SPrekshitha/Adversarial_Attack_Detection_Framework ? reverse-engineered prompt

Build me a Python project that detects adversarially manipulated inputs before they reach a machine learning model.

I want the first part to set up the environment and train a simple baseline CNN, then add common attack examples like FGSM, PGD, DeepFool, and Carlini Wagner so I can test the detector against them. After that, create the feature extraction and behavior profiling pieces, including useful engineered features and gradient based signals, and store those features in a reusable way so they can feed into detection models later.

Keep it organized in a way that feels like a week by week learning project, and make sure the code is easy to run, read, and extend. Use Python with TensorFlow, Scikit Learn, ART, and Foolbox if they help. If anything needs current setup details or best practices, look up the latest docs online and keep the implementation practical.