Prediction in what sense? Yeah it is definitely a possibility. I think the best use case would be in terms of time series forecasting given the limited data on the pandemic. However, meta-learning has not been widely applied to time series forecasting in general. The idea though would be you take MAML (or some other meta-learning algorithm) and then train it with a base model on other time series data related so it learns how to adapt to limited data forecasting. Then you would try to apply to forecast case numbers etc.