An online learning method and device with adaptive weights, electronic equipment and storage medium
By employing an online learning method with adaptive weights, a weight parameter space is constructed and adjusted using real-time energy loss. This enables zero-sample startup and continuous optimization, addressing the performance limitations of traditional machine learning methods in the absence of historical data and improving computational efficiency and adaptability.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- 深圳市积微科技有限公司
- Filing Date
- 2026-03-23
- Publication Date
- 2026-06-19
AI Technical Summary
Traditional machine learning methods struggle to adapt quickly to dynamically changing business needs in real-world scenarios. In particular, they fail to effectively launch or achieve expected performance when there is a lack of sufficient historical data. Acquiring labeled data is costly, environmental interaction is inefficient, and objective function evaluation is expensive.
An online learning method with adaptive weights is adopted. By constructing a weight parameter space, the weight parameters are dynamically optimized by adjusting the initial sampling probability of the uniform distribution of candidate weight vectors and real-time energy loss, so as to achieve zero-sample start-up and continuous optimization.
It can quickly start and iteratively optimize weight parameters in scenarios lacking historical data, reduce the preparation cost before model deployment, support cold start scenarios, solve the problem of traditional methods being unable to start or having insufficient performance due to insufficient data in the initial stage, and has high computational efficiency.
Smart Images
Figure CN122242608A_ABST