Create an app that dynamically generates a memory-optimized network to answer questions based on a given text. The app should read input text and identify relevant entities to store in multiple "memory banks" for efficient retrieval. It should include a controller to manage which banks to access, balancing speed and accuracy. Use a dynamic framework like PyTorch to enable flexible network sizing and variable batching. Test the app's efficiency by comparing its inference speed and accuracy on standard and modified tasks with excessive information.