Text this: Defending Deep Neural Networks Against Backdoor Attack by Using De-Trigger Autoencoder