A method and system for black-box fairness testing of machine learning models based on deep reinforcement learning
A machine learning model and reinforcement learning technology, applied in neural learning methods, software testing/debugging, biological neural network models, etc., can solve problems such as high overhead and low efficiency, improve discovery capabilities and efficiency, reduce testing costs, The effect of good scalability
Active Publication Date: 2022-07-22
INST OF SOFTWARE - CHINESE ACAD OF SCI
View PDF2 Cites 0 Cited by
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
[0005] In view of the low efficiency and high cost of the existing black-box fairness test, and the white-box fairness test method cannot be used in the black-box scenario, the present invention proposes a black-box fairness test of a machine learning model based on deep reinforcement learning method and system
Method used
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View moreImage
Smart Image Click on the blue labels to locate them in the text.
Smart ImageViewing Examples
Examples
Experimental program
Comparison scheme
Effect test
Embodiment approach
[0069] The present invention does not limit the type of input action to the test environment and the construction method of test data (eg combined action, ie, synchronizing or simultaneously replacing multiple non-protected feature values of the current state, etc.).
[0070] The present invention does not limit the reward calculation method of the reward calculation sub-module for actions or states (such as the reward value r 1 , r 2 , r 3 dynamically adjusted as the test iterations progress, etc.).
[0071] The present invention does not limit the strategy learning model and update method (such as other reinforcement learning models and learning algorithms) adopted by the agent.
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More PUM
Login to View More Abstract
The invention discloses a machine learning model black-box fairness testing method and system based on deep reinforcement learning. The main process includes: (1) building a machine learning model black-box fairness testing environment; The optimal discrimination instance generation strategy learning and (3) result statistics are three parts. First, the machine learning model black-box fairness test environment is constructed, and then the reinforcement learning agent interacts with the constructed machine learning model black-box fairness test environment. Using deep reinforcement learning algorithm to learn the optimal discrimination instance generation strategy, and finally statistical test results. The invention can solve the problems of no effective heuristic strategy guidance, low test efficiency and high test overhead in the field of machine learning model black box fairness testing.
Description
technical field [0001] The invention relates to a machine learning model black-box fairness testing method and system based on deep reinforcement learning, and relates to the technical fields of software engineering and artificial intelligence. Background technique [0002] Machine learning software is widely used in various decision-making fields in human real life, such as recruitment, insurance, policy prediction, etc. Researchers have found that machine learning software will produce various unfair decision-making behaviors in the actual operation process, resulting in bad social impact. Therefore, from the perspective of software engineering, designing efficient fairness testing algorithms, conducting adequate fairness testing before machine learning software is delivered, and finding as many instances of discrimination in machine learning software as possible have become an urgent problem to be solved. Here, the goal of fairness testing is to find as many instances of...
Claims
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More Application Information
Patent Timeline
Login to View More Patent Type & Authority Patents(China)
IPC IPC(8): G06F11/36G06N3/08
CPCG06F11/3684G06F11/3688G06N3/08
Inventor 谢文涛吴鹏
Owner INST OF SOFTWARE - CHINESE ACAD OF SCI