CN112906896A

CN112906896A - Information processing method and device and computing equipment

Info

Publication number: CN112906896A
Application number: CN202110192000.6A
Authority: CN
Inventors: 章祯梁; 印卧涛
Original assignee: Alibaba Group Holding Ltd
Current assignee: Alibaba Group Holding Ltd
Priority date: 2021-02-19
Filing date: 2021-02-19
Publication date: 2021-06-04
Also published as: WO2022174755A1

Abstract

The embodiment of the application provides an information processing method, an information processing device and computing equipment, wherein the method comprises the following steps: responding to a parameter optimization request of a parameter to be optimized, and determining local feature information of the parameter to be optimized when the global feature information of the parameter to be optimized meets a preset local switching rule; acquiring and obtaining a first candidate solution of the parameter to be optimized according to the local feature information; and if the first candidate solution meets the parameter use condition, determining the first candidate solution as the target solution of the parameter to be optimized. The embodiment of the application improves the parameter optimization effectiveness.

Description

Information processing method and device and computing equipment

Technical Field

The present application relates to the technical field of electronic devices, and in particular, to an information processing method and apparatus, and a computing device.

Background

The black box algorithm is a function or a computing system which can only observe input and output, and has a complex internal computing structure or a complex computing process and is difficult to analyze. In the black box optimization process, firstly, a candidate solution of a parameter to be optimized is generated by using a black box algorithm, and then, the candidate solution is subjected to parameter test through an objective function so as to judge whether the candidate solution is a target solution which accords with a preset target. The black box optimization algorithm means that the objective function is a black box algorithm, and the calculation structure or function is unknown.

In the prior art, in the black box optimization process, in order to obtain a candidate solution, a bayesian model approximate to an objective Function may be used to represent the objective Function, so that on the basis of a few initial candidate solutions of a parameter to be optimized, the bayesian model is constructed by using the few candidate solutions, and the bayesian model is used to perform sampling again on the basis of an Acquisition Function (AC Function, Acquisition Function) to obtain a new better candidate solution, and then the objective Function is used to evaluate the use effect of the newly obtained candidate solution, so as to continuously obtain the candidate solution and the evaluation information of the candidate solution, and select the candidate solution with the highest evaluation effect as the objective solution.

However, when the candidate solution of the parameter to be optimized is generated by the black box algorithm, the sampling process based on the bayesian model is simple, and convergence cannot be quickly achieved, so that the parameter optimization efficiency is low.

Disclosure of Invention

In view of this, embodiments of the present application provide an information processing method and apparatus, and a computing device, so as to solve the technical problem in the prior art that the efficiency of parameter optimization is low due to the use of bayesian optimization.

In a first aspect, an embodiment of the present application provides an information processing method, including:

responding to a parameter optimization request of a parameter to be optimized, and determining local feature information of the parameter to be optimized when the global feature information of the parameter to be optimized meets a preset local switching rule;

acquiring and obtaining a first candidate solution of the parameter to be optimized according to the local feature information;

and if the first candidate solution meets the parameter use condition, determining the first candidate solution as the target solution of the parameter to be optimized.

In a second aspect, an embodiment of the present application provides an information processing method, including:

detecting a resource management request triggered by a resource management party aiming at a target resource;

responding to the resource management request, and determining the local feature information of the parameter to be optimized when the global feature information of the parameter to be optimized meets a preset local switching rule;

if the first candidate solution meets the parameter use condition, determining that the first candidate solution is the target solution of the parameter to be optimized;

generating resource setting information of the resource parameter based on the target solution;

and feeding back the resource setting information to the resource manager so that the resource manager can set the target resource according to the resource setting information.

In a third aspect, an embodiment of the present application provides an information processing method, including:

detecting a system access request triggered by an access user aiming at a network transaction system;

responding to the system access request, and determining the local characteristic information of the parameter to be optimized when the global characteristic information of the parameter to be optimized meets a preset local switching rule;

generating target access information corresponding to the system access parameter based on the target solution;

and feeding back the target access information to the access user so that the access user can operate the network transaction system based on the target access information.

In a fourth aspect, an embodiment of the present application provides an information processing method, including:

responding to a request for calling an information processing interface, and determining a processing resource corresponding to the information processing interface;

executing the following steps by utilizing the processing resource corresponding to the information processing interface:

In a fifth aspect, an embodiment of the present application provides an information processing apparatus, including:

the request response module is used for responding to a parameter optimization request of a parameter to be optimized and determining local characteristic information of the parameter to be optimized when the global characteristic information of the parameter to be optimized meets a preset local switching rule;

the local sampling module is used for acquiring and obtaining a first candidate solution of the parameter to be optimized according to the local characteristic information;

and the target determination module is used for determining the first candidate solution as the target solution of the parameter to be optimized if the first candidate solution meets the parameter use condition.

In a sixth aspect, an embodiment of the present application provides a computing device, including: a storage component and a processing component; the storage component is used for storing one or more computer instructions; the one or more computer instructions are invoked by the processing component;

the processing component is to:

responding to a parameter optimization request of a parameter to be optimized, and determining local feature information of the parameter to be optimized when the global feature information of the parameter to be optimized meets a preset local switching rule; acquiring and obtaining a first candidate solution of the parameter to be optimized according to the local feature information; and if the first candidate solution meets the parameter use condition, determining the first candidate solution as the target solution of the parameter to be optimized.

According to the method and the device, the parameter optimization request of the parameter to be optimized is responded, and when the global feature information of the parameter to be optimized meets the preset local switching rule, the local feature information of the parameter to be optimized is determined. And acquiring a first candidate solution of the parameter to be optimized according to the acquired information. If the first candidate solution satisfies the parameter use condition, the first candidate solution may be determined to be a target solution of the parameter to be optimized. When the parameters to be optimized are sampled, the information of the global aspect and the local aspect is determined, the multi-directional analysis of the parameters to be optimized is realized, the rapid local convergence can be realized on the basis of the second candidate solution, and the parameter optimization efficiency is improved.

Drawings

In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative efforts.

Fig. 1 is a flowchart of an embodiment of an information processing method according to an embodiment of the present application;

fig. 2 is a flowchart of another embodiment of an information processing method according to an embodiment of the present application;

fig. 3 is a flowchart of another embodiment of an information processing method according to an embodiment of the present application;

fig. 4 is a flowchart of another embodiment of an information processing method according to an embodiment of the present application;

fig. 5 is a flowchart of another embodiment of an information processing method according to an embodiment of the present application;

fig. 6 is a flowchart of another embodiment of an information processing method according to an embodiment of the present application;

fig. 7 is a flowchart of another embodiment of an information processing method according to an embodiment of the present application;

fig. 8 is a diagram illustrating an application example of an information processing method according to an embodiment of the present application;

fig. 9 is a schematic structural diagram of an embodiment of an information processing apparatus according to an embodiment of the present application;

fig. 10 is a schematic structural diagram of an embodiment of a computing device according to an embodiment of the present application.

Detailed Description

In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are some embodiments of the present application, but not all embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.

The terminology used in the embodiments of the present application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. As used in the examples of this application and the appended claims, the singular forms "a", "an", and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise, and "a" and "an" typically include at least two, but do not exclude the presence of at least one.

It should be understood that the term "and/or" as used herein is merely one type of association that describes an associated object, meaning that three relationships may exist, e.g., a and/or B may mean: a exists alone, A and B exist simultaneously, and B exists alone. In addition, the character "/" herein generally indicates that the former and latter related objects are in an "or" relationship.

The words "if," "if," as used herein may be interpreted as "at … …" or "at … …" or "in response to a determination" or "in response to a recognition," depending on the context. Similarly, the phrases "if determined" or "if identified (a stated condition or event)" may be interpreted as "when determined" or "in response to a determination" or "when identified (a stated condition or event)" or "in response to an identification (a stated condition or event)", depending on the context.

It is also noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a good or system that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such good or system. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in a commodity or system that includes the element.

The technical scheme of the embodiment of the application can be applied to a parameter optimization scene, the global characteristic and the local characteristic of the objective function are considered in the parameter sampling process, multi-angle optimization of the parameters is achieved, more optimization information is provided, and the optimization efficiency is improved.

