Tag
This paper presents the first systematic study of injecting biokinetic ODE knowledge into neural networks for bioprocess modeling, comparing simulation pre-training and architecture-level priors, finding they are substitutable and effective under data scarcity.