A local learning and adaptive inference method and system

Abstract

The application relates to the technical field of microprocessors, and specifically provides a local learning and adaptive inference method and system. First, millimeter wave radar data collected is used for data cleaning, then a TinyML model is trained, and finally the model is deployed to a microprocessor. Meanwhile, millimeter wave radar data is transmitted to a fully connected neural network layer in the microcontroller as input features, then each input feature data corresponds to a current microprocessor device state parameter as a label of the input feature data, with the continuous increase of the input feature data and the corresponding label, the weight parameters of the fully connected neural network layer are optimized according to the actual state of the device, and finally the updated parameters are updated to the TinyML model for local learning and adaptive inference. Compared with the prior art, the application reduces the dependence on high-performance edge devices, and reduces the security problems in the data transmission process of the traditional scheme.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of microprocessors, and particularly provides a local learning and adaptive inference method and system. BACKGROUND

[0002] Existing detection of some outdoor environments and living bodies adopts an edge gateway plus camera module solution. On the one hand, the edge gateway needs wired power supply, which greatly reduces the deployment range. Meanwhile, the related recognition algorithm based on the camera needs a high-performance computing device, but the recognition accuracy in some extreme weather environments will be greatly reduced. Meanwhile, the gateway device has high requirements for the network, and cannot correctly infer according to the existing artificial intelligence model in some sudden situations.

[0003] This method needs a high-performance computing device and a wired deployment mode, which greatly improves the deployment limitations and costs. Secondly, the existing camera solution is greatly affected by the environment. Once the surrounding environment deviates from the deployed model, the inference result will be greatly affected, resulting in the loss of the function of the device itself. SUMMARY

[0004] The application is aimed at the deficiencies of the prior art, and provides a local learning and adaptive inference method with high practicability.

[0005] The further technical task of the application is to provide a local learning and adaptive inference system which is reasonable in design and safe and applicable.

[0006] The application adopts the technical scheme to solve the technical problems:

[0007] A local learning and adaptive inference method, first, millimeter wave radar data collected is used for data cleaning, then a TinyML model is trained, and finally the model is deployed to a microprocessor;

[0008] Meanwhile, the millimeter wave radar data is transmitted to a fully connected neural network layer in the microcontroller as input feature, then each input feature data corresponds to a current microprocessor device state parameter as the label of the input feature data, with the continuous increase of the input feature data and the corresponding label, the weight parameters of the fully connected neural network layer are also optimized according to the actual state of the device, and finally the updated parameters are updated to the TinyML model to perform local learning and adaptive inference.

[0009] Further, first, the fully connected neural network layer updates parameters in the microprocessor device, and the millimeter wave radar data stream is used as the input layer of the neural network, and the label corresponding to the input layer is the device state connected to the microprocessor or the controllable variable;

[0010] Then, the weight parameters of the current model are iteratively optimized through the calculation of the fully connected neural network layer, the latest model parameters are transmitted to the TinyML model, adaptive inference is realized, and the inference results of the latest model parameters are used to control and perform subsequent operations on the connected device.

[0011] Further, during the running of the microprocessor, the model weight parameters are shared with the fully connected neural network layer, so that the updated model parameters are updated to the TinyML model, the real-time millimeter wave radar data stream and the updated model weight parameters are combined, the effect of local learning is achieved, and the adaptive inference result is output.

[0012] Further, the data stream of the millimeter wave radar is divided into three parts, i.e., (moving target, energy value), (stationary target, energy value), and (detection distance, distance), and the data structure is arranged as {(moving target, energy value), (stationary target, energy value), (detection distance, distance), (label, device state)} by combining the current state of the device.

[0013] The data structure is used as the input of the fully connected neural network layer to realize local learning and iteration of the model parameters.

[0014] A local learning and adaptive inference system first uses the collected millimeter wave radar data for data cleaning, then trains a TinyML model, and finally deploys the model to a microprocessor.

[0015] Meanwhile, the millimeter wave radar data as input features is transmitted to the fully connected neural network layer in the microcontroller, and then each input feature data corresponds to a current microprocessor device state parameter as a label of the input feature data. With the continuous increase of input feature data and corresponding labels, the weight parameters of the fully connected neural network layer are also optimized according to the actual state of the device, and finally the updated parameters are updated to the TinyML model for local learning and adaptive inference.

[0016] Further, first, the fully connected neural network layer updates parameters in the microprocessor device, and the millimeter wave radar data stream is used as the input layer of the neural network. The label corresponding to the input layer is the state of the device connected to the microprocessor or the controllable variable.

[0017] Then, the weight parameters of the current model are iteratively optimized through the calculation of the fully connected neural network layer, the latest model parameters are transmitted to the TinyML model, adaptive inference is realized, and the inference results of the latest model parameters are used to control and perform subsequent operations on the connected device.