In the prior art, in order to obtain a candidate solution, a bayesian model approximate to an objective function is usually adopted to represent the objective function, so that on the basis of a few initial candidate solutions of a parameter to be optimized, the bayesian model is constructed by using the few initial candidate solutions, and the bayesian model is used to perform sampling again on the basis of an acquisition function to obtain a new better candidate solution, and then the objective function is used to evaluate the use effect of the newly obtained candidate solution to continuously obtain the candidate solution and evaluation information of the candidate solution, and a candidate solution with the highest evaluation effect is selected as the objective solution. However, when the black box algorithm generates a candidate solution of a parameter to be optimized, the sampling process based on the bayesian model is a global sampling algorithm, global exploration is performed only by using a proxy model of a target function and a collection function, the sampling process is simple, and convergence cannot be quickly achieved.

In the embodiment of the application, in response to a parameter optimization request of a parameter to be optimized, when global feature information of the parameter to be optimized meets a preset local switching rule, local feature information of the parameter to be optimized is determined. And acquiring a first candidate solution of the parameter to be optimized according to the local characteristic information. If the first candidate solution satisfies the parameter use condition, the first candidate solution may be determined to be a target solution of the parameter to be optimized. When the parameters to be optimized are sampled, the global characteristic information is utilized to carry out switching judgment so as to confirm whether local analysis is carried out, the information in the global and local aspects is actually adopted, the multi-directional analysis of the parameters to be optimized is realized, the rapid local convergence can be realized on the basis of the second candidate solution, and the parameter optimization efficiency is improved.

The embodiments of the present application will be described in detail below with reference to the accompanying drawings.

As shown in fig. 1, which is a flowchart of an embodiment of an information processing method provided in an embodiment of the present application, the method may include the following steps:

101: and responding to a parameter optimization request of the parameter to be optimized, and determining the local characteristic information of the parameter to be optimized when the global characteristic information of the parameter to be optimized meets a preset local switching rule.

The embodiment of the application can be applied to a computing device, and the computing device can include: the embodiments of the present application do not limit the specific types of the computing devices. In practical applications, if the technical solution of the embodiment of the present application is applied to a computing device having a GPU (Graphics Processing Unit), the GPU may execute the technical solution of the embodiment of the present application.

The target user may interact with the computing device for data or information using a user device, which may include, for example: the mobile phone, the tablet computer, the notebook, the virtual reality device, the augmented reality device, or the wearable device, etc., and the specific type of the user equipment is not limited too much in the embodiments of the present application.

The parameter optimization request may be initiated by the target user for the parameter to be optimized, and specifically may be generated when the user equipment detects a request operation triggered by the target user.

Alternatively, the parameter to be optimized may be a normal parameter, or a hyper-parameter. The common parameters may be parameters in the model during the calculation of the model. Hyper-parameterOne parameter that may be set for building the data model prior to the data model calculation or learning process is not the model parameter used in the training process. Typically a superparameter may comprise a plurality of sub-superparameters. The type of the hyper-parameter can include a plurality of types, for example, besides the ordinary hyper-parameter which is not associated with each sub-hyper-parameter, the proportion type hyper-parameter can also be one of the hyper-parameters. The number of network layers of a deep network in the machine learning model, the learning rate of the model and the like can belong to two sub-hyper-parameters in the common hyper-parameters. The proportion parameter may be a superparameter in which the sum of the parameter proportions occupied by the plurality of sub-superparameters is 1, and it is assumed that n sub-superparameters exist, and the parameter values corresponding to the n sub-superparameters are a_iWherein a is_i∈[0,1]，

The specific type of the parameter to be optimized in the embodiment of the present application is not limited too much.

The local switching rules may include: and the incidence relation between the comparison result of the comparison information designed for the global characteristic information and the switching to the local sampling. When the comparison result of the global feature information is the first result, the local sampling may be switched to, and when the comparison result of the global feature information is the second result, the local optimization may not be switched to, and the global sampling is continuously performed.

102: and acquiring a first candidate solution of the parameter to be optimized according to the local characteristic information.

The first candidate solution may be obtained using local feature information sampling on the basis of the second candidate solution. By locally sampling the parameters to be optimized, fast convergence can be achieved.

103: and if the first candidate solution meets the parameter use condition, determining the first candidate solution as a target solution of the parameter to be optimized.

When the first candidate solution satisfies the parameter use condition, the first candidate solution may be a target solution of the parameter to be optimized. Whether the first candidate solution satisfies the parameter usage condition may be determined based on the second target value of the first candidate solution at the objective function.

As shown in fig. 2, a flowchart of another embodiment of an information processing method provided in the embodiment of the present application may include the following steps:

201: and responding to a parameter optimization request of the parameter to be optimized, and determining the global characteristic information of the parameter to be optimized.

Some steps in the embodiment of the present application are the same as those in the embodiment shown in fig. 1, and are not described herein again.

202: and if the global characteristic information meets the local switching rule, obtaining a second candidate solution of the parameter to be optimized by using a preset global sampling algorithm.

The global feature information may be the distribution of sampling points of the parameter to be optimized in the whole sampling range. The sampling point with the highest overall use effect of the parameter to be optimized can be estimated through the global characteristic information so as to obtain a second candidate solution. After the second candidate solution in the global scope is obtained, the vicinity of the second candidate solution may be tested to obtain a sampling point with the highest local use effect, that is, the first candidate solution. And obtaining an accurate target solution through global and local feature sampling.

203: and evaluating the local acquisition characteristics of the parameter to be optimized based on the second candidate solution to obtain the local characteristic information of the parameter to be optimized.

204: and acquiring a first candidate solution of the parameter to be optimized according to the local characteristic information.

205: and if the first candidate solution meets the parameter use condition, determining the first candidate solution as a target solution of the parameter to be optimized.

In the embodiment of the application, in response to a parameter optimization request of a parameter to be optimized, global feature information of the parameter to be optimized can be determined, so that whether the global feature information meets a local switching rule or not is judged. If the global feature information meets the local switching rule, a second candidate solution of the parameter to be optimized can be obtained by using a global sampling algorithm, so that parameter sampling on the global feature is realized. And then evaluating the local acquisition characteristics of the parameter to be optimized based on the second candidate solution to obtain the local characteristic information of the parameter to be optimized. Therefore, according to the local feature information, a first candidate solution of the parameter to be optimized is acquired and obtained, and parameter sampling on the local feature is achieved. And further determining the first candidate solution as the target solution when the first candidate solution meets the parameter use condition. By adopting information in the global and local aspects, multi-directional analysis of the parameters to be optimized is realized, rapid local convergence can be realized on the basis of the second candidate solution, and the parameter optimization efficiency is improved.

As shown in fig. 3, which is a flowchart of an embodiment of an information processing method provided in the embodiment of the present application, the method may include the following steps:

301: and responding to a parameter optimization request of the parameter to be optimized, and determining the global characteristic information of the parameter to be optimized.

Optionally, the global feature information of the parameter to be optimized may include a probability distribution model of sampling points of the parameter to be optimized. The global feature information may include a probability distribution model.

302: and if the global characteristic information meets the local switching rule, acquiring a second candidate solution of the parameter to be optimized by utilizing a preset global sampling algorithm.

Some steps in the embodiment of the present application are the same as those in the embodiment shown in fig. 1, and are not described herein again for the sake of brevity of description.

The global sampling algorithm is to sample the parameter to be optimized in the whole search domain range of the objective function and evaluate the use effect of the candidate solution obtained by sampling so as to obtain a second candidate solution with the highest use effect in the global range. Any effect information may include effect index data, and in general, the effect information of any solution candidate may include a target value obtained by inputting the solution candidate to an objective function and calculating, and the target value may be the effect index data. In one possible design, the effect index data is positively correlated with the use effect, the higher the effect index data is, the better the use effect is, and the lower the effect index data is, the worse the use effect is.

Optionally, the global sampling algorithm may include: bayesian optimization algorithms, web search algorithms, random search algorithms, etc. In the global sampling algorithm, the parameter value of the parameter to be optimized is sampled in the global range, so as to obtain the parameter value with the highest use effect in the global range, and obtain the target solution of the parameter to be optimized.

