Adaptive Fuzzy Neural Agent for Human and Machine Co-learning

This paper proposes an Adaptive Fuzzy Neural Agent (AFNA) with a Patch Learning Mechanism and IEEE 1855 Fuzzy Markup Language (FML) for human and machine co-learning. There are three phases of patch learning mechanism embedded in AFNA, including (1) training an initial global model, (2) training a patch model for each identified patch, and (3) updating the global model using the training data that do not fall into any patch. The AFNA can be applied to construct the student and robot co-learning regression model, as well as the regression model for the dataset retrieved from the game of Go. First, students generate human learning data through interactions with handheld devices or robots based on the AFNA in Taiwan and Japan. Then, the AFNA utilizes the student learning data collected in the classroom and the Go game data provided by both Google DeepMind and Facebook AI Research open-source OpenGo to train the Fuzzy Machine-Learning Model. In addition, the trained Fuzzy Machine-Learning Model of AFNA is deployed to the robots to make students and machines co-learn together based on IEEE 1855 FML. The experiments show that the AFNA with Patch Learning Mechanism and Fuzzy Machine-Learning Model can improve the performance of regression model based on the datasets of student learning and Go game. In the future, we hope to apply the AFNA with robots to the other domain areas, embed it with the Artificial Intelligence of Things devices, and introduce it to more teaching fields in various countries.

Automated summary

This paper introduces an Adaptive Fuzzy Neural Agent (AFNA) for human and machine co-learning, utilizing a Patch Learning Mechanism and a Fuzzy Machine-Learning Model to construct regression models for students and robots.