2017 Fiscal Year Annual Research Report
Evolutionary Approaches to Learning Self-awareness for a Decentralized System
| Project/Area Number |
15K00343
|
|---|
| Research Institution | The University of Aizu |
|---|
Principal Investigator |
劉 勇 会津大学, コンピュータ理工学部, 上級准教授 (60325967)
|
|---|
| Project Period (FY) |
2015-04-01 – 2018-03-31
|
| Keywords | Neural network learning / Machine learning / Statistical learning / Committee machine / Machine intelligence |
|---|
| Outline of Annual Research Achievements |
A decentralized awareness system with a set of self-aware neural network subsystems has been developed in this project. Self-awareness is a kind of ability of recognizing oneself as an individual being different from the environment and other individuals. Awareness neural networks of being aware of their states, behavior and performance can better acquire the right representations of the learned data, and distribute context information through the whole decentralized system.
Two levels of self-awareness have been created in the decentralized system, including private self-awareness for individual neural networks and public self-awareness for the decentralized system. Private self-awareness is trained by negative correlation learning with difference learning and opposition learning. Difference learning can let each individual to adapt its learning directions, and be aware of what others have learned. Public self-awareness at the ensemble level is created by the bounded negative correlation learning. Negative correlation learning emphasizes interaction and cooperation among individual neural networks in a decentralized system, and uses an unsupervised penalty term in learning functions to produce negatively correlated individual neural networks.
Experimental results have shown that awareness neural networks by negative correlation learning could better meet their requirements for predictions in the applications on both medical and financial data. The research results have been published in the international journals, and presented in a number of international conferences.
|
