ML in-security: invisible Unicode characters

podcast: Imperceptible NLP Attacks – The Data Exchange

fascinating new paper – “Bad Characters: Imperceptible NLP Attacks” – which provides a taxonomy of attacks against text-based NLP models, that are based on Unicode and other encoding systems. We discussed the key findings in their paper, and we also briefly talked about the state of homomorphic encryption for machine learning and analytics.

4 types of attacks: Invisible characters, homoglyphs, reorderings, and deletions