图片分类模型的获取方法和获取装置
By training a neural network model using both clean and noisy training data, adjusting the direction of parameter updates, and combining a Gaussian mixture model and a SAM optimizer, the problem of classification errors under noisy data is solved, improving the robustness and accuracy of the model.
CN116704265BActive Publication Date: 2026-07-17JD DIGITS HAIYI INFORMATION TECHNOLOGY CO LTD
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- JD DIGITS HAIYI INFORMATION TECHNOLOGY CO LTD
- Filing Date
- 2023-07-17
- Publication Date
- 2026-07-17
AI Technical Summary
Technical Problem
Existing fully supervised deep learning models cannot work effectively in complex real-world scenarios when subjected to noisy training data, leading to prediction errors.
Method used
The neural network model is trained using both clean and noisy training data. By adjusting the update direction of the model parameters, the loss is reduced and the robustness of the model is enhanced by using clean data. A Gaussian mixture model or a co-teaching model is used to distinguish between clean and noisy data, and the model parameters are optimized by combining the SAM optimizer.
Benefits of technology
It improves the model's loss smoothing gain and robustness, and enhances classification accuracy on noisy data.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN116704265B_ABST
Abstract
本公开提出一种图片分类模型的获取方法和获取装置，涉及计算机领域，尤其涉及机器学习及应用领域。根据图片训练样本的类别标签是否为噪声标签，将图片训练样本集合划分为干净图片训练样本集合和噪声图片训练样本集合；利用干净图片训练样本集合对神经网络模型的第一模型参数进行更新；利用噪声图片训练样本集合对神经网络模型的第二模型参数进行更新，所述神经网络模型的模型参数为第一模型参数与第二模型参数之和，第一模型参数的更新方向与第二模型参数的更新方向相反；将收敛的神经网络模型确定为图片分类模型。从而，提升损失平滑增益，增强图片分类模型的鲁棒性。
Need to check novelty before this filing date? Find Prior Art