[0018] Further, during the running of the microprocessor, the model weight parameters are shared with the fully connected neural network layer, so that the updated model parameters are updated to the TinyML model, and the real-time millimeter wave radar data stream and the updated model weight parameters are combined to achieve the effect of local learning and output adaptive inference results.

[0019] Further, the data stream of the millimeter wave radar is divided into three parts, namely (moving target, energy value) (stationary target, energy value) (detection distance, distance), and the data structure is arranged as {(moving target, energy value), (stationary target, energy value), (detection distance, distance), (label, device state)} by combining the current state of the device.

[0020] The data structure is used as the input of the fully connected neural network layer to realize local learning and iteration of the model parameters.

[0021] Compared with the prior art, the local learning and adaptive inference method and system of the present application has the following outstanding beneficial effects:

[0022] The present application realizes local learning of the model through the data stream of the millimeter wave radar and the fully connected neural network layer. The latest model parameters of the local learning are transmitted to the TinyML model through the weight parameter sharing mechanism of the TinyML model and the fully connected neural network layer, and the model is continuously iterated to achieve local learning and adaptive inference results of the model. At the same time, according to the adaptive inference results, the subsequent control and operation of the related device are carried out. The dependence on high-performance edge devices is reduced, and the security problems in the data transmission process of the traditional scheme are reduced. BRIEF DESCRIPTION OF DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and those skilled in the art can obtain other drawings without creative labor based on these drawings.

[0024] Figure Figure 1 It is a flowchart of a local learning and adaptive inference method. DETAILED DESCRIPTION

[0025] In order to make the person skilled in the art better understand the scheme of the present application, the present application will be further described in detail in combination with specific embodiments. Obviously, the described embodiments are only some of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0026] A preferred embodiment is given below:

[0027] As shown in Figure 1 , in this embodiment, a local learning and adaptive inference method is used. First, the collected millimeter wave radar data is cleaned, then the TinyML model is trained, and finally the model is deployed to the microprocessor.

[0028] At the same time, the millimeter wave radar data is transmitted to the fully connected neural network layer in the microcontroller as input features, and then each input feature data corresponds to a current microprocessor device state parameter as the label of the input feature data. With the continuous increase of input feature data and corresponding labels, the weight parameters of the fully connected neural network layer are optimized according to the actual state of the device, and finally the updated parameters are updated to the TinyML model for local learning and adaptive inference.

[0029] First, the fully connected neural network layer updates the parameters in the microprocessor device, and the millimeter wave radar data stream is used as the input layer of the neural network. The label corresponding to the input layer is the device state or controllable variable connected to the microprocessor.

[0030] Then, the weight parameters of the current model are optimized through the calculation of the fully connected neural network layer. By transmitting the latest model parameters to the TinyML model, adaptive inference is realized, so that the inference result of the latest model parameters is used to control the connected device and perform subsequent operations.

[0031] During the operation of the microprocessor, the model weight parameters are shared with the fully connected neural network layer, so that the updated model parameters are updated to the TinyML model. By combining the real-time millimeter wave radar data stream and the updated model weight parameters, the effect of local learning is achieved and the adaptive inference result is output.

[0032] The data stream of the millimeter wave radar is divided into three parts: (moving target, energy value), (stationary target, energy value), and (detection distance, distance). At the same time, the current state of the device is combined to organize the data structure as {(moving target, energy value), (stationary target, energy value), (detection distance, distance), (label, device state)}.

[0033] The data structure is used as the input of the fully connected neural network layer to realize local learning and iteration of model parameters.

[0034] Based on the above method, a local learning and adaptive inference system in this embodiment is used. First, the collected millimeter wave radar data is cleaned, then the TinyML model is trained, and finally the model is deployed to the microprocessor.

[0035] At the same time, the millimeter wave radar data as input features will be transmitted to the fully connected neural network layer in the microcontroller, and then each input feature data will correspond to a current microprocessor device state parameter as the label of the input feature data. With the continuous increase of input feature data and corresponding labels, the weight parameters of the fully connected neural network layer will also be optimized according to the actual state of the device. Finally, the updated parameters will be updated to the TinyML model for local learning and adaptive inference.

[0036] First, the fully connected neural network layer updates the parameters in the microprocessor device, and the millimeter wave radar data stream is taken as the input layer of the neural network. The label corresponding to the input layer is the device state or controllable variable connected to the microprocessor.

[0037] Then, the weight parameters of the current model are optimized through the calculation of the fully connected neural network layer. By transmitting the latest model parameters to the TinyML model, adaptive inference is realized, so as to use the inference result of the latest model parameters to control the connected device and perform subsequent operations.