Taking a bayesian algorithm as an example, the obtaining of the second candidate solution of the parameter to be optimized by using the preset global sampling algorithm may specifically be to use a bayesian model proxy objective function that is similar to the actual sampling point of the objective function at the parameter to be optimized, that is, the distribution of the candidate solution and the actual target value corresponding to the candidate solution, so as to analyze the global distribution of the candidate solution by using the bayesian model, and obtain the second candidate solution with the highest use effect in the global range.

After the second candidate solution is obtained, the sampling region near the second candidate solution may be analyzed based on the second candidate solution, so as to obtain the local feature information of the parameter to be optimized. When the global acquisition algorithm samples the parameters to be optimized, the parameters to be optimized are sampled in the whole search range of the parameters to be optimized so as to obtain a target solution with the highest global use effect. The local acquisition characteristic can be an analysis result of a sampling region near a second candidate solution, the second candidate solution is a candidate solution with the highest global effect index data obtained by current sampling, and the local characteristic information of the parameter to be optimized is determined on the basis of the second candidate solution, so that the target solution can be searched in an accelerated manner, and the rapid convergence of the parameter is realized.

303: and evaluating the local acquisition characteristics of the parameter to be optimized based on the second candidate solution to obtain the local characteristic information of the parameter to be optimized.

304: and acquiring a first candidate solution of the parameter to be optimized according to the local characteristic information.

305: judging whether the first candidate solution meets the parameter use condition; if yes, go to step 306; if not, go to step 307.

306: and determining the first candidate solution as a target solution of the parameter to be optimized.

307: and updating the first candidate solution into a second candidate solution, and returning to the step 303 to continue execution.

Optionally, the first candidate solution satisfying the parameter usage condition may include: the target value difference between the second target value and the historical target value corresponding to the historical candidate solution is smaller than a preset difference threshold, and it can be determined that the first candidate solution satisfies the parameter use condition. The historical candidate solution may be one or more, and specifically may be a first candidate solution obtained by sampling several times before the first candidate solution, and after the parameter usage condition is determined for any first candidate solution, the first candidate solution may be stored, and at this time, the first candidate solution becomes a historical candidate solution. The first candidate solution satisfying the parameter usage condition may further include: the second target value is greater than or equal to a preset target threshold, and at this time, it may be determined that the first candidate solution satisfies the parameter usage condition.

In the embodiment of the application, after the second candidate solution of the parameter to be optimized is acquired and obtained by using the preset global sampling algorithm in response to the parameter optimization request of the parameter to be optimized, the local acquisition feature of the parameter to be optimized can be evaluated based on the second candidate solution, so as to obtain the local feature information of the parameter to be optimized. According to the local feature information, a first candidate solution for obtaining the parameter to be optimized can be acquired. When the first candidate solution satisfies the parameter usage condition, the first candidate solution may be determined to be the target solution. And when the first candidate solution does not meet the parameter use condition, updating the first candidate solution into a second candidate solution, and returning to evaluating the local acquisition characteristics of the parameter to be optimized based on the second candidate solution to obtain the local characteristic information of the parameter to be optimized for continuous execution. And through multiple local analysis, the local area of the candidate solution is accurately sampled, the rapid convergence of the parameters is realized, and the parameter optimization efficiency is improved.

When the global sampling algorithm is adopted, each candidate solution obtained by sampling may be input to the objective function to obtain target values corresponding to the plurality of candidate solutions, and the target value corresponding to each of the plurality of candidate solutions is used to select the target solution with the highest target value. The sampling method needs to calculate the objective function for each of the candidate solutions, and the calculation process of the objective function is complex, which results in low efficiency of parameter optimization. The global sampling algorithm may be a bayesian optimization algorithm. In the Bayesian optimization algorithm, a Gaussian model can be constructed by adopting a small amount of initial parameters, and then the Gaussian model is used for parameter sampling to obtain a second candidate solution.

The objective function may be a black box function preset for performing a parameter test on the candidate solution to obtain an evaluation index of the candidate solution. The mathematical expression form position of the objective function has higher complexity. Inputting the first candidate solution to the objective function may calculate an evaluation index to obtain the first candidate solution. The higher the evaluation index is, the higher the use effect of the first solution candidate is, and the lower the evaluation index is, the lower the use effect of the first solution candidate is. The effect of the sampling can be processed in time by evaluating the using effect of the first candidate solution, so as to obtain an evaluation index of the first candidate solution.

In global optimization, the global feature information may include a probability distribution model, whether the current model needs to be optimized again may be determined according to the model distribution condition of the probability distribution model, and when the probability distribution model satisfies a certain local switching rule, the probability distribution model may be switched to local sampling. The probability distribution model may be obtained by performing a distribution fit based on a plurality of initial candidate solutions.

As an embodiment, in response to a parameter optimization request for a parameter to be optimized, determining global feature information of the parameter to be optimized may include:

responding to a parameter optimization request of a parameter to be optimized, and determining initial effect information corresponding to a plurality of initial candidate solutions and a plurality of initial candidate solutions of the parameter to be optimized respectively;

fitting a distribution model of parameters to be optimized according to initial effect information respectively corresponding to the initial candidate solutions and the initial candidate solutions to obtain a probability distribution model;

if the global feature information satisfies the local switching rule, acquiring a second candidate solution of the parameter to be optimized by using a preset global sampling algorithm may include:

and if the probability distribution model meets the local switching rule, performing parameter sampling on the parameter to be optimized by using a preset acquisition function based on the probability distribution model to obtain a second candidate solution.

Optionally, taking the probability distribution model as a gaussian distribution model as an example, fitting the distribution model of the parameter to be optimized according to the initial effect information corresponding to the multiple initial candidate solutions and the multiple initial candidate solutions, and obtaining the probability distribution model may include: and performing Gaussian fitting on the distribution model of the parameters to be optimized according to the initial effect information corresponding to the initial candidate solutions and the initial effect information corresponding to the initial candidate solutions to obtain a Gaussian distribution model. At this time, based on the probability distribution model, performing parameter sampling on the parameter to be optimized by using a preset acquisition function, and obtaining the second candidate solution may include: and based on the Gaussian distribution model, performing parameter sampling on the parameter to be optimized by using a preset acquisition function to obtain a second candidate solution.

The initial effect information corresponding to any initial candidate solution may include effect indicator data corresponding to the initial candidate solution. In general, the effect information of any one candidate solution may include inputting the candidate solution to an objective function, and calculating to obtain a corresponding target value, where the target value may be effect index data.

In practical applications, the probability distribution models of the initial candidate solutions may include multiple types, and besides the gaussian model, the probability distribution models may also be a polynomial model, a bernoulli model, and the like.

In a bayesian optimization process, the AC function may include a variety of factors, such as: the specific type of the sampling function is not limited to a few examples, such as, for example, profile of improvement (POI), Expected Improvement (EI), Entropy search (Entropy search), Upper/Lower confidence bound (Upper/Lower confidence), and so on. When the parameters to be optimized are sampled, a batch sampling mode can be adopted, and a batch of candidate solutions are collected at the same time so as to select a second candidate solution, and the sampling efficiency is improved.

In the embodiment of the application, in response to a parameter optimization request of a parameter to be optimized, initial effect information corresponding to a plurality of initial candidate solutions and a plurality of initial candidate solutions of the parameter to be optimized respectively can be determined. And fitting the distribution model of the parameters to be optimized by using a plurality of initial candidate solutions with small quantity through sampling the parameters to be optimized by a small quantity to obtain a probability distribution model. The parameters to be optimized can be sampled through the probability distribution model to obtain a second candidate solution. The probability model estimation with higher use efficiency in the global sampling process is used for prompting sampling, so that the sampling efficiency can be improved.

In one possible design, the method may further include:

and if the probability distribution model does not meet the local switching rule, determining first effect information of a second candidate solution, increasing the second candidate solution into an initial candidate solution, returning to a parameter optimization request responding to the parameter to be optimized, and determining continuous execution of initial effect information corresponding to the initial candidate solutions and the initial candidate solutions of the parameter to be optimized.

