17544077. ADJUSTING MACHINE LEARNING MODELS BASED ON SIMULATED FAIRNESS IMPACT simplified abstract (INTERNATIONAL BUSINESS MACHINES CORPORATION)
Contents
ADJUSTING MACHINE LEARNING MODELS BASED ON SIMULATED FAIRNESS IMPACT
Organization Name
INTERNATIONAL BUSINESS MACHINES CORPORATION
Inventor(s)
Pranay Kumar Lohia of Bhagalpur (IN)
Kushal Mukherjee of New Delhi (IN)
Rakesh Rameshrao Pimplikar of Bangalore (IN)
Sameep Mehta of Bangalore (IN)
Stacy F. Hobson of Hopewell Junction NY (US)
ADJUSTING MACHINE LEARNING MODELS BASED ON SIMULATED FAIRNESS IMPACT - A simplified explanation of the abstract
This abstract first appeared for US patent application 17544077 titled 'ADJUSTING MACHINE LEARNING MODELS BASED ON SIMULATED FAIRNESS IMPACT
Simplified Explanation
The patent application describes methods, systems, and computer program products for adjusting machine learning models based on simulated fairness impact. Here is a simplified explanation of the abstract:
- The invention involves a central simulation system that obtains policies to be used for a simulation involving machine learning models interacting with a target population.
- Simulators representing the machine learning models of different systems are configured using information provided by the central simulation system.
- The simulation is performed in iterations, where the central simulation system predicts the state of the target population, provides the state to the simulators, and collects metrics based on the simulator results.
- Based on the collected metrics, the central simulation system selects and sends one of the policies to at least one of the different systems.
Potential applications of this technology:
- Fairness assessment: The invention can be used to assess the fairness impact of machine learning models on a target population, allowing for adjustments to be made to ensure fairness.
- Model optimization: By simulating the impact of different policies, the invention can help optimize machine learning models to achieve desired outcomes while minimizing potential biases.
- Policy evaluation: The central simulation system can evaluate the effectiveness of different policies in achieving specific goals, such as reducing bias or improving accuracy.
Problems solved by this technology:
- Bias detection and mitigation: The invention addresses the problem of potential biases in machine learning models by simulating their impact and allowing for adjustments to be made to ensure fairness.
- Policy selection: The central simulation system helps in selecting the most effective policy by collecting metrics based on the simulation results, enabling informed decision-making.
- Transparency and accountability: By providing a systematic approach to evaluating machine learning models and their impact, the invention promotes transparency and accountability in the use of AI systems.
Benefits of this technology:
- Fairness and equity: The invention helps in identifying and addressing potential biases in machine learning models, promoting fairness and equity in decision-making processes.
- Optimization and efficiency: By simulating different policies, the invention allows for the optimization of machine learning models, leading to improved accuracy and efficiency.
- Decision support: The collected metrics and simulation results provide valuable insights for decision-makers, enabling them to make informed choices regarding machine learning models and policies.
Original Abstract Submitted
Methods, systems, and computer program products for adjusting machine learning models based on simulated fairness impact are provided herein. A computer-implemented method includes obtaining, by a central simulation system, policies to be used for performing a simulation involving machine learning models, implemented on different systems, interacting with a target population; providing information for configuring simulators on the different systems, each simulator representing at least the machine learning model of a given one of the different systems; performing iterations of the simulation for the policies, wherein, for each iteration, the central simulation system: predicts a state of the target population, provides the state to the simulators, and collects metrics based on results of the simulators; and selecting and sending one of the policies to at least one of the different systems based on the collected metrics.