[0038] During the running of the microprocessor, the model weight parameters are shared with the fully connected neural network layer, so that the updated model parameters are updated to the TinyML model. Combining the real-time millimeter wave radar data stream and the updated model weight parameters, the effect of local learning is achieved and the adaptive inference result is output.

[0039] The data stream of the millimeter wave radar is divided into three parts, namely (moving target, energy value), (stationary target, energy value) and (detection distance, distance). At the same time, the current state of the device is combined to arrange the data structure as {(moving target, energy value), (stationary target, energy value), (detection distance, distance), (label, device state)}.

[0040] The data structure is taken as the input of the fully connected neural network layer to realize local learning and iteration of model parameters.

[0041] The above specific embodiments are only specific cases of the present application. The patent protection scope of the present application includes but is not limited to the above specific embodiments. Any appropriate changes or replacements made by any ordinary skilled person in the art to the local learning and adaptive inference method and system of the present application shall fall within the patent protection scope of the present application.

[0042] While embodiments of the application have been shown and described, it is to be understood that the embodiments described are merely exemplary of the principles and application of the present application. Numerous modifications and adaptions can be effected without departing from the spirit and scope of the present application, which is not limited to the exact construction and arrangement described. It is intended, therefore, to cover all modifications and adaptions that fall within the scope of the claims and their equivalents.

Claims

1. A local learning and adaptive inference method, characterized in that, First, the collected millimeter-wave radar data is cleaned, then the TinyML model is trained, and finally the model is deployed to the microprocessor. Meanwhile, millimeter-wave radar data, as input features, is transmitted to the fully connected neural network layer in the microcontroller. Each input feature data is then labeled with a current state parameter of the microprocessor device. As the input feature data and corresponding labels increase, the weight parameters of the fully connected neural network layer are optimized according to the actual state of the device. Finally, the updated parameters are updated in the TinyML model for local learning and adaptive inference. The data stream of the millimeter-wave radar is decomposed into three parts: (moving target, energy value), (stationary target, energy value), and (detection range, range). At the same time, combined with the current status of the device, the data structure is organized into {(moving target, energy value), (stationary target, energy value), (detection range, range), (tag, device status)}. The data structure is used as input to a fully connected neural network layer to enable local learning and iteration of model parameters.

2. The local learning and adaptive inference method according to claim 1, characterized in that, First, the fully connected neural network layer updates parameters on the microprocessor device, using millimeter-wave radar data stream as the input layer of the neural network. The labels corresponding to the input layer are the device status or controllable variables connected to the microprocessor. Then, the weight parameters of the current model are optimized and iterated through the calculation of the fully connected neural network layer. By transmitting the latest model parameters to the TinyML model, adaptive inference is achieved, thereby using the inference results of the latest model parameters to control and perform subsequent operations on the connected devices.

3. The local learning and adaptive inference method according to claim 2, characterized in that, During microprocessor operation, the model weight parameters are shared with the fully connected neural network layers, thereby updating the model parameters in the TinyML model. Combined with real-time millimeter-wave radar data streams and updated model weight parameters, local learning is achieved and adaptive inference results are output.

4. A local learning and adaptive reasoning system, characterized in that, First, the collected millimeter-wave radar data is cleaned, then the TinyML model is trained, and finally the model is deployed to the microprocessor. Meanwhile, millimeter-wave radar data, as input features, is transmitted to the fully connected neural network layer in the microcontroller. Each input feature data is then labeled with a current state parameter of the microprocessor device. As the input feature data and corresponding labels increase, the weight parameters of the fully connected neural network layer are optimized according to the actual state of the device. Finally, the updated parameters are updated in the TinyML model for local learning and adaptive inference. The data stream of the millimeter-wave radar is decomposed into three parts: (moving target, energy value), (stationary target, energy value), and (detection range, range). At the same time, combined with the current status of the device, the data structure is organized into {(moving target, energy value), (stationary target, energy value), (detection range, range), (tag, device status)}. The data structure is used as input to a fully connected neural network layer to enable local learning and iteration of model parameters.

5. The local learning and adaptive inference system according to claim 4, characterized in that, First, the fully connected neural network layer updates parameters on the microprocessor device, using millimeter-wave radar data stream as the input layer of the neural network. The labels corresponding to the input layer are the device status or controllable variables connected to the microprocessor. Then, the weight parameters of the current model are optimized and iterated through the calculation of the fully connected neural network layer. By transmitting the latest model parameters to the TinyML model, adaptive inference is achieved, thereby using the inference results of the latest model parameters to control and perform subsequent operations on the connected devices.

6. The local learning and adaptive inference system according to claim 5, characterized in that, During microprocessor operation, the model weight parameters are shared with the fully connected neural network layers, thereby updating the model parameters in the TinyML model. Combined with real-time millimeter-wave radar data streams and updated model weight parameters, local learning is achieved and adaptive inference results are output.

Images