NLP Inspired Training Mechanics For Modeling Transient Dynamics
This work addresses the problem of modeling transient dynamics in fluid flows for researchers and engineers, representing an incremental improvement by applying existing NLP techniques to a new domain.
The paper tackled improving the accuracy, robustness, and generalizability of machine learning models for simulating transient dynamics, specifically vortical flows, by introducing NLP-inspired training mechanics like teacher forcing and curriculum learning, resulting in an enhancement in accuracy for models such as FNO and UNet by more than 50%.
In recent years, Machine learning (ML) techniques developed for Natural Language Processing (NLP) have permeated into developing better computer vision algorithms. In this work, we use such NLP-inspired techniques to improve the accuracy, robustness and generalizability of ML models for simulating transient dynamics. We introduce teacher forcing and curriculum learning based training mechanics to model vortical flows and show an enhancement in accuracy for ML models, such as FNO and UNet by more than 50%.