A few days back Tesla announced a new patent filing under the title “System and Method for Adapting a Neural Network Model on a Hardware Platform” in which they have proposed a new type of neural network model. The method in the patent helps in selecting the perfect neural network by looking at the system, the aim is to fulfill all the constraints that are present for a particular system.
The main reason behind the development of this new method was the restrictions that Tesla was facing in the self-driving project on which they have been working extensively for a very long time.
The work got the apt impetus after Tesla acquired a startup Deepscale that works in the field of artificial intelligence for developing innovative components for self-driving vehicles and also neural networks for various kinds of devices.
How Tesla’s Neural Network Method is Different
The patent discusses the fundamental rule using which the neural network operates. The idea is to create a list of configurations that are valid and a threshold or a condition that will satisfy all the constraints for a specific platform mainly hardware.
Neural Networks have revolutionized the field of Machine Learning with their unique ability to perform a particular task with exceptional accuracy and speed that hasn’t be parallel by human beings. One of the main difficulties with neural networks is that it takes time to configure them for a particular system or platform.
Tesla has been working day in day out to build the world’s first fully self-driving car and its engineers have had a tough time in the intermingling of the hardware components with the software components. According to Tesla, there is a long and tedious process that has to be followed so that neural networks can function in compliance with the hardware on which they’ll be working. To achieve this successfully all the decision points are reviewed and configured so that desired results can be achieved.
Improving the efficiency
Apart from the hardware aspect, in their patent, Tesla has attempted to improve the neural network’s efficiency by enhancing the adaptability attribute of the same.
This novel development was initiated and successfully completed under the guidance of Dr. Michael Driscoll – who has taken up the responsibility of Senior Software Engineer at Tesla really well, after showcasing his mastery at DeepScale.
In simple terms, once a neural network model is fed along with the hardware’s information into this software system, all the pivotal decision points will be highlighted whose care has to be taken. Even the hyperparameters are chosen on the basis of the results obtained from here. This system will also enable a user-friendly interface that’ll facilitate in understanding the process and also the path that should be adhered to for reaching the results.
This Neural Network is another long stride that Tesla has made towards its dream of building the world’s first self-driving car. After numerous attempts at this monumental task, Tesla has been finding the points which are causing a delay in making the dream of self-driving a reality. Through such novel developments, Tesla looks to build a robust and flexible system that can help them build this unthinkable vehicle.