Inference computing apparatus, model training apparatus, inference computing system

Patent metadata

Field	Value
Publication number	US-12236362-B2
Application number	US-201917044276-A
Country	US
Kind code	B2
Filing date	Dec 20, 2019
Priority date	Dec 20, 2019
Publication date	Feb 25, 2025
Grant date	Feb 25, 2025

Abstract

An inference computing apparatus comprises at least one processor and a memory with program instructions stored therein, the program instructions can be executed by the at least one processor to cause the inference computing apparatus to perform the following operations: receiving a first inference model from a model training apparatus, the first inference model being obtained through a model training by the model training apparatus based on a first training sample library, the first training sample library comprising training samples from historical data generated in a manufacturing stage; performing an inference computing on data to be processed generated in the manufacturing stage based on the first inference model to obtain the inference result which is sent to a user-side device; and evaluating performance of the first inference model to determine whether the first inference model needs to be updated, and if yes, updating the first inference model.

First claim

What is claimed is:

1. An inference computing apparatus, comprising at least one processor and a memory, wherein the memory has stored therein program instructions that, when executed by the at least one processor, cause the inference computing apparatus to perform following operations:
receiving a first inference model from a model training apparatus, the first inference model being obtained by the model training apparatus through a first model training based on a first training sample library, the first training sample library including first training samples from historical data generated in a manufacturing stage, and the model training apparatus including a cloud device;
performing an inference computing on data to be processed generated in the manufacturing stage based on the first inference model to obtain an inference result, and sending the inference result to a user-side device, the inference computing apparatus being closer to the user-side device than the model training apparatus;
evaluating a performance of the first inference model to determine whether the first inference model needs to be updated; and
updating the first inference model when the first inference model needs to be updated, wherein updating the first inference model includes:
performing a second model training based on a second training sample library to obtain a second inference model, or sending a model update request to the model training apparatus to obtain the second inference model, the second training sample library including: second training samples from the historical data, third training samples being from the inference result and subjected to re-judging, or the second training samples from the historical data and the third training samples being from the inference result and subjected to re-judging; and
updating the first inference model with the second inference model in a case where the second inference model meets an update condition, wherein the update condition includes that:
a test is performed on the second inference model and the second inference model passes the test, wherein the test includes: evaluating a performance of the second inference model based on test samples; and when the performance of the second inference model meets evaluation requirements, determining that the second inference model passes the test; and
a gray-scale deployment is performed on the second inference model, the performance of the second inference model is evaluated during the gray-scale deployment, and the performance of the second inference model meets the evaluation requirements, wherein the gray-scale deployment is that the inference computing apparatus performs a simulation processing using the second inference model within a preset period of time;

wherein before performing the second model training to obtain the second inference model, or before sending the model update request to the model training apparatus to obtain the second inference model, the inference computing apparatus further performs following operations:
determining whether training parameters required for the second model training to be performed are within a preset range of the training parameters; and when the training parameters are within the preset range, performing the second model training; when the training parameters are not within the preset range, sending the model update request to the model training apparatus, wherein the training parameters include at least one of data size, training duration and computing power required for the second model training, and the preset range of the training parameters is a range of the training parameters which corresponds to the case where a training capacity of the inference computing apparatus meets the requirements of the second model training to be performed.

2. The inference computing apparatus according to claim 1 , wherein
the historical data includes a product image marked with defect information having a correct result;
the data to be processed includes an original product image generated in the manufacturing stage; and
the inference result includes defect information of the original product image.

3. The inference computing apparatus according to claim 2 , wherein after obtaining the inference result, the inference computing apparatus further performs following operations:
generating a decision instruction according to the inference result; wherein
the defect information of the original product image includes: the identified original product image having a defect, and defect location and defect type; and the decision instruction includes: performing a corresponding defect processing on a product corresponding to the original product image having the defect according to the defect information of the original product image.

4. The inference computing apparatus according to claim 1 , wherein the update condition further includes that:
configuration information of the second inference model is verified, and the configuration information matches with the inference computing to be performed.

5. The inference computing apparatus according to claim 1 , wherein evaluation parameters used for evaluating the performance of the first inference model include at least one of accuracy rate, precision rate, recall rate and F Score of the first inference model during an inference computing process;
a case where the first inference model needs to be updated according to the performance evaluation includes: the performance of the first inference model fluctuates or decreases, wherein whether the performance of the first inference model fluctuates or decreases is determined according to a variation of the evaluation parameters in a continuous period of time.

6. The inference computing apparatus according to claim 1 , wherein updating the first inference model according to the performance evaluation includes:
receiving a third inference model, the third inference model being obtained by the model training apparatus through a third model training based on an updated first training sample library in a case where the model training apparatus has not received the model update request, and the updated first training sample library including the third training samples being from the inference result and subjected to re-judging;
comparing the performance of the first inference model with a performance of the third inference model, and when the performance of the third inference model is better than the performance of the first inference model, or when an abnormality occurs in the first inference model, updating the first inference model with the third inference model.

7. The inference computing apparatus according to claim 1 , wherein in a case where the first inference model is updated with the second inference model, the inference computing apparatus further performs following operations:
performing the performance evaluation on the second inference model, and comparing the performance of the first inference model with the performance of the second inference model;
when the performance of the second inference model is worse than the performance of the first inference model, rolling back the second inference model to the first inference model; and performing the second model training again, or requesting the model training apparatus to perform the second model training again.

8. A model training apparatus, comprising at least one processor and a memory, wherein the memory has stored therein program instructions that, when executed by the at least one processor, cause the model training apparatus to perform following operations:
performing a first model training based on a first training sample library to obtain a first inference model, the first training sample librar

Assignees

• Boe Technology Group Co Ltd

Inventors

• Li Zhaoyue
• Chai Dong
• Lu Yuanyuan
• Wang Hong

Classifications

• G06N3/0475

  Generative networks · CPC title

• G06N3/0464

  Convolutional networks [CNN, ConvNet] · CPC title

• G06N3/09

  Supervised learning · CPC title

• G06N3/08

  Learning methods · CPC title

• G06T7/0004 Primary

  Industrial image inspection · CPC title