数据的识别结果(比如为分值0.8)与该训练样本数据的标签(比如为1)之间的相似程度为0.8，那么，在第81轮、82轮、第83轮以及第84轮训练中，如果神经网络模型输出的训练样本数据的识别结果与该训练样本数据的标签之间的相似程度均小于0.8，则说明第81轮、82轮、第83轮以及第84轮训练中得到的各个参数的当前参数值均不优于上一次记录(即在第80轮训练中记录)的各个参数的当前参数值。在第85轮训练中，如果神经网络模型输出的训练样本数据的各个识别结果(或者平均识别结果)与该训练样本数据的标签之间的相似程度大于0.8，则说明第85轮训练中得到的各个参数的当前参数值优于上一次记录(即在第80轮训练中记录)的各个参数的当前参数值，则会记录在第85轮训练中得到神经网络模型的各个参数的当前值。In this embodiment of the specification, for the training sample data, whether the current parameter value of a parameter is better than the current parameter value recorded last time can be judged by the similarity between the recognition result output by the neural network model and the label. For example, in the 80th round of training, the similarity between the recognition result of the training sample data output by the neural network model (for example, a score of 0.8) and the label of the training sample data (for example, 1) is 0.8, then, in In the 81st, 82nd, 83rd and 84th rounds of training, if the similarity between the recognition results of the training sample data output by the neural network model and the label of the training sample data is less than 0.8, it means that the 81st round The current parameter value of each parameter obtained in the 1st, 82nd, 83rd and 84th rounds of training is not better than the current parameter value of each parameter recorded last time (that is, recorded in the 80th round of training). In the 85th round of training, if the similarity between each recognition result (or average recognition result) of the training sample data output by the neural network model and the label of the training sample data is greater than 0.8, it means that the 85th round of training obtained If the current parameter value of each parameter is better than the current parameter value of each parameter recorded last time (that is, recorded in the 80th round of training), the current value of each parameter of the neural network model obtained in the 85th round of training will be recorded.
接下来对于步骤109：判断所有轮训练是否结束，如果是，执行步骤111，否则，返回步骤103。Next to step 109: judge whether all rounds of training are over, if yes, execute
在步骤109中，判断所有轮的训练是否结束的方法可以是判断当前训练的轮数是否到达了预先设置的训练轮数阈值，比如，预先设置训练100轮，那么，如果本轮训练是第100轮训练，则说明所有轮训练结束。In
在步骤109中，判断所有轮的训练是否结束的方法也可以是判断神经网络模型是否收敛，比如损失函数是否达到了预设要求，如果是，则说明所有轮训练结束。In
通常，在神经网络模型的训练中，训练的轮数越多，神经网络模型的性能越优，在刚开始训练的时候，神经网络模型的性能一般不能满足要求，比如总共训练1万轮，前8千轮训练中，神经网络模型的性能一般不能满足要求，而后2千轮训练中神经网络模型的性能越来越优，因此，在本说明书实施例中，可以是在预先设置的第L轮及其之后的每一轮中，在执行完步骤103之后，也执行本步骤105及步骤107的处理；而在预先设置的第L轮之前的每一轮中在执行完步骤103之后不执行该步骤105及步骤107的处理，直接执行步骤109；其中，L为大于1的正整数，比如L为8千，即，在第8千轮之前的每一轮训练中，执行步骤103，但不执行步骤105及步骤107的处理，而从第8千轮至第1万轮的每一轮中，执行步骤103、步骤105以及步骤107的处理。Usually, in the training of the neural network model, the more training rounds, the better the performance of the neural network model. At the beginning of training, the performance of the neural network model generally cannot meet the requirements. In the 8,000-round training, the performance of the neural network model generally cannot meet the requirements, and the performance of the neural network model in the next 2,000 rounds of training is getting better and better. And in each round after that, after executing
接下来对于步骤111：在所有轮训练结束后，针对神经网络模型的每一个参数，根据记录的该参数的至少一个当前参数值，得到该参数对应的最终参数值。Next to step 111 : after all rounds of training are finished, for each parameter of the neural network model, according to at least one current parameter value recorded for the parameter, the final parameter value corresponding to the parameter is obtained.
本步骤111的实现方式包括：The implementation of this
方式A、选取最大值。Method A. Select the maximum value.
在该方式A中，步骤111的实现过程包括如下步骤1111：针对神经网络模型的每一个参数，从记录的该参数的至少一个当前参数值中，选取最大的参数值，将该最大的参数值作为该参数对应的最终参数值。In this mode A, the implementation process of
方式B、选取平均值。Method B. Select the average value.
