Upgrade core software to use more modern library versions by utilizing a generic model inference framework
Modernizing Legacy ML
Upgrading core software without breaking old models.
What is the item?
Why is it helpful to our customers?
This allows us to continue to deliver excellent efficacy without compromising the time it takes to respond to attacks.
Why is it interesting?
Models that were trained long ago are useful because they represent a snapshot of the attack landscape at that time, but they make upgrading libraries nearly impossible since retraining is difficult or impossible.
Using ONNX allowed us to use these old models without blocking our library upgrade.