Optionally, when the statistical indicator of the probability distribution model does not satisfy the local switching rule, the probability distribution model may be considered to be not accurate enough, or the information is not comprehensive enough, and a second candidate solution newly acquired may be added to the initial candidate solution, so as to fit the probability distribution model with more candidate solutions, so as to improve the accuracy of the probability distribution model. The probability distribution model with higher accuracy is used for parameter sampling, unnecessary local convergence can be reduced, the combination of global analysis and local analysis is tighter, and the efficiency and accuracy of parameter optimization are improved.

As a possible implementation, whether the probability distribution model satisfies the local switching rule may be determined by:

determining a statistical index of the probability distribution model;

if the statistical index meets a preset index threshold, determining that the second candidate solution meets a local switching rule;

and if the statistical index does not meet the preset index threshold, determining that the second candidate solution does not meet the local switching rule.

The probability distribution model may be actually obtained by calculating posterior distribution probabilities or prior distribution probabilities of the initial effect information corresponding to the distribution of the initial candidate solutions and the initial candidate solutions, and may be composed of probability statistical indicators, where the statistical indicators may be used for at least one model data including the probability distribution model, for example, model type, expectation, variance, covariance, and the like. Taking the gaussian model as an example, the statistical indicators of the gaussian model may include: gaussian expectation, gaussian variance.

The statistical indexes of the probability distribution model are subjected to condition judgment to determine whether the current probability distribution model meets the local switching rule or not, so that accurate switching from global analysis to local analysis is realized, and the switching efficiency is improved. In this case, the local switching rule may specifically include: the method comprises the steps that a comparison result between a statistical index of a probability distribution model and a preset index threshold value and an incidence relation of switching to local optimization are obtained, when the statistical index of the probability distribution model meets the preset index threshold value, the comparison result is a first result, the probability distribution model meets a local switching rule, when the statistical index of the probability distribution model does not meet the preset index threshold value, the comparison result is a second result, and the probability distribution model does not meet the local switching rule.

Alternatively, the statistical indicators of the probability distribution model may include, for example: mean, variance, covariance, etc.

After the statistical index of the probability distribution model is obtained, whether the probability distribution model meets the local switching rule or not can be accurately judged by comparing the statistical index with a preset index threshold value.

The statistical indicator satisfying the preset indicator threshold value may include, for example, that a difference between the statistical indicator and the preset indicator threshold value is smaller than an error threshold value. For example, when the statistical indicator is a mean value, if an error between a mean value of the probability distribution model and a preset mean value threshold is less than 0.1, the probability distribution model may be considered to satisfy the local switching rule, and if an error between a mean value of the probability distribution model and a preset mean value is greater than 0.1, the probability distribution model may be considered not to satisfy the local switching rule.

As shown in fig. 4, a flowchart of another embodiment of an information processing method provided in the embodiment of the present application may include the following steps:

401: and responding to a parameter optimization request of the parameter to be optimized, and determining initial effect information corresponding to a plurality of initial candidate solutions and a plurality of initial candidate solutions of the parameter to be optimized respectively.

Some steps in the embodiments of the present application are the same as those in the embodiments described above, and are not described herein again.

402: and fitting the distribution model of the parameters to be optimized according to the initial effect information respectively corresponding to the initial candidate solutions and the initial effect information corresponding to the initial candidate solutions to obtain a probability distribution model.

403: and based on the probability distribution model, performing parameter sampling on the parameter to be optimized by using a preset acquisition function to obtain a second candidate solution.

404: judging whether the probability distribution model meets the local switching rule, if so, executing step 405; if not, determining the first effect information of the second candidate solution, adding the second candidate solution into the initial candidate solution, and returning to the step 301 to continue the execution.

405: and evaluating the local acquisition characteristics of the parameter to be optimized based on the second candidate solution to obtain the local characteristic information of the parameter to be optimized.

406: and acquiring a first candidate solution of the parameter to be optimized according to the local characteristic information.

407: judging whether the first candidate solution meets the parameter use condition; if so, go to step 408; if not, step 409 is performed.

408: and determining the first candidate solution as a target solution of the parameter to be optimized.

409: and updating the first candidate solution into a second candidate solution, returning to the step 405 and continuing to execute.

In the embodiment of the application, in response to a parameter optimization request of a parameter to be optimized, initial effect information corresponding to a plurality of initial candidate solutions and a plurality of initial candidate solutions of the parameter to be optimized respectively can be determined. And then, fitting the distribution model of the parameters to be optimized according to the initial effect information corresponding to the initial candidate solutions and the initial effect information corresponding to the initial candidate solutions to obtain a probability distribution model. Based on the probability distribution model, a preset acquisition function can be utilized to perform parameter sampling on the parameter to be optimized, so as to obtain a second candidate solution. The probability distribution model is a global analysis of the sampling points of the parameter to be optimized, and can identify the global characteristics of the parameter to be optimized. Judging whether the local switching rule is met or not by judging the statistical indexes of the probability distribution model, if so, executing subsequent local analysis, if not, determining the first effect information of the second candidate solution, adding the second candidate solution into the initial candidate solution, and returning to the step of determining the initial effect information corresponding to the initial candidate solutions and the plurality of initial candidate solutions of the parameter to be optimized to continue execution.

By judging the local switching rule of the statistical indexes of the probability distribution model, more candidate solutions can be fitted to the probability distribution model, so that the accuracy of the probability distribution model is improved. And evaluating the local acquisition characteristics of the parameter to be optimized based on the second candidate solution when the subsequent probability distribution model meets the local switching rule, so as to obtain the local characteristic information of the parameter to be optimized and obtain accurate local characteristic information.

And then acquiring a first candidate solution of the parameter to be optimized according to the local characteristic information. And when the first candidate solution meets the parameter use condition, determining the first candidate solution as a target solution, when the first candidate solution does not meet the parameter use condition, updating the first candidate solution into a second candidate solution, returning to evaluate the local acquisition characteristics of the parameter to be optimized based on the second candidate solution, obtaining the local characteristic information of the parameter to be optimized, continuing to execute, and realizing local rapid convergence. Through accurate grasp of the global probability distribution model, a global second candidate solution can be obtained, local iterative sampling is carried out on the parameter to be optimized on the basis of the second candidate solution, so that an accurate first candidate solution is obtained, rapid convergence is achieved, and a target solution is obtained.

As an embodiment, evaluating the local acquisition feature of the parameter to be optimized based on the second candidate solution, and obtaining the local feature information of the parameter to be optimized may include:

performing gradient estimation processing on the parameter to be optimized in the second candidate solution to obtain sampling displacement data;

according to the local feature information, acquiring a first candidate solution of the parameter to be optimized may include:

and performing parameter sampling on the parameter to be optimized on the basis of the second candidate solution according to the sampling displacement data to obtain a first candidate solution.

As one possible implementation, sampling the displacement data may include: the obtaining of the first candidate solution by performing parameter sampling on the parameter to be optimized on the basis of the second candidate solution according to the sampling displacement data may include: and performing parameter sampling according to the direction data and the step data on the basis of the second candidate solution to obtain a first candidate solution. Specifically, data sampling may be performed according to the direction data moving step data on the basis of the value of the second candidate solution to obtain the first candidate solution.

In one possible design, evaluating disturbance information of the parameter to be optimized based on the second candidate solution, and obtaining the sampling displacement data may include:

and carrying out disturbance analysis processing on the second candidate solution based on a gradient estimation algorithm to obtain sampling displacement data.

Optionally, performing perturbation analysis processing on the second candidate solution based on a gradient estimation algorithm, and obtaining the sample displacement data may include: determining a second candidate solution and a historical candidate solution; performing gradient estimation calculation on the historical candidate solution to obtain a gradient calculation result; and carrying out disturbance analysis processing on the gradient calculation result to obtain sampling displacement data. The gradient estimation algorithm may include, for example, finite difference method (finite difference), infinite perturbation analysis (infinite perturbation analysis), gradient sampling method (gradient sampling), simultaneous perturbation stochastic approximation (simultaneous perturbation approximation), and the like. Disturbance Analysis (Perturbation Analysis) is a data processing and analyzing method, and may be to analyze an optimized position and direction near the second candidate solution by using a mathematical direction to obtain sampling displacement data of a parameter to be optimized, and further perform parameter sampling by using the sampling displacement data to obtain a first candidate solution. The principle of the disturbance analysis process can be referred to the description in the prior art, and is not described herein. The information about the sensitivity of the system characteristics to the selection of the parameters to be optimized can be directly obtained through disturbance analysis processing, so that the optimization efficiency can be greatly improved.

