Self-Organizing Adaptive Recurrent Neural Network for Lifelong Learning
Abstract
Life-long learning can learn consistently over a long period of time by updating new knowledge while retraining from previous learning experiences. Imitation is the capacity to understand the behaviors of others and reproduce them. Imitation learning is a way to learn and acquire new skills through another agent's observation of these skills. The First achievement through imitation is other’s actions by vision using different development theories. This paper presents a developmental model of imitation learning based on the hypothesis that humanoid robot acquires imitative abilities as induced by sensorimotor associative learning through self-exploration. Our proposed method “SOARIN” network that connected hierarchically. The efficient and effective method SOARIN consists of three stages namely neuron activation, neuron matching, neuron learning. We learn to incorporate and update upper body behavior according to the joint angles visualized by the robot sensor by using all these steps. The temporal connections represent the series of neurons activated during the learning process. Through vision, sensors understand and learn action automatically and store data in episodic memory.
Keywords: Incremental learning, Episodic memory, Robot upper body actions learning.
