Anti-noise doping signal processing method and system
By setting the hyperparameter candidate set in the addable model, normalizing the processing labels, and selecting the optimal hyperparameters through cross-validation, and optimizing the model parameters to generate an anti-noise parameter signal processing model, the problem of the existing addable model's poor performance when processing complex data is solved, and its performance on non-i.i.d data and complex learning tasks is improved.
Patent Information
- Application Number
- CN202411882715.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-19
- Publication Date
- 2025-04-11
AI Technical Summary
The existing addable model based on the empirical risk minimization (ERM) framework does not perform well when dealing with situations such as complex data distribution offsets, non-Gaussian noise, imbalance labels, etc., and it is difficult to effectively process complex signals.
By setting up the hyperparameter candidate set of the model and optimization algorithm, separating the input features and output labels, normalizing the output labels, initializing the model coefficients, selecting the optimal hyperparameters through K fold cross-validation, and optimizing the model parameters to generate a signal processing model with anti-noise parameters.
Improve the processing capability of non-independent homogeneous distribution (i.i.d) data, enhance the learning ability under non-i.i.d conditions, and improve the performance of the additive model when dealing with unbalanced classification, robust estimation, and multi-objective learning.
Smart Images
Figure CN119807899A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of signal processing technology, and in particular to a signal processing method and system that is resistant to noise interference. Background Art
[0002] As a natural extension of linear models, additive models have attracted great attention in high-dimensional data analysis due to their flexible representation and inherent interpretability. They have been widely studied in terms of theoretical foundations and practical applications. The model is mainly built on the empirical risk minimization (ERM) framework; the ERM framework can flexibly embed various loss functions, such as squared loss in regression tasks. Due to their flexible representation and inherent interpretability, additive models are becoming increasingly popular in data analysis. However, existing additive model algorithms based on the ERM framework often perform poorly in cases with complex data, such as those with data distribution shift, complex non-Gaussian noise, unbalanced labels, or multiple complex problems. Therefore, it is urgent to design a new additive model to improve the effectiveness of traditional models in signal processing. Summary of the invention
[0003] In order to solve the above problems, the purpose of the present invention is to provide a signal processing technology that is resistant to noise interference, aiming to improve the traditional additive model and achieve its effectiveness in processing signals with distribution offset, noise disturbance and class imbalance.
[0004] In order to achieve the above technical objectives, the present application provides a signal processing method for resisting noise interference, comprising the following steps:
[0005] Set the candidate set of hyperparameters for the model and optimization algorithm, and separate the input features and output labels;
[0006] After normalizing the output labels and mapping them to between -1 and 1, the model coefficients are initialized, and the optimal hyperparameters are selected from the hyperparameter candidate set through K-fold cross validation;
[0007] After optimizing the model parameters according to the optimal hyperparameters, an optimized model is generated to process signals mixed with noise.
[0008] Preferably, in the process of obtaining input features, observed and unobserved samples are obtained as input features; wherein the samples are regarded as the result of a random process, and each specific observed sample is set to have a unique distribution; and unobserved samples are extracted from the underlying distribution.
[0009] Preferably, in the process of constructing the WTERM learning framework, an additive hypothesis function space is constructed to achieve nonlinear estimation capabilities.
[0010] Preferably, in the process of minimizing the objective of WTERM, a nuclear norm-induced sparse penalty is set in the data-dependent hypothesis space based on the additive hypothesis space to achieve variable selection.
[0011] Preferably, in the process of building the estimator, by generalizing WTERM to multi-objective WTERM, one can further address compound learning problems, including simultaneously handling label noise and data imbalance.
[0012] Preferably, in the process of constructing the (multi-objective) WTERM, the optimal hyperparameters are selected by comparing the prediction performance of the (multi-objective) WTERM for the validation samples under different hyperparameter settings.
[0013] The present invention discloses a signal processing system for resisting noise interference, which is used to implement a signal processing method for resisting noise interference. The system comprises:
[0014] The parameter processing module is used to set the candidate set of hyperparameters for the model and optimization algorithm, and to separate input features and output labels;
[0015] The parameter selection module is used to normalize the output labels and map them to between -1 and 1, initialize the model coefficients, and select the optimal hyperparameters from the hyperparameter candidate set through K-fold cross validation;
[0016] The signal processing module is used to optimize the model parameters according to the optimal hyperparameters and generate an optimized model for processing signals mixed with noise.
[0017] The present invention discloses the following technical effects:
[0018] The present invention greatly improves the capability for non-iid data and provides effective insights for constructing optimized independent error bounds for various kernel-based methods and deep networks.
[0019] The present invention further enriches the simulation experiments under non-IID conditions and enriches the actual experiments by simulating noise disturbance scenarios.
[0020] The present invention improves the learning ability of additive models when dealing with challenges such as imbalanced classification, robust estimation and multi-objective learning, and effectively addresses the above-mentioned limitations of sparse additive models; it effectively combines the interpretability of additive models with the estimation flexibility of WTERM. BRIEF DESCRIPTION OF THE DRAWINGS
[0021] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0022] Figure 1 It is a schematic diagram of the method flow of the present invention. DETAILED DESCRIPTION
[0023] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. The components of the embodiments of the present application generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present application.
