Use async for data computation

[Firenze11]

Current implementation involves converting tensor data between array and tf.Tensor format through synchronized data reading, which blocks the UI thread until the values are ready, which can cause performance issues. Migrate to async instead.