Adversarial Attacks on ML Models

Concept

Defending machine learning models from attacks and manipulation.

Longer Description

With the continued rise of machine learning in production level environments, a deeper understanding and prevention of adversarial attacks on these models is needed.

A vast body of work has been done in research communities to tackle this, largely related to identification obfuscation and manipulation of intent (think removal of signage that helps guide AVs or tricking human tracking models).

Other Thoughts

• It’s possible that this becomes a core competency of existing cybersecurity companies however we believe this could necessitate its own company.
• Most of the GTM for the past few years has been focused on fraud detection in financial settings that utilize models for underwriting/detection. We believe this is a large opportunity but are less interested in it/likely not the best investors to back a company with this core focus.
• That said other surface areas related to automated natural language interactions or image verification for financial gain are compelling.