Text this: Applying Reinforcement Learning to Protect Deep Neural Networks from Soft Errors