US 11,934,520 B2
Detecting data anomalies on a data interface using machine learning
Gorkem Batmaz, Santa Clara, CA (US); Nicola DiMiscio, Santa Clara, CA (US); Mark Overby, Snohomish, WA (US); and Ildiko Pete, Santa Clara, CA (US)
Assigned to NVIDIA Corporation, Santa Clara, CA (US)
Filed by Nvidia Corporation, Santa Clara, CA (US)
Filed on Mar. 28, 2019, as Appl. No. 16/368,589.
Claims priority of provisional application 62/649,531, filed on Mar. 28, 2018.
Prior Publication US 2019/0303567 A1, Oct. 3, 2019
Int. Cl. G06N 3/08 (2023.01); G06F 11/30 (2006.01); G06F 11/34 (2006.01); G06F 21/55 (2013.01); G06F 21/85 (2013.01); G06N 3/044 (2023.01); G06N 3/045 (2023.01); G06N 3/088 (2023.01); G06N 5/045 (2023.01); G06N 7/01 (2023.01); H04L 9/40 (2022.01); H04L 12/40 (2006.01); H04L 67/12 (2022.01)
CPC G06F 21/552 (2013.01) [G06F 11/3027 (2013.01); G06F 11/349 (2013.01); G06F 21/554 (2013.01); G06F 21/556 (2013.01); G06F 21/85 (2013.01); G06N 3/044 (2023.01); G06N 3/045 (2023.01); G06N 3/08 (2013.01); G06N 3/088 (2013.01); G06N 5/045 (2013.01); G06N 7/01 (2023.01); H04L 12/40 (2013.01); H04L 12/40045 (2013.01); H04L 63/1416 (2013.01); H04L 63/1441 (2013.01); G06F 2221/034 (2013.01); H04L 67/12 (2013.01)] 20 Claims
OG exemplary drawing
 
1. A method for detecting anomalous data communicated over a data interface, the method comprising:
applying a first data package of a first data type received from a data interface as input to a first neural network;
generating, by the first neural network and based on the first data package, predicted data values for a subsequent data package of the first data type received from the data interface;
receiving, over the data interface, a subsequent data package of the first data type comprising real data values;
determining, based on a difference between the predicted data values and the real data values, a first deviation value for the first data type by comparing the real data values with the predicted data values;
applying the first deviation value with other deviation values corresponding to other data types as input to a second neural network; and
calculating, by the second neural network, a probability of an attack on the data interface by comparing the first deviation value to the other deviation values.