Alternatively, the historical candidate solution may be selected from a plurality of initial candidate solutions. For example, any one of a plurality of initial candidate solutions may be selected as the historical candidate solution. And selecting the initial candidate solution with the highest initial effect information as a historical candidate solution according to the initial effect information respectively corresponding to the plurality of initial candidate solutions.

As an embodiment, before acquiring and obtaining the second candidate solution of the parameter to be optimized by using the preset global sampling algorithm in response to the parameter optimization request of the parameter to be optimized, the method may further include:

receiving a parameter optimization request sent by a target user;

if the first candidate solution satisfies the parameter use condition, after determining that the first candidate solution is the target solution of the parameter to be optimized, the method may further include:

and outputting the target solution for the target user.

Alternatively, the parameter optimization request may be initiated by the target user. The information processing method provided by the application can be displayed for the target user, so that the user can use the information processing service provided by the embodiment of the application to acquire the target solution. The parameter optimization request may be sent by the user device of the target user to a computing device configured with the information processing method shown in fig. 1.

The target solution may be output to a target user. If the user side of the target user is the same device as the computing device, outputting the target solution for the target user may include presenting the target solution for the target user in a display interface. If the user end of the target user and the computing device are different devices, outputting the target solution for the target user may include: and sending the target solution to a user side of the target user so that the user side can output the target solution for the target user.

In practical application, in order to provide more personalized parameter service, parameter adjustment service can be provided for target users. In some embodiments, the method may further comprise:

detecting a parameter adjustment request initiated by a target user aiming at a target solution;

responding to the parameter adjustment request, and acquiring parameter adjustment information provided by a target user;

and adjusting the target solution according to the parameter adjustment information to obtain an expected solution of the target user.

Optionally, before detecting the parameter adjustment request initiated by the target user for the target solution, the method may further include: and displaying the parameter adjusting control to the target user so that the target user knows that the target solution is in an adjustable state. Detecting a parameter adjustment request initiated by a target user for a target solution may include: and detecting the trigger operation executed by the target user aiming at the parameter adjusting control of the target solution, and generating a parameter adjusting request.

In many application fields, the parameter optimization problem may be directly involved, and in order to improve the parameter optimization efficiency, the technical solution of the embodiment of the present application may be used.

In the process of allocating the electric power resources and the water resources, the allocation result of the electric power resources or the water resources in each area may be used as a parameter to be optimized to initiate a parameter optimization request, and the parameter to be optimized may specifically be a resource amount corresponding to each area, for example, a load capacity of an area in an electric power scene.

As shown in fig. 5, a flowchart of another embodiment of an information processing method provided in an embodiment of the present application may include:

501: and detecting a resource management request triggered by the resource manager for the target resource.

502: and responding to the resource management request, and determining the local characteristic information of the parameter to be optimized when the global characteristic information of the parameter to be optimized meets a preset local switching rule.

503: and acquiring a first candidate solution of the parameter to be optimized according to the local characteristic information.

504: and if the first candidate solution meets the parameter use condition, determining the first candidate solution as a target solution of the parameter to be optimized.

505: and generating resource setting information of the resource parameters based on the target solution.

506: and feeding back the resource setting information to the resource manager so that the resource manager can set the target resource according to the resource setting information.

Optionally, the method may further include, if the first candidate solution does not satisfy the parameter usage condition, determining that the first candidate solution is a second candidate solution, returning to evaluate the local acquisition feature of the parameter to be optimized based on the second candidate solution, and obtaining the local feature information of the parameter to be optimized to continue execution.

The resource elements specifically represented by the parameters to be optimized may be determined according to the processing objective of the target resource. For example, when the processing target of the target resource may be the power load capacities set in different regions so as to minimize the total power consumption of the power grid, at this time, the power load capacities in different regions may be the sub-parameters of the parameter to be optimized, and the processing target may be the target function corresponding to the total power consumption of the power grid. The target solution of the parameters to be optimized can be the power load capacity of each region under the condition that the obtained total power consumption of the power grid is minimum. Processing information of the target resource can be generated according to the target solution of the parameter to be optimized, that is, prompt information or setting instructions of the power load capacity of each region can be generated according to the target solution of the parameter to be optimized, and capacity setting can be performed according to the power load capacity of each region through the setting instructions. The prompt information can be displayed for the user, so that the user can set the capacity of each region according to the power load capacity of each region prompted in the prompt information.

In the field of e-commerce, parameter optimization problems are also involved. Taking a relatively common product recommendation as an example, due to different browsing characteristics of a user, such as consumption habits, attention fields, historical browsing behaviors, and the like, the recommended contents or products for the user are also different. In practical application, in order to improve the click rate of a user, the browsing characteristics of the user such as consumption habits, attention fields and the like can be parameterized to generate different browsing parameters, and the characteristics of a click target of the user can be accurately analyzed by setting a plurality of browsing parameters, so that a target product with higher attention degree with the user can be found. And taking the plurality of browsing parameters as parameters to be optimized, continuously sampling the parameters to be optimized, and calculating the use effect of the parameters so as to evaluate the target solution with the highest index. The method and the device can be applied to the technical scheme of the embodiment of the application so as to improve the parameter optimization efficiency of the parameters to be optimized.

As shown in fig. 6, a flowchart of another embodiment of an information processing method provided in an embodiment of the present application may include:

601: a system access request triggered by an accessing user for a network transaction system is detected.

602: and responding to the system access request, and determining the local characteristic information of the parameter to be optimized when the global characteristic information of the parameter to be optimized meets a preset local switching rule.

603: and acquiring a first candidate solution of the parameter to be optimized according to the local characteristic information.

604: and if the first candidate solution meets the parameter use condition, determining the first candidate solution as a target solution of the parameter to be optimized.

605: and generating target access information corresponding to the system access parameters based on the target solution.

606: and feeding back target access information to the access user so that the access user can operate the network transaction system based on the target access information.

In the network transaction system, the parameter to be optimized may be a browsing parameter of the user, and the target access information fed back to the user may be generated by obtaining a target solution corresponding to the browsing parameter. For example, the target access information may be product information recommended to the target user.

Optionally, the method may further include: and if the first candidate solution does not meet the parameter use condition, determining the first candidate solution as a second candidate solution, returning to evaluate the local acquisition characteristics of the parameter to be optimized based on the second candidate solution, and obtaining the local characteristic information of the parameter to be optimized to continue execution.

In the network transaction system, corresponding browsing parameters are obtained by setting the corresponding browsing parameters for the browsing characteristics of a target user, so that parameter tests are carried out on the browsing parameters to obtain a target solution with the highest evaluation index. Of course, in practical applications, the plurality of browsing parameters may constitute a plurality of sub-parameters of the parameter to be optimized, and a target solution may be obtained when the respective sampling of the parameter values corresponding to the plurality of sub-parameters is finished. After the target solution is determined, the target access information of the target user can be generated through the target solution, so that the target product matched with the target access information can be searched and output for the target user.

In some embodiments, the browsing parameters may be proportions of different browsing characteristics, and the proportions of the different browsing characteristics in the product searching process may be determined according to target solutions corresponding to the browsing parameters, so that the plurality of browsing characteristics are weighted and summed according to values of the browsing characteristics in the target solutions to obtain recommended characteristics, and the recommended characteristics may be used for searching recommended products for users, displaying the searched target products for the target users, and feeding back the searched target products as target access information to the target users.

In still other embodiments, the browsing parameter may be a ratio of different types of products, that is, products may be recommended to the user from a plurality of types of products, but the ratio of each type of product is different. Taking cosmetics and clothing products as main recommended types as examples, parameter optimization is carried out on the recommended proportions of the cosmetics and the clothing products, the finally obtained target solution is 3:7, the cosmetics and the clothing products are 3, at the moment, according to values of browsing parameters in the target solution, 3 parts of cosmetics and 7 parts of clothing products are searched from a product database, and the searched cosmetics and clothing products are fed back to a target user as target access information.