在该方式B中，步骤111的实现过程包括如下步骤：In this mode B, the implementation process of
步骤11121：针对神经网络模型的每一个参数，确定所记录的该参数的每一个当前参数值对应的权重值；Step 11121: For each parameter of the neural network model, determine the recorded weight value corresponding to each current parameter value of the parameter;
步骤11123：将该参数的各个当前参数值进行加权平均，以得到最终参数值。Step 11123: Perform weighted average of each current parameter value of the parameter to obtain the final parameter value.
上述步骤11121的实现方式可以包括：The implementation of the above step 11121 may include:
方式11、将每一个当前参数值对应的权重值均设置为
方式12、根据得到每一个当前参数值的轮数，设置每一个当前参数值对应的权重值，其中，得到一个当前参数值的轮数在前时，该当前参数值对应的权重值较小，得到一个当前参数值的轮数在后时，该当前参数值对应的权重值较大。Mode 12. According to the number of rounds for obtaining each current parameter value, set the weight value corresponding to each current parameter value. Wherein, when the number of rounds for obtaining a current parameter value is earlier, the weight value corresponding to the current parameter value is smaller. When the number of rounds to obtain a current parameter value is later, the weight value corresponding to the current parameter value is larger.
以上述方式11为例，说明步骤111的实现过程。比如，针对神经网络模型的参数M，总共记录了100次/个该参数M的当前参数值。那么，每一个当前参数值对应的权重值均为1/100，将100个当前参数值分别与1/100相乘后，乘积相加，则得到了参数M的最终参数值。Taking the foregoing manner 11 as an example, the implementation process of
以上述方式12为例，说明步骤111的实现过程。比如，针对神经网络模型的参数M，总共记录了100次/个该参数M的当前参数值。那么，因为训练的轮数越靠后，其性能越优，因此，靠后的训练轮中得到的当前参数值对应的权重值更大。比如，对于参数M，在第85轮中得到的当前参数值对应的权重值大于在第80轮中得到的当前参数值对应的权重值，在第91轮中得到的当前参数值对应的权重值大于在第85轮中得到的当前参数值对应的权重值。Taking the foregoing manner 12 as an example, the implementation process of
步骤113：将神经网络模型中每一个参数的参数值设置为该参数对应的最终参数值。Step 113: Set the parameter value of each parameter in the neural network model as the final parameter value corresponding to the parameter.
执行完本步骤113，则实现了将神经网络模型中每一个参数的参数值赋予了能够体现训练过程中各个阶段的模型特点的最终参数值。After executing this
需要说明的是，在本说明书实施例中，上述图1中涉及的神经网络模型中的各个参数可以是该神经网络模型中所有的参数，也可以是指定的部分重要参数。It should be noted that, in the embodiment of this specification, each parameter in the neural network model mentioned above in FIG. 1 may be all the parameters in the neural network model, or may be some specified important parameters.
在利用本说明书实施例的方法训练出神经网络模型之后，则可以利用该模型进行业务预测。参见图2，在本说明书实施例中，提出了一种业务预测方法，包括如下步骤：After the neural network model is trained by using the method in the embodiment of this specification, the model can be used for service prediction. Referring to Fig. 2, in the embodiment of this specification, a business prediction method is proposed, including the following steps:
步骤201：获取待预测的业务数据；Step 201: Obtain business data to be predicted;
步骤203：将待预测的业务数据输入神经网络模型中；Step 203: Input the business data to be predicted into the neural network model;
步骤205：得到该神经网络模型输出的对该待预测业务数据的识别结果。Step 205: Obtain the recognition result of the service data to be predicted output by the neural network model.
