Rice growth simulation and plant resilience

During our fourth meet up we discussed two interesting projects presented by Jingye Han and Ben Noordijk.

Jingye Han presented a novel approach to yield prediction (DeepOryza). The approach consists in using a crop growth metamodel pretrained on synthetic data and in fine-tuning its weights on a limited set of observations. Preliminary results indicate that pre-training the weights of a LSTM neural network on synthetic data can help reaching better predictions compared to a non-pretrained network or a crop growth model. The presented use case focused on rice.

Jingye’s presentation can be found here.

Ben Noordijk presented his work on integrating gene regulatory network with ODE neural network. The work is aimed at modelling genetic control of plant resilience traits for breeding purposes. The approach is particularly novel as it aims at clustering the genes of the gene regulatory network in order to simplify the model and make it identifiable.

Ben’s presentation can be found here.