Tesla Files Patent for Neural Network Hardware Adaptability

 

The world-famous electric vehicle manufacturing company Tesla has recently filed a patent. The patent is about a “Neural Network System with Hardware Adaptability”. It has come about by Tesla’s acquisition of a company called DeepScale, which is an AI startup specialising in Neural Network solutions.

A Neural Network is a type of AI model which is inspired by the widely-accepted biological model of the brain. It is composed of artificial neurons connected to each other with a certain structure. Each neuron is a simple construct and models a simple computation. Multiple neurons connected in a structure forming a neural network can model very complex computations which can be used to extract meaningful information from data. Such capabilities can be harnessed for a wide-variety of applications: such as self-driving cars, the ones Tesla produces. 

Neural Networks are incredibly efficient in finding patterns and providing control outputs once they have been trained and their parameters fixed. However, adapting Neural Networks to specific hardware is time-consuming for developers and this has been a downside. Developers often have to review what they call “decision points” to ensure that the resultant Neural Network is fit for purpose.

 

To optimise the adaptability of Tesla’s neural networks, according to the recent patent filing, they try to train and automate the sets of algorithms to be adaptable. Providing the Neural Network architecture and hardware parameters as inputs, the automation code analyses the neural network and pinpoints where the decision points are. Then these decision points are checked against the hardware parameters to provide adapted Neural Network configurations that are available.

To know more about this advancement please follow this link: https://interestingengineering.com/tesla-files-patent-for-new-neural-network-system-with-hardware-adaptability