Adversarial sample generation method and device and computer equipment
By perturbing the verbs determined according to the business scenario of the deep learning model, the generated adversarial examples can better meet the business objectives of the model, solving the problem of insufficient effectiveness of adversarial examples in existing technologies and achieving efficient generation of adversarial examples.
CN115309854BActive Publication Date: 2026-06-26TENCENT TECHNOLOGY (SHENZHEN) CO LTD
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- TENCENT TECHNOLOGY (SHENZHEN) CO LTD
- Filing Date
- 2021-05-07
- Publication Date
- 2026-06-26
AI Technical Summary
Technical Problem
Existing technologies lack sufficient effectiveness in generating adversarial examples, making it difficult to accurately reflect the anti-interference capabilities of deep learning models.
Method used
By obtaining the original samples of the target model, determining the verbs based on its business scenario, and performing perturbation processing to generate adversarial examples, the perturbation processing is ensured to conform to the business objectives of the target model.
Benefits of technology
The generated adversarial examples are better suited to the business scenarios of the target model, thus improving the effectiveness and realism of the adversarial examples.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN115309854B_ABST
Abstract
The application relates to artificial intelligence and provides an adversarial sample generation method, device and computer equipment. The method comprises the following steps: obtaining an original sample of a target model; determining a movable verb in the original sample according to a business scenario of the target model; performing disturbance processing on the original sample based on the movable verb to obtain an adversarial sample of the original sample, and the adversarial sample is used for detecting the target model. The method improves the effectiveness of the adversarial sample.
Need to check novelty before this filing date? Find Prior Art