| 研究実績の概要 |
In this research we further developed the novel machine learning paradigm based solely on dynamical equations and no optimization. I am extremely happy to mention that all works submitted in the first year of research were accepted in top journals of the area. This further justifies how promising this new paradigm is for the next generation of AI. SyncMap was further improved in stability (Symmetrical SyncMap) which was published in Physica D: Nonlinear Phenomena.
To deal with high dimensionality, a variation of SyncMap called Magnum was published in Neurocomputing. With excellent results we also went beyond the proposal and investigated further the paradigm in many different ways:
1 - We considered the addition of Reservoir Computing (RC) and how it could be used inside the SyncMap paradigm. Our review paper on the subject got accepted by IEEE Access and we have some proof of concept of a new method working with RC
2 - We developed a concept of a modified SyncMap that could directly deal with continuous variables without losing any of its adaptation and robust properties. Moreover, the new method can take into account continuous probabilities and other information naturally.
3 - We developed a foundation for image recognition and classification with SyncMap. Very promising first results shows that image recognition with SyncMap is robust and not data intensive (low data). More time is required to further the experiments.
Based on the above, the research results were not only great but also ramified into multiple new promising research areas.
|
