US 12,333,789 B2
Using guard feedback to train AI models
David Lee Selinger, Pleasanton, CA (US)
Assigned to DEEP SENTINEL CORP., Pleasanton, CA (US)
Filed by Deep Sentinel Corp., Pleasanton, CA (US)
Filed on Mar. 27, 2022, as Appl. No. 17/705,346.
Prior Publication US 2023/0316726 A1, Oct. 5, 2023
Int. Cl. G06V 10/774 (2022.01); G06V 10/40 (2022.01); G06V 10/778 (2022.01)
CPC G06V 10/7747 (2022.01) [G06V 10/40 (2022.01); G06V 10/7792 (2022.01)] 42 Claims
OG exemplary drawing
 
1. A system for training an AI model, the system comprising:
a processor;
at least one video camera in communication with the processor;
a memory that stores computer readable instructions that, upon execution by the processor, cause the processor to:
divide a recorded video into video frames;
make a first prediction about an event based on the first set of latent characteristics, and existing conditions surrounding the event;
identify a second set of latent characteristics in the video frames, the second set of latent characteristics being different from the first set of latent characteristics;
make a second prediction about the event based on the second set of latent characteristics and the existing conditions surrounding the event;
annotate one or more of the frames based on one of the first set of latent characteristics and the first prediction or the second set of latent characteristics and the second prediction;
observe and track a user response of a first user to the one or more annotated frames;
score how useful the first set of latent characteristics is in achieving a preferred user response based on the user response and the first prediction;
score how useful the second set of latent characteristics is in achieving the preferred user response based on the user response and the second prediction;
perform an adjustment to the first set of latent characteristics or the second set of latent characteristics based on one or more of the user response, a score of the first set of latent characteristics, or a score of the second set of latent characteristics;
wherein semi-supervised learning or self-supervised learning is used to score the first set of latent characteristics and the second set of latent characteristics.