What are the tradeoffs of using a specific neural network architecture for time-series data versus an alternative approach?
