View source for 18268664. LEARNING DEVICE, LEARNING METHOD, AND LEARNING PROGRAM simplified abstract (NEC Corporation)

=LEARNING DEVICE, LEARNING METHOD, AND LEARNING PROGRAM=


==Organization Name==

[[:Category:NEC Corporation|NEC Corporation]]


[[Category:NEC Corporation]]

==Inventor(s)==

[[:Category:Riki Eto of Tokyo (JP)|Riki Eto of Tokyo (JP)]][[Category:Riki Eto of Tokyo (JP)]]

==LEARNING DEVICE, LEARNING METHOD, AND LEARNING PROGRAM - A simplified explanation of the abstract==


This abstract first appeared for US patent application 18268664 titled 'LEARNING DEVICE, LEARNING METHOD, AND LEARNING PROGRAM

==Simplified Explanation==

The patent application describes a method that involves using a reward function with specific features and satisfying a Lipschitz continuity condition. This function is used as input for the estimation process, which aims to find a trajectory that minimizes the Wasserstein distance. The Wasserstein distance represents the difference between the probability distribution of an expert's trajectory and the probability distribution of a trajectory determined using the reward function's parameters. The update process then adjusts the parameters of the reward function to maximize the Wasserstein distance based on the estimated trajectory.

* The patent application describes a method for optimizing a reward function.
* The method involves using a reward function with specific features and satisfying a Lipschitz continuity condition.
* The estimation process aims to find a trajectory that minimizes the Wasserstein distance.
* The Wasserstein distance represents the difference between the probability distributions of expert and estimated trajectories.
* The update process adjusts the parameters of the reward function to maximize the Wasserstein distance based on the estimated trajectory.

Potential Applications:
* Reinforcement learning: This method can be applied in reinforcement learning algorithms to optimize reward functions and improve the learning process.
* Robotics: The method can be used in robotics to enhance trajectory planning and control by optimizing reward functions.
* Autonomous vehicles: By optimizing reward functions, this technology can improve the decision-making process of autonomous vehicles.

Problems Solved:
* Reward function optimization: The method solves the problem of efficiently optimizing reward functions to improve learning and decision-making processes.
* Trajectory estimation: By minimizing the Wasserstein distance, the method solves the problem of accurately estimating trajectories based on expert data.

Benefits:
* Improved learning performance: By optimizing reward functions, the method can enhance the learning performance of reinforcement learning algorithms.
* Enhanced decision-making: The technology can improve the decision-making capabilities of autonomous systems by accurately estimating trajectories.
* Efficient optimization: The method provides an efficient way to optimize reward functions, reducing the computational complexity of the process.


==Original Abstract Submitted==

A function input means  accepts input of a reward function whose features are set to satisfy a Lipschitz continuity condition. An estimation means  estimates a trajectory that minimizes Wasserstein distance, which represents distance between probability distribution of a trajectory of an expert and probability distribution of a trajectory determined based on parameters of the reward function. An update means  updates the parameters of the reward function to maximize the Wasserstein distance based on the estimated trajectory.



[[Category:G06N20/00]]