| 研究実績の概要 |
Although Deep Learning algorithms are very powerful, their application in biomedical and cancer research is often limited by their requirement for very large number of samples. This project proposes a meta-learning-based strategy to overcome this limitation, which allows other related datasets to be used in training the model that can then be tailored for the specific cases of interest where fewer samples may be available. The work lead to the successful development of a novel deep meta-learning architecture for survival analysis of censored time-to-event data that can achieve superior levels of performance on high-dimensional medical multiomics datasets commonly used in cancer research.
|
