Speaker
Description
The advances in computing, coupled with the easy availability of massive datasets, have led to the widespread adoption of machine learning (ML) in general and deep learning (DL) in particular. In essence, the traditional DL models are fundamentally data-driven. That is, these models tend to have an improved performance when the input data is increased. Therefore, it is challenging to deploy these models where data is scarce. Furthermore, since these traditional DL models are data-driven, they tend to lack the reality check. Currently, some approaches have been proposed to address these limitations of traditional DL models. Such approaches include physics-informed neural networks (PINN) and biology-informed neural networks (BINN). In this paper, we propose a new approach to address the traditional DL limitations. This approach is referred to as statistics-informed neural networks (SINN). In particular, our approach focuses on the statistics of rare events and hence is referred to as extreme value-informed neural networks (EVINN). Essentially, EVINN fuses DL and the extreme value theory together, with the extreme value theory serving as a reality check for the DL. Furthermore, to demonstrate the utility of the proposed EVINN model, we show how the model can be used for drought prediction. Finally, the results obtained from the study reported in this paper demonstrate the significance of fusing together the traditional DL and the extreme value theory for a more realistic modeling of rare events.
| Apply for student award at which level: | None |
|---|---|
| Consent on use of personal information: Abstract Submission | Yes, I ACCEPT |