Skip to content

EY23.4.1.B: Implement FOQUS support for generating gradient data required for GENN models.

Past due by about 1 year 100% complete

Personnel: Brady Gess (intern, main developer), Brandon Paul (mentor), Joshua Morgan (mentor), Miguel Zamarripa (supervisor)

Description: Gradient-enhanced neural network (GENN) models use derivatives to inform graph training and are more accurate for multi-input, multi-output problems. Training GENN models requires knowing the derivatives a priori; for a…

Personnel: Brady Gess (intern, main developer), Brandon Paul (mentor), Joshua Morgan (mentor), Miguel Zamarripa (supervisor)

Description: Gradient-enhanced neural network (GENN) models use derivatives to inform graph training and are more accurate for multi-input, multi-output problems. Training GENN models requires knowing the derivatives a priori; for a sampled or simulation-produced dataset, these derivatives must be calculated. This project aims to develop a CCSI2 tool to calculates gradients for an arbitrary input dataset to support loading and training GENN models in FOQUS.

Current status: Tool and methods are in-development, and various implementations are being tested for accuracy against a publicly-available sample problem.

Original due date: 6/30
Modified due date: 7/31 or later (end of the summer, probably mid to late August to give Brady time to complete the work)

There are no open issues in this milestone.

Add issues to milestones to help organize your work for a particular release or project.

Create new issue

Or find and add issues with no milestone in this repo.