Incremental Learning

Incremental Learning with Adaptive Resonance Theory (ART) & Developmental Resonance networks

1. Online Incremental Learning

Online incremental learning without forgetting aims to analyze sequentially incoming data. Online incremental learning allows robots to deal with dynamic environment information. We propose s-DRN which can cluster sequential data with the following featuers:

• Online incremental: s-DRN processes input data and generates clusters dynamically.
• Computationally efficient: s-DRN requires O(n) computation.
• Robust to hyper-parameter setting: the performance of s-DRN is hardly affected by the internal hyper-parameters such as vigilance parameters.

For detailed formulation of s-DRN, please refer to the sDRN directory of our repository and the following paper:

@article{yoon2019stabilized,
  title={Stabilized Developmental Resonance Network},
  author={Inug Yoon*, Uehwan Kim* and Jong-Hwan Kim},
  journal={IEEE Transactions on Neural Networks and Learning Systems, Under Review},
  year={2019}
}

2. Stabilized-Feedback Episodic Memory (SF-EM)

Episodic memory incrementally learns user behaviors and event sequences. However, conventional episodic memory fails to stably perform over a long period of time. In addition, they cannot not accept user feedback. The proposed SF-EM stably performs over a long period of time and accepts user feedback. The following are the key features of SF-EM:

• An adaptive decay factor to enhance the stability of the learning process of the memory architecture.
• A feedback mechanism to reflect user feedback.
• A home service provision framework for robot and IoT collaboration.

For detailed formulation of SF-EM, please refer to the SFEM directory of our repository and the following paper:

@article{kim2018a,
  title={A Stabilized Feedback Episodic Memory (SF-EM) and Home Service Provision Framework for Robot and IoT Collaboration},
  author={Uehwan Kim and Jong-Hwan Kim},
  journal={IEEE Transactions on Cybernetics, Early Access},
  year={2018}
}

References

This repository contains implementation of following works as components for the proposed s-DRN and SF-EM:

• G. A. Carpenter, S. Grossberg, and D. B. Rosen, “Fuzzy ART: An adaptive resonance algorithm for rapid, stable classification of analog patterns,” in Proc. Int. Joint Conf. Neural Netw., vol. 2, 1991, pp. 411–416.
• W. Wang, B. Subagdja, A.-H. Tan, and J. A. Starzyk, “Neural modeling of episodic memory: Encoding, retrieval, and forgetting,” IEEE Trans. Neural Netw. Learn. Syst., vol. 23, no. 10, pp. 1574–1586, Oct. 2012.
• G.-M. Park, Y.-H. Yoo, D.-H. Kim, and J.-H. Kim, “Deep ART neural model for biologically inspired episodic memory and its application to task performance of robots,” IEEE Trans. Cybern., vol. 48, no. 6, pp. 1786–1799, Jun. 2018.

Acknowledgments

This work was supported by Institute for Information & communications Technology Promotion(IITP) grant funded by the Korea government(MSIT) (No.2016-0-00563, Research on Adaptive Machine Learning Technology Development for Intelligent Autonomous Digital Companion)