It should be noted that the specific application schemes of the embodiments shown in fig. 5 and fig. 6 are only exemplary, and should not constitute application limitations of the embodiments of the present application, and the embodiments of the present application may be applied to parameter optimization scenarios of various data models and calculation models.

As shown in fig. 7, a flowchart of another embodiment of an information processing method provided in an embodiment of the present application may include:

701: and responding to the request for calling the information processing interface, and determining the processing resource corresponding to the information processing interface.

The following steps are executed by utilizing the processing resources corresponding to the information processing interface:

702: and responding to a parameter optimization request of the parameter to be optimized, and determining the local characteristic information of the parameter to be optimized when the global characteristic information of the parameter to be optimized meets a preset local switching rule.

703: and acquiring a first candidate solution of the parameter to be optimized according to the local characteristic information.

704: and if the first candidate solution meets the parameter use condition, determining the first candidate solution as a target solution of the parameter to be optimized.

In a possible design, the information processing interface may define the technical solution provided in the embodiment of the present application as a processing protocol, and obtain an application program interface providing a software service to the outside. The information processing Interface includes an SDK (Software Development Kit), an API (Application Programming Interface), and the like. In the application process of the technical scheme of the embodiment of the application, the target solution can be acquired by a user in an interface form, the access request of the user is received through the interface, the target solution is output to the user through the interface after the target solution is acquired, and a product object related to the target solution can be fed back to the user by taking a network transaction system as an example. In addition, in another possible design, in the process of using the technical solution of the embodiment of the present application, a usage log may be generated, and the usage log may be sent to the log server through the interface, so that the log server stores the usage log, and the usage log is used to effectively detect the usage process of the information processing method.

Optionally, the technical solution provided in the embodiment of the present application may be configured in a cloud server, and an information processing interface for providing an external information processing method is formed after the information processing method is encapsulated, and the information processing interface may be called by a user equipment to provide an information processing service for a user.

The specific steps executed by the processing resources corresponding to the information processing interface in the embodiment of the present application are the same as the processing steps executed by the information processing method shown in fig. 1, and the specific implementation manner and the technical effect of each technical feature are described in detail in the embodiment shown in fig. 1, and are not described again here.

For ease of understanding, fig. 8 shows a diagram of an application example provided by an embodiment of the present application. Referring to fig. 7, a technical solution of the embodiment of the present application is described in detail by taking an example that a target user can interact with a cloud server M2 through a computer M1 to implement information processing.

The computer M1 may detect 801 a parameter optimization request initiated by a target user and send 802 the parameter optimization request to the cloud server M2. The cloud server M2 may receive a parameter optimization request sent by the computer M1, and in response to the parameter optimization request for the parameter to be optimized, determine 803 local feature information of the parameter to be optimized when global feature information of the parameter to be optimized satisfies a preset local switching rule. A first candidate solution of the parameter to be optimized may be acquired 804 according to the local feature information. By judging the parameter use condition of the first candidate solution, when the first candidate solution meets the parameter use condition, the first candidate solution is determined 805 to be a target solution of the parameter to be optimized. And when the first candidate solution does not meet the parameter use condition, updating the first candidate solution into a second candidate solution, returning to the step of evaluating the local acquisition characteristics of the parameter to be optimized based on the second candidate solution and obtaining the local characteristic information of the parameter to be optimized, and continuing to execute the step. After obtaining the target solution, the cloud server M2 may send 806 the target solution to computer M1.

After receiving the target solution, the computer M1 may display 807 the target solution for the user, and obtain and use the target solution satisfying the parameter use condition for the target user, thereby achieving fast sampling of the parameter to be optimized and improving the sampling efficiency. When the computer M1 outputs the target solution for the target user, the specific output mode may include various forms, for example, the output mode may be data, page, information, or message, and the specific output mode of the target solution in the embodiment of the present application is not limited to a great number.

In practical application, the target solution obtained by the information processing method provided by the embodiment of the application can be directly applied to a model training scene of a machine learning model. For example, when the user optimized parameter is a hyper-parameter, the obtained target solution meeting the parameter usage is a target hyper-parameter. The machine learning model can be constructed by utilizing the target hyper-parameters, the machine learning model is trained by utilizing the training data, the model parameters of the machine learning model constructed by utilizing the target hyper-parameters are obtained, and the using effect of the machine learning model is better.

The technical scheme of the embodiment of the application can be applied to various fields such as artificial intelligence interaction, data retrieval, content recommendation, click rate prediction, intelligent factories and industrial control, and particularly has stronger applicability in the field of content recommendation, such as content recommendation in the fields of e-commerce, live video broadcast, social contact and online education, and in the field of resource allocation, such as financial product configuration, electric power resource, water resource and supply chain allocation.

For ease of understanding, the embodiments of the present application will be described in detail with respect to several practical fields of use as follows.

(1) The e-commerce field. The method is most common in application scenes such as feature search in the e-commerce field, product recommendation in a live broadcast scene, content recommendation, advertisement click rate calculation and the like, and in the embodiment, the content recommendation scene is taken as an example, and example deployment is performed. The general recommendation process in the recommended scene may be to perform parameterization setting on elements of the selected scene, obtain a parameter to be optimized composed of a plurality of sub-parameters affecting the scene, and identify different features of the scene by using the plurality of parameters. The information processing scheme in the embodiment of the application is provided for the target user in a service form, the specific service form comprises a webpage service, a software program service or a program module existing in an SDK or API form, and the target user obtains the information processing service in a form of browsing a parameter optimization interface, service software or a service module and the like.

The server for providing the information processing service can detect a parameter optimization request initiated by a target user, respond to the parameter optimization request of the parameter to be optimized, and determine the local characteristic information of the parameter to be optimized when the global characteristic information of the parameter to be optimized meets a preset local switching rule; acquiring and obtaining a first candidate solution of a parameter to be optimized according to the local characteristic information; and if the first candidate solution meets the parameter use condition, determining the first candidate solution as a target solution of the parameter to be optimized.

Taking a word-clicking recommendation scenario as an example, when a user clicks a search box in an APP (Application), the system will recommend a partial search word (Query word) for the user. The purpose of recommending search terms for the user is to mine the potential purchase demand of the user, increase the use stickiness of the user and improve the total commodity transaction amount. The search system utilizes the following architecture and combines a deep learning Encode-decoder network, namely an objective function, to predict the recommendation of search words. The number of search words can be used as a parameter to be optimized, assuming that the search effect of the number of search words is the highest as an optimization target. In the prior art, a target solution of a parameter to be optimized, which is formed by the number of search words, is manually set according to manual experience. By using the information processing method, the number of the search words can be automatically optimized by using the information processing method, and a target solution meeting the parameter use condition is selected. In the parameter optimization process, the global characteristic and the local characteristic of the objective function are considered at the same time, a second candidate solution is obtained by sampling on the global basis, and local analysis is carried out on the basis of the second candidate solution to realize local rapid sampling and obtain a first candidate solution with higher accuracy. And when the evaluation index of the first candidate solution meets the parameter use condition, determining the first candidate solution as a target solution, and determining the number of the search words by using the target solution. By setting the number of search words, the search efficiency of the user when using the APP is improved.

(2) In the social field, it is also common to recommend content to social users and recommend materials to students. The recommendation in the social field is that a social user browses a social application program, and the display interface of the application program outputs social content which is interesting to the user. Generally, the recommendation of the social field is to form sub-parameters by using options of historical browsing behaviors, attention fields, user information and the like of a user, and a plurality of sub-parameters can form parameters to be optimized. When the characteristic value of each sub-parameter is determined, characteristic information finally corresponding to the parameter to be optimized can be generated, and related content or products can be searched based on the characteristic information. In order to find content of interest to a social user, the number and kinds of parameters may be optimized to obtain accurate social user content.

