Abstract
Disclosed herein is a method for an event-based post transformation framework for multitask learning outputs in deep neural networks. The method discloses categorization of event types based on specific pre-defined business requirements. Each task associated with multitask learning is then treated as a separate event and grouped or separated according to their correlation with predefined event types. Next, the method calculates event confidence scores based on the probability of raw output scores from each task. The method then determines scaled score cut-offs, which are predetermined based on business needs and in next step utilizes a spline model to transform probability of event confidence raw scores from the range 0 to 1 to a transformed score ranging from 1 to 99. The method in last step transforms the scaled two-digit transformed scores into three-digit scores, where the first digit of three digit score signifies the event type.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Kan, Xi; Wang, Sheng; Mori, Michael; and Wang, Dan, "EVENT BASED POST TRANSFORMATION FRAMEWORK FOR MULTI-TASK LEARNING OUTPUTS", Technical Disclosure Commons, (February 02, 2024)
https://www.tdcommons.org/dpubs_series/6657