Hot off the press:
This is Nvidia's recently published NN patent:
WO2024183052A1 FEDERATED LEARNING TECHNIQUE 20230309; pub 20240912
= US2024303504A1 FEDERATED LEARNING TECHNIQUE
View attachment 69635
View attachment 69634 FIG 9C
View attachment 69638
Apparatuses, systems, and techniques to train/use one or more neural networks. In at least one embodiment, a processor comprises one or more circuits to cause neural network training information to be aggregated based, at least in part, on contribution of the neural network training data and one or more performance metrics of the neural network.
It's a massive document 263 pages including 56 pages of drawings.
It's about federated learning using cloud-based processors to download, eg, NNs and maps to cars (Fig 9D). Fig 9C shows the vehicle's sensor/processor configuration. sensor fusion would be part of this.
View attachment 69639
This abstract talks about broadcasting road changes over the internet.
Interesting thing is that, for ordinary CNN NNs, the whole map model needs to be changed. For Akida, only the change needs to be updated.