The technical scheme of the embodiment of the application can be configured in the cloud server, and the parameter optimization request can be initiated by operation and maintenance personnel, namely, a target user can initiate the parameter optimization request to the cloud server. When the cloud server receives the parameter optimization request and the global characteristic information of the parameter to be optimized meets a preset local switching rule, determining the local characteristic information of the parameter to be optimized; acquiring and obtaining a first candidate solution of a parameter to be optimized according to the local characteristic information; and if the first candidate solution meets the parameter use condition, determining the first candidate solution as a target solution of the parameter to be optimized. Target content information recommended to the user can be generated through the target solution, and the recommended content can be promoted to be quickly and accurately fed back to the access user through the improvement of the parameter sampling efficiency.

(3) The financial field. Stock index simulation is a very important problem, and based on linear regression, models such as SVM (support vector machines) and LSTM (Long Short-Term Memory) are common to stock index simulation, and before the models are used, a proper model needs to be established. Many hyper-parameters can be involved in the model training process, such as time step in LSTM, feature _ dim, hidden feature, etc., and the contextual features of market, macroscopic factor, microscopic factor, emergency, etc., which can influence the selection of hyper-parameters. By utilizing the technology, the hyper-parameters influencing the construction of the mathematical model and the parameters such as the context characteristics influencing the selection of the hyper-parameters can be used as sub-parameters to form the parameters needing to be optimized. In order to obtain an accurate target solution, sub-parameters corresponding to the above features may be determined, and the parameters to be optimized are formed by using the plurality of sub-parameters.

By using the technical scheme of the embodiment of the application, when the global characteristic information of the parameter to be optimized meets the preset local switching rule, the local characteristic information of the parameter to be optimized can be determined; acquiring and obtaining a first candidate solution of a parameter to be optimized according to the local characteristic information; and if the first candidate solution meets the parameter use condition, determining the first candidate solution as a target solution of the parameter to be optimized. After the target solution is obtained, a machine learning model corresponding to the exponential simulation problem can be constructed by using the hyper-parameters in the obtained target solution, model training is carried out, and model parameters are obtained. And then, performing simulation calculation on data such as RMSE (Root Mean square Error) difference values of actual stock indexes of the index simulation problem by using a machine learning model obtained by training, wherein the obtained data can be used for stock market investment guidance.

(4) The resource allocation field takes power resource allocation as an example. The allocation of power resources usually involves many areas, each area can be represented by a corresponding parameter, and these parameters can respectively allocate a certain proportion of resources, and the allocation of resources can affect the information of area economy, population, environment, etc. The technical scheme of the embodiment of the application can be applied to the problems of dynamic pricing of the power market and power economic load distribution. In the following, a detailed description is mainly given of a specific application field of the power system.

In the dynamic pricing problem of the electric power market, the influence of the type of users and the amount of power consumption on the electric power market is more critical. The user type, the electricity consumption and the like can be used as parameters to be optimized, the parameters to be optimized are sampled, and candidate solutions can be obtained. The maximum profit/cost of the value of each sub-parameter in any candidate solution in the power system can be used as the judgment result of the parameter use condition. In the parameter optimization process, the power resource management party can initiate a parameter optimization request, and the background server can respond to the parameter optimization request of the parameter to be optimized and determine the local characteristic information of the parameter to be optimized when the global characteristic information of the parameter to be optimized meets a preset local switching rule; acquiring and obtaining a first candidate solution of a parameter to be optimized according to the local characteristic information; and if the first candidate solution meets the parameter use condition, determining the first candidate solution as a target solution of the parameter to be optimized. After providing the target solution to the power resource manager, the power resource manager may use the target solution to perform dynamic pricing of the power resources. In the parameter selection process, since the parameter sampling is carried out in the second parameter space which is simpler than the first parameter space, the complexity of parameter sampling can be reduced, and the parameter sampling efficiency is improved, thereby further improving the parameter optimization efficiency.

In the power economic load distribution problem, a power supplier can simultaneously provide power resources to a plurality of regions, the power load capacity of each region can be used as a sub-parameter, and the lowest total power consumption of a power grid can be used as a parameter using condition. By using the technical scheme of the embodiment of the application, the respective power load capacities of a plurality of regions can be sampled to obtain a candidate solution of a parameter to be optimized. And in the sampling process, the parameters to be optimized are composed of sub-parameters corresponding to the power load capacity of all regions. The power supplier can initiate a parameter optimization request, and the background server can respond to the parameter optimization request of the parameter to be optimized and determine the local characteristic information of the parameter to be optimized when the global characteristic information of the parameter to be optimized meets a preset local switching rule; acquiring and obtaining a first candidate solution of a parameter to be optimized according to the local characteristic information; and if the first candidate solution meets the parameter use condition, determining the first candidate solution as a target solution of the parameter to be optimized. After the target solution is determined, the target solution can be output by the power supplier, so that the power supplier can obtain the distribution strategy of the power load by using the target solution, the selection complexity of the parameters is reduced, and the selection efficiency is improved.

As shown in fig. 9, a schematic structural diagram of an embodiment of an information processing apparatus provided in an embodiment of the present application may include:

the request response module 901: the local feature information of the parameter to be optimized is determined when the global feature information of the parameter to be optimized meets a preset local switching rule in response to a parameter optimization request of the parameter to be optimized;

local sampling module 902: the method comprises the steps of acquiring a first candidate solution of a parameter to be optimized according to local characteristic information;

the goal determination module 903: and the target solution is used for determining the first candidate solution as the parameter to be optimized if the first candidate solution meets the parameter use condition.

As an embodiment, the request response module may include:

the first response unit is used for responding to a parameter optimization request of the parameter to be optimized and determining the global characteristic information of the parameter to be optimized;

the first processing unit is used for acquiring and obtaining a second candidate solution of the parameter to be optimized by utilizing a preset global sampling algorithm if the global characteristic information meets the local switching rule;

and the second processing unit is used for evaluating the local acquisition characteristics of the parameter to be optimized based on the second candidate solution to obtain the local characteristic information of the parameter to be optimized.

In some embodiments, the apparatus may further comprise:

and the condition skipping module is used for updating the first candidate solution into a second candidate solution if the first candidate solution does not meet the parameter use condition, and skipping to the local analysis module for continuous execution.

In some embodiments, the first response unit may include:

the first response subunit is configured to determine, in response to a parameter optimization request for a parameter to be optimized, initial effect information corresponding to a plurality of initial candidate solutions and a plurality of initial candidate solutions of the parameter to be optimized, respectively.

The model fitting subunit is used for fitting the distribution model of the parameter to be optimized according to the initial effect information respectively corresponding to the initial candidate solutions and the initial candidate solutions to obtain a probability distribution model;

the first processing unit may include:

and the first sampling subunit is used for performing parameter sampling on the parameter to be optimized by using a preset acquisition function based on the probability distribution model to obtain a second candidate solution.

Optionally, the request response module may further include:

and the third processing unit is used for determining first effect information of a second candidate solution if the probability distribution model does not meet the local switching rule, increasing the second candidate solution into an initial candidate solution, returning to a parameter optimization request responding to the parameter to be optimized, and determining a plurality of initial candidate solutions of the parameter to be optimized and initial effect information corresponding to the initial candidate solutions to be optimized respectively to continue execution.

In one possible design, the request response module may further include:

the statistical analysis unit is used for determining a statistical index of the probability distribution model; if the statistical index meets a preset index threshold, determining that the second candidate solution meets a local switching rule; and if the statistical index does not meet the preset index threshold, determining that the second candidate solution does not meet the local switching rule.

As still another embodiment, the second processing unit may further include:

the first analysis subunit is used for performing gradient estimation processing on the parameter to be optimized in the second candidate solution to obtain sampling displacement data;

the local sampling module may include:

and the first sampling unit is used for performing parameter sampling on the parameter to be optimized on the basis of the second candidate solution according to the sampling displacement data to obtain a first candidate solution.

In some embodiments, the first analysis subunit may be specifically configured to:

As still another embodiment, the apparatus may further include:

and the request receiving module is used for receiving a parameter optimization request sent by a target user.

And the target output module is used for outputting a target solution for the target user.

In some embodiments the apparatus may further comprise:

and the adjustment receiving module is used for detecting a parameter adjustment request initiated by the target user aiming at the target solution.

And the adjustment determining module is used for responding to the parameter adjustment request and acquiring the parameter adjustment information provided by the target user.

