	US 12,235,999 B2
	System and method for remediating poisoned training data used to train artificial intelligence models
Ofir Ezrielev, Be'er Sheva (IL); Amihai Savir, Newton, MA (US); and Tomer Kushnir, Omer (IL)
Assigned to Dell Products L.P., Round Rock, TX (US)
Filed by Dell Products L.P., Round Rock, TX (US)
Filed on Dec. 29, 2022, as Appl. No. 18/147,752.
Prior Publication US 2024/0220661 A1, Jul. 4, 2024
Int. Cl. G06F 21/64 (2013.01); G06F 21/57 (2013.01)

CPC G06F 21/64 (2013.01) [G06F 21/57 (2013.01)]

20 Claims

1. A method for managing an artificial intelligence (AI) model, the method comprising:

obtaining a snapshot of a first poisoned instance of the AI model, the first poisoned instance of the AI model being obtained using, at least in part, first poisoned training data;

remediating an impact of the first poisoned training data on the first poisoned instance of the AI model to obtain a first new AI model;

obtaining a snapshot of the first new AI model;

obtaining a snapshot of a second poisoned instance of the AI model, the second poisoned instance of the AI model being obtained through further training of the first poisoned instance of the AI model;

obtaining first weights associated with the new AI model using the snapshot of the new AI model;

obtaining second weights associated with the second poisoned instance of the AI model using the snapshot of the second poisoned instance of the AI model;

obtaining a first difference using the first weights and the second weights; and

obtaining a second new AI model using the first weights and the first difference.