On request, models can be trained using Modzy’s patented adversarial defense solution to ensure your models are robust against attacks. Further, Modzy’s novel adversarial input detector can parse data for potential adversarial attacks and filter out adversarial inputs before they get to the model.
All model images managed by Modzy are immutable: changes and updates will generate new versions, allowing full reproducibility of results during audits or debugging. Further, Modzy's proprietary Model watermarking can validate provenance information for models running in production.
Updated 2 months ago