Diffusion-based training optimization device and method for smart vision inspection system
The diffusion-based learning optimization method addresses the limitations of SGD by converting stochastic optimization into a sampling problem with global convergence, achieving high accuracy and efficiency in training AI models for smart vision inspection systems.
Patent Information
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- IMPIX
- Filing Date
- 2024-12-30
- Publication Date
- 2026-07-02
AI Technical Summary
Existing stochastic gradient descent (SGD) methods for training artificial intelligence models face limitations such as converging to local solutions and strong assumptions, making them unsuitable for complex deep learning tasks, despite their widespread use in fields like smart vision inspection systems.
A diffusion-based learning optimization method that transforms stochastic optimization into a sampling problem using the invariant measure of the Langevin SDE, applying a sampling algorithm with proven global convergence to estimate parameters, ensuring high accuracy and efficiency in training AI models for defect detection and anomaly recognition.
Ensures global convergence and high accuracy in training AI models for smart vision inspection systems, even with complex objective functions, by concentrating samples at the global minimum and adapting step size based on the objective function's slope.
Smart Images
Figure KR2024021524_02072026_PF_FP_ABST