[0024] like Figure 1 As shown, the present invention provides a signal processing method for resisting noise interference, comprising the following steps:
[0025] Set the candidate set of hyperparameters for the model and optimization algorithm, and separate the input features and output labels;
[0026] After normalizing the output labels and mapping them to between -1 and 1, the model coefficients are initialized, and the optimal hyperparameters are selected from the hyperparameter candidate set through K-fold cross validation;
[0027] After optimizing the model parameters according to the optimal hyperparameters, an optimized model is generated to process signals mixed with noise.
[0028] Further preferably, the present invention provides a signal processing method resistant to noise interference, which obtains observed and unobserved samples as input features in the process of obtaining input features; wherein the samples are regarded as the result of a random process, and each specific observed sample is set to have a unique distribution; and unobserved samples are extracted from the basic distribution.
[0029] Further preferably, in the noise-resistant signal processing method provided by the present invention, in the process of constructing the WTERM learning framework, an additive hypothesis space is set to achieve nonlinear estimation capability.
[0030] Further preferably, the present invention provides a signal processing method resistant to noise interference, in the process of constructing an additive model, based on the additive hypothesis space, considers sparse learning in the data-dependent hypothesis space to obtain coefficient-induced penalty terms to achieve variable selection.
[0031] Further preferably, the noise-resistant signal processing method provided by the present invention can further solve the compound learning problem, including simultaneously processing label noise and data imbalance, by generalizing WTERM to multi-objective WTERM in the process of constructing an estimator.
[0032] Further preferably, in the process of constructing a (multi-objective) WTERM, the present invention provides a signal processing method resistant to noise interference, and selects the optimal hyperparameters by comparing the prediction performance of the (multi-objective) WTERM for the validation samples under different hyperparameter settings.
[0033] The present invention discloses a signal processing system for resisting noise interference, which is used to implement a signal processing method for resisting noise interference. The system comprises:
[0034] The parameter processing module is used to set the candidate set of hyperparameters for the model and optimization algorithm, and to separate input features and output labels;
[0035] The parameter selection module is used to normalize the output labels and map them to between -1 and 1, initialize the model coefficients, and select the optimal hyperparameters from the hyperparameter candidate set through K-fold cross validation;
[0036] The signal processing module is used to optimize the model parameters according to the optimal hyperparameters and generate an optimized model for processing signals mixed with noise.
[0037] Embodiment: The present invention provides a signal processing technology for resisting noise interference, which specifically includes the following contents:
[0038] The present invention defines the input variable X as a member of a p-dimensional metric space, and the output variable Y as a real-valued response of the output space.
[0039] The present invention has a set of (X, Y) observations of size n, denoted as Using x ij represents the jth dimension of the ith sample. The unobserved sample to be predicted is represented by z = (x, y).
[0040] According to traditional statistical learning theory, both observed and unobserved samples are considered to be independently drawn from the underlying distribution ρ. However, the present invention goes beyond the standard independent and identically distributed assumption and regards the samples as the result of a random process, where each specific observation z iAll come from its unique distribution ρ i , and the unobserved samples are drawn from ρ. Note that ρ i and ρ can represent different distributions. This data generation setting provides enough flexibility to cover scenarios where the traditional independent and identically distributed setting is applicable. Specifically, if the distribution ρ i , i=1,…,n, are the same and independent of the distribution ρ, then the setting degenerates to the independent and identically distributed case. Considering the data set The estimator obtained by ERM is given by:
[0041]
[0042] Where H is a measurable hypothesis space and l is the loss function. Despite their generality, many existing ERM-based methods can struggle when the average performance fails to capture the specific challenges encountered. This is particularly evident when data distribution drift, complex non-Gaussian noise, label imbalance, or a combination of these are involved.
[0043] In the present invention, a WTERM learning framework is proposed, by transforming the weight vector q=(q 1 ,…,q n ) is integrated into the existing TERM and redefined as:
[0044]
[0045] In the proposed WTERM learning framework, the parameter t is defined in the range (-∞,0)∪(0,+∞), and q i represents the weight assigned to the ith sample. These two hyperparameters play an important role in the learning process. For example, by choosing different values of t, the algorithm under WTERM can be fine-tuned to amplify (when t∈(0,+∞]) or suppress (when t∈[-∞,0)) the impact of outliers. In addition, the theoretical insights of the present invention show that by judiciously assigning the weight q=(q 1 ,…,q n ) T Given each sample, WTERM can capture the changes in data distribution, thereby improving the accuracy of learning from non-IID data.
[0046] The goal of minimizing WTERM is to select a predictor f from a specified hypothesis space z , so that it has a smaller overall risk for unseen data z = (x, y).
[0047]
[0048] The present invention considers an additive hypothesis space because it has good approximation ability and interpretability. The most common representation of an additive hypothesis space is to directly divide the input space into p parts, that is, and is defined by:
[0049]
[0050] The component hypothesis space Η j , j = 1, ..., p is conceptualized as RKHS and denoted as Where K j Refers to kernel functions, such as Gaussian kernel and Sigmoid kernel.