And the target adjusting module is used for adjusting the target solution according to the parameter adjusting information to obtain an expected solution of the target user.

As still another embodiment, the apparatus may further include:

the first detection module is used for detecting a resource management request triggered by the resource manager for the target resource.

The request response module may include:

and the second response unit is used for responding to the resource management request and acquiring a second candidate solution of the resource parameters corresponding to the target resource by utilizing a preset global sampling algorithm.

The apparatus may further include:

and the first generation module is used for generating resource setting information of the resource parameters based on the target solution.

And the first feedback module is used for feeding back the resource setting information to the resource manager so that the resource manager can set the target resource according to the resource setting information.

As still another embodiment, the apparatus may further include:

and the second detection module is used for detecting a system access request triggered by the access user aiming at the network transaction system.

The request response module may include:

the third response unit is used for responding to the system access request and acquiring a second candidate solution of the system access parameter corresponding to the network transaction system by using a preset global sampling algorithm;

the apparatus may further include:

and the second generation module is used for generating target access information corresponding to the system access parameter based on the target solution.

And the second feedback module is used for feeding back the target access information to the access user so that the access user can operate the network transaction system based on the target access information.

The information processing apparatus implementing fig. 9 can execute the information processing method of the embodiment shown in fig. 1, and the implementation principle and technical effect thereof are not described again. The specific manner in which each step is performed by each module or unit in the above embodiments has been described in detail in the embodiments related to the method, and will not be described in detail herein.

In practical applications, the embodiment shown in fig. 9 may be configured as a computing device, and referring to fig. 10, for a structural schematic diagram of an embodiment of a computing device provided in the embodiment of the present application, the device may include: a storage component 1001 and a processing component 1002; storage component 1001 is used to store one or more computer instructions; one or more computer instructions are invoked by the processing component 1002.

The processing component 1002 may be used to perform the information processing methods of the embodiments shown in fig. 1-7.

As one embodiment, the processing component 1002 may be to: responding to a parameter optimization request of a parameter to be optimized, and determining local acquisition characteristics of the parameter to be optimized for evaluation when global characteristic information of the parameter to be optimized meets a preset local switching rule to obtain local characteristic information of the parameter to be optimized; acquiring and obtaining a first candidate solution of a parameter to be optimized according to the local characteristic information; and if the first candidate solution meets the parameter use condition, determining the first candidate solution as a target solution of the parameter to be optimized.

Among other things, the processing component 1002 may include one or more processors to execute computer instructions to perform all or some of the steps of the methods described above. Of course, the processing elements may also be implemented as one or more Application Specific Integrated Circuits (ASICs), Digital Signal Processors (DSPs), Digital Signal Processing Devices (DSPDs), Programmable Logic Devices (PLDs), Field Programmable Gate Arrays (FPGAs), controllers, micro-controllers, microprocessors or other electronic components configured to perform the above-described methods.

The storage component 1001 is configured to store various types of data to support operations at the terminal. The memory components may be implemented by any type or combination of volatile or non-volatile memory devices such as Static Random Access Memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic or optical disks.

Of course, a computing device may also necessarily include other components, such as input/output interfaces, communication components, and so forth. The input/output interface provides an interface between the processing components and peripheral interface modules, which may be output devices, input devices, etc. The communication component is configured to facilitate wired or wireless communication between the computing device and other devices, and the like.

In addition, embodiments of the present application also provide a computer-readable storage medium, where one or more computer instructions may be stored, and when executed, the one or more computer instructions are used to implement any information processing method in the embodiments of the present application.

The above-described embodiments of the apparatus are merely illustrative, and the units described as separate parts may or may not be physically separate, and the parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment. One of ordinary skill in the art can understand and implement it without inventive effort.

Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented by adding a necessary general hardware platform, and of course, can also be implemented by a combination of hardware and software. With this understanding in mind, the above-described technical solutions and/or portions thereof that contribute to the prior art may be embodied in the form of a computer program product, which may be embodied on one or more computer-usable storage media having computer-usable program code embodied therein (including but not limited to disk storage, CD-ROM, optical storage, etc.).

Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit the same; although the present application has been described in detail with reference to the foregoing embodiments, it should be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions in the embodiments of the present application.

Claims

1. An information processing method characterized by comprising:

2. The method according to claim 1, wherein the determining, in response to the parameter optimization request for the parameter to be optimized, the local feature information of the parameter to be optimized when the global feature information of the parameter to be optimized satisfies a preset local switching rule comprises:

responding to a parameter optimization request of the parameter to be optimized, and determining global feature information of the parameter to be optimized;

if the global characteristic information meets the local switching rule, acquiring a second candidate solution of the parameter to be optimized by utilizing a preset global sampling algorithm;

and evaluating the local acquisition characteristics of the parameter to be optimized based on the second candidate solution to obtain the local characteristic information of the parameter to be optimized.

3. The method according to claim 2, wherein after acquiring and obtaining the first candidate solution of the parameter to be optimized according to the local feature information, the method further comprises:

if the first candidate solution does not meet the parameter use condition, updating the first candidate solution to be the second candidate solution, returning to the step of evaluating the local acquisition characteristics of the parameter to be optimized based on the second candidate solution, and obtaining the local characteristic information of the parameter to be optimized.

4. The method of claim 2, wherein the determining global feature information of the parameter to be optimized in response to the parameter optimization request of the parameter to be optimized comprises:

responding to a parameter optimization request of the parameter to be optimized, and determining a plurality of initial candidate solutions of the parameter to be optimized and initial effect information corresponding to the initial candidate solutions respectively;

fitting the distribution model of the parameters to be optimized according to the initial candidate solutions and the initial effect information corresponding to the initial candidate solutions respectively to obtain a probability distribution model;

if the global feature information meets the local switching rule, acquiring and obtaining a second candidate solution of the parameter to be optimized by using a preset global sampling algorithm comprises the following steps:

and if the probability distribution model meets a local switching rule, performing parameter sampling on the parameter to be optimized by using a preset acquisition function based on the probability distribution model to obtain the second candidate solution.

5. The method of claim 4, further comprising:

and if the probability distribution model does not meet the local switching rule, determining first effect information of a second candidate solution, increasing the second candidate solution into an initial candidate solution, returning to the parameter optimization request responding to the parameter to be optimized, and determining a plurality of initial candidate solutions of the parameter to be optimized and initial effect information corresponding to the initial candidate solutions to be optimized respectively to continue execution.

6. The method according to any of claims 4 or 5, wherein whether the probability distribution model satisfies a local switching rule is determined by:

determining a statistical indicator of the probability distribution model;

and if the statistical index does not meet a preset index threshold value, determining that the second candidate solution does not meet the local switching rule.

7. The method according to claim 2, wherein the evaluating the local acquisition feature of the parameter to be optimized based on the second candidate solution, and obtaining the local feature information of the parameter to be optimized comprises:

the acquiring and obtaining a first candidate solution of the parameter to be optimized according to the local feature information includes:

and performing parameter sampling on the parameter to be optimized on the basis of the second candidate solution according to the sampling displacement data to obtain the first candidate solution.

8. The method according to claim 6, wherein the performing gradient estimation processing on the parameter to be optimized at the second candidate solution to obtain sample displacement data comprises:

and carrying out disturbance analysis processing on the second candidate solution based on a gradient estimation algorithm to obtain the sampling displacement data.

9. The method of claim 1, further comprising:

receiving a parameter optimization request sent by a target user;

after determining that the first candidate solution is the target solution of the parameter to be optimized if the first candidate solution satisfies the parameter use condition, the method further includes:

outputting the target solution for the target user.

10. The method of claim 9, further comprising:

detecting a parameter adjustment request initiated by the target user aiming at the target solution;

responding to the parameter adjustment request, and acquiring parameter adjustment information provided by the target user;

11. An information processing method characterized by comprising:

12. An information processing method characterized by comprising:

13. An information processing method characterized by comprising:

14. An information processing apparatus characterized by comprising:

15. A computing device, comprising: a storage component and a processing component; the storage component is used for storing one or more computer instructions; the one or more computer instructions are invoked by the processing component;

the processing component is to: