Abstract
The present disclosure discloses a method to construct deep neural network models (102) that can generate complex-valued scores in FinTech. The method employs conformal mapping, which maps points in the upper half-plane onto the unit disc while using the complex-valued number in the unit disc as the deep learning model's score. This gives the score two degrees of freedom, which helps to avoid the score overlapping problem that happens when using a real-valued score ([0, 1] interval). For training deep learning models, the approach also includes a loss function. On the basis of the Loss value, the deep learning model is updated to get the best results.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Yang, Pei; Wu, Peng; and Wang, Dan, "METHOD FOR GENERATING COMPLEX-VALUED SCORE FOR DEEP LEARNING MODELS", Technical Disclosure Commons, (June 12, 2022)
https://www.tdcommons.org/dpubs_series/5189