fairsafepriv : Fairness, Privacy & Safety # Bias # MinDiff # Tensorflow # Keras # Gemini-API
- Learn equal opportunity fairness concept.
- Apply MinDiff method to a TF model to mitigate bias.
- Wrap existing optimizers (e.g., SGD, Adam) into differentially private counterparts using TF Privacy
- Understand hyperparameters introduced by differentially private ML
- Measure privacy guarantee provided using analysis tools included in TensorFlow Privacy
- Call Gemini via Gen AI SDK & inspect responses safety ratings
- Define threshold for filtering safety ratings according to needs