[0051] The additive model based on WTERM can be expressed in the form with nuclear norm penalty:
[0052]
[0053] Where η>0 is the regularization parameter, is a positive weight. Due to the properties of RKHS, the estimator of equation (2) can be expressed as:
[0054]
[0055] Based on the above representation, we next consider sparse learning in the data-dependent hypothesis space.
[0056]
[0057] With coefficient-induced penalty:
[0058]
[0059] in Then WT-SAM can be expressed as:
[0060]
[0061] The estimator of WT-SAM is rewritten as:
[0062]
[0063] at the same time,
[0064]
[0065] The proposed WT-SAM can be constructed by transforming the individual component space Replace with defined subgroup to seamlessly adapt to the group sparse setting. Here, each subgroup χ dRepresents a component input space that contains interactions between specific variables. Common methods for building these grouped component spaces include using functional) ANOVA models to study d-order interactions (1≤d≤p), or decomposing the inputs using prior knowledge.
[0066] In practical applications, a multi-objective WTERM can be constructed to solve compound learning problems, such as dealing with label noise and data imbalance at the same time. By classifying samples into different groups g∈G, a multi-objective WTERM can be constructed.
[0067]
[0068] in
[0069]
[0070] in represents the weight of the i-th sample in the g-th group. Obviously, when |G|=1, the multi-objective WTERM degenerates into WTERM. For example,
[0071] In the classification task, the present invention groups samples according to their categories, which enables the present invention to achieve robustness within each class (by setting each γ g <0) and solve the imbalance problem between classes (by setting t>0). Correspondingly, the expected version of multi-objective WTERM describes the average expected performance of unknown samples belonging to different groups and is defined as
[0072]
[0073] in
[0074]
[0075] We propose a new learning paradigm, called Weighted Skewed Empirical Risk Minimization (WTERM), for sparse additive models, which overcomes the limitations of ERM-based additive models in handling complex data by introducing skew and weighted influence on individual losses. This proposed learning paradigm ensures that sparse additive models can effectively handle a variety of learning tasks, including distribution-out-of-balance prediction, imbalanced classification, robust estimation, and multi-objective learning. Theoretically, we novelly establish an optimization-related upper bound on the uniform stability of WTERM-based sparse additive models under prox-SAGA optimization, which provides two key theoretical findings for the present invention, which are applicable to both independent and identically distributed (iid,) and non-iid, scenarios: generalization constraints and oracle analysis. Specifically, oracle analysis explores the ability of our method to perform function approximation on true functions under specific complex noises. Empirical evaluation shows that our method performs competitively on both synthetic and real data.
[0076] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.
[0077] In the description of the present invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.
[0078] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.
Claims
1. A signal processing method for resisting noise interference, characterized in that: The following steps are involved: Set the candidate set of hyperparameters for the model and optimization algorithm, and separate the input features and output labels; After normalizing the output labels and mapping them to between -1 and 1, the model coefficients are initialized, and the optimal hyperparameters are selected from the hyperparameter candidate set through K-fold cross validation; After optimizing the model parameters according to the optimal hyperparameters, an optimized model is generated for processing signals mixed with noise.
2. The method for signal processing against noise interference according to claim 1, characterized in that: In the process of obtaining input features, observed and unobserved samples are obtained as input features; wherein the samples are regarded as the result of a random process, and each specific observed sample is assumed to have a unique distribution; and unobserved samples are drawn from the underlying distribution.
3. The method for signal processing against noise interference according to claim 2, characterized in that: In the process of constructing the WTERM learning framework, an additive hypothesis function space is constructed to achieve nonlinear estimation capabilities.
4. The method for signal processing against noise interference according to claim 3, characterized in that: In the process of minimizing the WTERM objective, based on the additive hypothesis space, a nuclear norm-induced sparse penalty is set in the data-dependent hypothesis space to achieve variable selection.
5. The signal processing method for resisting noise interference according to claim 4, characterized in that: In the process of building the estimator, WTERM is generalized to multi-objective WTERM to solve compound learning problems, including dealing with label noise and data imbalance at the same time.
6. The method for signal processing against noise interference according to claim 5, characterized in that: In the process of constructing (multi-objective) WTERM, the optimal hyperparameters are selected by comparing the prediction performance of (multi-objective) WTERM for validation samples under different hyperparameter settings.
7. A signal processing system resistant to noise interference, characterized in that: The system is used to implement a signal processing method for resisting noise interference as described in any one of claims 1 to 6, and the system comprises: The parameter processing module is used to set the candidate set of hyperparameters for the model and optimization algorithm, and to separate input features and output labels; A parameter selection module is used to normalize the output labels and map them to between -1 and 1, initialize the model coefficients, and select the optimal hyperparameter from the hyperparameter candidate set through K-fold cross validation; The signal processing module is used to optimize the model parameters according to the optimal hyperparameters and generate an optimized model for processing signals mixed with noise.