本说明书实施例还提出了一种神经网络模型的训练装置，参见图3，该装置包括：The embodiment of this specification also proposes a training device for a neural network model, as shown in Figure 3, the device includes:
样本获取模块301，配置为根据历史业务数据，获取训练样本数据；The
训练执行模块302，配置为在每一轮训练中均执行：将训练样本数据输入所述神经网络模型中，以对所述神经网络模型中每一个参数的参数值进行调整；以及检测本轮训练是否满足参数获取条件，如果是，则记录本轮训练得到的神经网络模型中每一个参数的当前参数值；The
参数值确定模块303，配置为在各轮训练结束后，针对神经网络模型的每一个参数，根据记录的该参数的至少一个当前参数值，得到该参数对应的最终参数值；The parameter
模型生成模块304，配置为将神经网络模型中每一个参数的参数值设置为该参数对应的最终参数值。The
在图3所示的本说明书装置的实施例中，训练执行模块302被配置为执行：根据预先设置的采样周期，检测本轮是否达到了该采样周期，如果是，则确定本轮训练满足参数获取条件；其中，N为不小于1的正整数，所述采样周期表征每隔N轮获取一次参数值。In the embodiment of the device of this specification shown in FIG. 3 , the
在图3所示的本说明书装置的实施例中，训练执行模块303被配置为执行：判断本轮得到的每一个参数的当前参数值是否优于上一次记录的每一个参数的当前参数值，如果是，则确定本轮训练满足参数获取条件。In the embodiment of the device of this specification shown in FIG. 3 , the
在图3所示的本说明书装置的实施例中，训练执行模块302被配置为执行：在预先设置的第L轮及其之后的每一轮中，均执行所述检测本轮训练是否满足参数获取条件的步骤，否则，不执行该步骤；其中，L为大于1的正整数。In the embodiment of the device of this description shown in FIG. 3 , the
在图3所示的本说明书装置的实施例中，参数值确定模块303被配置为执行：从记录的该参数的至少一个当前参数值中，选取最大的参数值，将该最大的参数值作为该参数对应的最终参数值。In the embodiment of the device of this specification shown in FIG. 3 , the parameter
在图3所示的本说明书装置的实施例中，参数值确定模块303被配置为执行：确定所记录的该参数的每一个当前参数值对应的权重值；In the embodiment of the device of this specification shown in FIG. 3 , the parameter
将该参数的各个当前参数值进行加权平均，以得到所述最终参数值。Each current parameter value of the parameter is weighted and averaged to obtain the final parameter value.
在图3所示的本说明书装置的实施例中，参数值确定模块303被配置为执行：将每一个当前参数值对应的权重值均设置为
在图3所示的本说明书装置的实施例中，参数值确定模块303被配置为执行：In the embodiment of the device of the present specification shown in FIG. 3, the parameter
根据得到每一个当前参数值的轮数，设置每一个当前参数只对应的权重值，其中，得到一个当前参数值的轮数在前时，该当前参数值对应的权重值较小，得到一个当前参数值的轮数在后时，该当前参数值对应的权重值较大。According to the number of rounds to obtain each current parameter value, set the weight value corresponding to each current parameter value. Among them, when the number of rounds to obtain a current parameter value is first, the weight value corresponding to the current parameter value is smaller, and a current parameter value is obtained. When the number of rounds of the parameter value is later, the weight value corresponding to the current parameter value is larger.
本说明书一个实施例提出了一种业务预测装置，参见图4，该装置包括：An embodiment of this specification proposes a service prediction device, see Figure 4, the device includes:
业务数据获取模块401，配置为获取待预测的业务数据；A business
输入模块402，配置为将待预测的业务数据输入神经网络模型中；所述神经网络模型是利用本说明书实施例中的模型训练装置训练出的；The
输出模块403，配置为得到该神经网络模型输出的对该待预测业务数据的识别结果。The
需要说明的是，上述各装置通常实现于服务器端，可以分别设置于独立的服务器，也可以其中部分或全部装置的组合设置于同一服务器。该服务器可以是单个的服务器，也可以是由多个服务器组成的服务器集群，服务器可以是云服务器，又称为云计算服务器或云主机，是云计算服务体系中的一项主机产品，以解决传统物理主机与虚拟专用服务器(VPs，Ⅵirtual Private Server)服务中存在的管理难度大，服务扩展性弱的缺陷。上述各装置还可以实现于具有较强计算能力的计算机终端。It should be noted that the above-mentioned devices are generally implemented on the server side, and may be respectively set on independent servers, or a combination of some or all of the devices may be set on the same server. The server can be a single server or a server cluster composed of multiple servers. The server can be a cloud server, also known as a cloud computing server or a cloud host. It is a host product in the cloud computing service system to solve Traditional physical host and virtual private server (VPs, VIirtual Private Server) services have the defects of high management difficulty and weak service scalability. The above-mentioned devices can also be implemented in computer terminals with relatively strong computing capabilities.
本说明书一个实施例提供了一种计算机可读存储介质，其上存储有计算机程序，当所述计算机程序在计算机中执行时，令计算机执行说明书中任一个实施例中的方法。An embodiment of the present specification provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed in a computer, the computer is instructed to execute the method in any one of the embodiments in the specification.
本说明书一个实施例提供了一种计算设备，包括存储器和处理器，所述存储器中存储有可执行代码，所述处理器执行所述可执行代码时，实现执行说明书中任一个实施例中的方法。An embodiment of this specification provides a computing device, including a memory and a processor, wherein executable code is stored in the memory, and when the processor executes the executable code, the implementation of any one of the embodiments in the specification is implemented. method.
可以理解的是，本说明书实施例示意的结构并不构成对本说明书实施例的装置的具体限定。在说明书的另一些实施例中，上述装置可以包括比图示更多或者更少的部件，或者组合某些部件，或者拆分某些部件，或者不同的部件布置。图示的部件可以以硬件、软件或者软件和硬件的组合来实现。It can be understood that, the structure shown in the embodiment of the present specification does not constitute a specific limitation on the device of the embodiment of the present specification. In other embodiments of the specification, the above-mentioned apparatus may include more or less components than those shown in the illustrations, or combine certain components, or separate certain components, or arrange different components. The illustrated components may be realized in hardware, software, or a combination of software and hardware.
本说明书中的各个实施例均采用递进的方式描述，各个实施例之间相同相似的部分互相参见即可，每个实施例重点说明的都是与其他实施例的不同之处。尤其，对于装置实施例而言，由于其基本相似于方法实施例，所以描述的比较简单，相关之处参见方法实施例的部分说明即可。Each embodiment in this specification is described in a progressive manner, the same and similar parts of each embodiment can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, as for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and for relevant parts, please refer to part of the description of the method embodiment.
本领域技术人员应该可以意识到，在上述一个或多个示例中，本发明所描述的功能可以用硬件、软件、挂件或它们的任意组合来实现。当使用软件实现时，可以将这些功能存储在计算机可读介质中或者作为计算机可读介质上的一个或多个指令或代码进行传输。Those skilled in the art should be aware that, in the above one or more examples, the functions described in the present invention may be implemented by hardware, software, pendants or any combination thereof. When implemented in software, the functions may be stored on or transmitted over as one or more instructions or code on a computer-readable medium.
以上所述的具体实施方式，对本发明的目的、技术方案和有益效果进行了进一步详细说明，所应理解的是，以上所述仅为本发明的具体实施方式而已，并不用于限定本发明的保护范围，凡在本发明的技术方案的基础之上，所做的任何修改、等同替换、改进等，均应包括在本发明的保护范围之内。The specific embodiments described above have further described the purpose, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above descriptions are only specific embodiments of the present invention and are not intended to limit the scope of the present invention. Protection scope, any modification, equivalent replacement, improvement, etc. made on the basis of the technical solution of the present invention shall be included in the protection scope of the present invention.
Claims (10)
Translated fromChinese
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211272252.0ACN115526266B (en) | 2022-10-18 | 2022-10-18 | Model Training Method and Device, Service Prediction Method and Device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211272252.0ACN115526266B (en) | 2022-10-18 | 2022-10-18 | Model Training Method and Device, Service Prediction Method and Device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115526266Atrue CN115526266A (en) | 2022-12-27 |
| CN115526266B CN115526266B (en) | 2023-08-29 |
Family
ID=84703949
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211272252.0AExpired - Fee RelatedCN115526266B (en) | 2022-10-18 | 2022-10-18 | Model Training Method and Device, Service Prediction Method and Device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115526266B (en) |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108510083A (en)* | 2018-03-29 | 2018-09-07 | 国信优易数据有限公司 | A kind of neural network model compression method and device |
| CN109118013A (en)* | 2018-08-29 | 2019-01-01 | 黑龙江工业学院 | A kind of management data prediction technique, readable storage medium storing program for executing and forecasting system neural network based |
| CN111291605A (en)* | 2018-12-07 | 2020-06-16 | 财团法人交大思源基金会 | People flow analysis system and people flow analysis method |
| US20200210840A1 (en)* | 2018-12-31 | 2020-07-02 | Microsoft Technology Licensing, Llc | Adjusting precision and topology parameters for neural network training based on a performance metric |
| CN111539479A (en)* | 2020-04-27 | 2020-08-14 | 北京百度网讯科技有限公司 | Method and apparatus for generating sample data |
| CN111582450A (en)* | 2020-05-08 | 2020-08-25 | 广东电网有限责任公司 | Neural network model training method based on parameter evaluation and related device |
| US20210241097A1 (en)* | 2019-11-07 | 2021-08-05 | Canon Kabushiki Kaisha | Method and Apparatus for training an object recognition model |
| CN113762061A (en)* | 2021-05-26 | 2021-12-07 | 腾讯科技（深圳）有限公司 | Quantitative perception training method and device for neural network and electronic equipment |
| CN113836804A (en)* | 2021-09-15 | 2021-12-24 | 向前 | Animal identification model establishing method based on convolutional neural network and application system thereof |
| CN113888475A (en)* | 2021-09-10 | 2022-01-04 | 上海商汤智能科技有限公司 | Image detection method, training method of related model, related device and equipment |
- 2022
