[MeshGraphNets] Question about training strategy with new trajectory data

**Description:**
I'm working with MeshGraphNets and have a question about the best approach for training when new trajectory data becomes available.

**Context:**
- I have already trained a MeshGraphNet model on 10 trajectories
- I now have 5 new trajectories from the same CFD case
- The goal is to improve model generalization with this additional data

**Specific Questions:**
1. What is the recommended approach for incorporating new trajectory data:
a) Continue training the existing model only on new trajectories
b) Continue training the existing model on old + new trajectories
c) Reset parameters and train from scratch on combined data

2. What are the implications of each approach in terms of:
- Model performance and generalization
- Training efficiency
- Risk of catastrophic forgetting
- Optimal parameter learning

I appreciate your response to my query.
Thank you.