- 2022-10-18CNCN202211272252.0Apatent/CN115526266B/ennot_activeExpired - Fee Related
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108510083A (en)* | 2018-03-29 | 2018-09-07 | 国信优易数据有限公司 | A kind of neural network model compression method and device |
| CN109118013A (en)* | 2018-08-29 | 2019-01-01 | 黑龙江工业学院 | A kind of management data prediction technique, readable storage medium storing program for executing and forecasting system neural network based |
| CN111291605A (en)* | 2018-12-07 | 2020-06-16 | 财团法人交大思源基金会 | People flow analysis system and people flow analysis method |
| US20200210840A1 (en)* | 2018-12-31 | 2020-07-02 | Microsoft Technology Licensing, Llc | Adjusting precision and topology parameters for neural network training based on a performance metric |
| US20210241097A1 (en)* | 2019-11-07 | 2021-08-05 | Canon Kabushiki Kaisha | Method and Apparatus for training an object recognition model |
| CN111539479A (en)* | 2020-04-27 | 2020-08-14 | 北京百度网讯科技有限公司 | Method and apparatus for generating sample data |
| CN111582450A (en)* | 2020-05-08 | 2020-08-25 | 广东电网有限责任公司 | Neural network model training method based on parameter evaluation and related device |
| CN113762061A (en)* | 2021-05-26 | 2021-12-07 | 腾讯科技（深圳）有限公司 | Quantitative perception training method and device for neural network and electronic equipment |
| CN113888475A (en)* | 2021-09-10 | 2022-01-04 | 上海商汤智能科技有限公司 | Image detection method, training method of related model, related device and equipment |
| CN113836804A (en)* | 2021-09-15 | 2021-12-24 | 向前 | Animal identification model establishing method based on convolutional neural network and application system thereof |
Non-Patent Citations (1)
| Title |
|---|
| 刘万军等: "基于双重优化的卷积神经网络图像识别算法", 《模式识别与人工智能》* |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115526266B (en) | 2023-08-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP6749468B2 (en) | Modeling method and apparatus for evaluation model | |
| CN111681091B (en) | Financial risk prediction method and device based on time domain information and storage medium | |
| WO2021027052A1 (en) | Interlayer parsing-based input instance verfication method for neural network model | |
| KR102330423B1 (en) | Online default forecasting system using image recognition deep learning algorithm | |
| CN114580239B (en) | A bridge damage identification method considering uncertainty | |
| CN109656818A (en) | A software-intensive system fault prediction method | |
| CN116244647A (en) | Unmanned aerial vehicle cluster running state estimation method | |
| CN113763928B (en) | Audio category prediction method, device, storage medium and electronic device | |
| CN115935265B (en) | Method for training risk identification model, risk identification method and corresponding device | |
| Asghari et al. | A novel approach for structural damage detection using multi-headed stacked deep ensemble learning | |
| CN120182582A (en) | Small sample HRRP radar target recognition method, device, equipment and medium | |
| CN114742644A (en) | Method and device for training multi-scene wind control system and predicting business object risk | |
| CN115526266B (en) | Model Training Method and Device, Service Prediction Method and Device | |
| CN115130003A (en) | Model processing method, device, equipment and storage medium | |
| JP7641978B2 (en) | Method and system for processing data with different time characteristics to generate predictions for management arrangements using a random forest classifier | |
| CN118337481A (en) | Knowledge distillation-based lightweight Internet of vehicles intrusion detection method and system | |
| CN117615359A (en) | Bluetooth data transmission method and system based on multiple rule engines | |
| CN114863178B (en) | Image data input detection method and system for neural network vision system | |
| CN117009218A (en) | Multi-interface system testing method, device, system and storage medium | |
| KR102244705B1 (en) | Method for controlling worker inflow into project by reversal adjustment of work unit price between crowdsourcing based similar projects for training data generation | |
| CN114443970A (en) | Artificial intelligence and big data based digital content pushing method and AI system | |
| CN118941376B (en) | Financing guarantee risk dynamic monitoring and early warning method and system based on data fusion | |
| CN118114920B (en) | Training system and method for scheduling model in fire scene | |
| CN114118958B (en) | Automated review method and device | |
| CN113327162B (en) | Credit wind control rule making system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | Granted publication date:20230829 |