Automated AI Training Teaches Cars How To Drive
In the future, computers will teach vehicles how to drive, with no human input needed. That’s what Brodmann17’s new automated deep learning training program hopes to achieve.
The tier 2 automotive software supplier, based in Detroit, has developed the automated platform, which trains neural networks for either ADAS driving or fully autonomous situations. While this process can be done manually, it doesn’t scale, and automating the system eliminates the risk of human error, while also decreasing costs.
Additionally, it means gains made in the neural networks get to market quicker, and OEMs can collaborate or partner with AI companies, rather than developing the complex technology themselves. OEMs and suppliers can upload data to Brodmann17’s cloud servers, then run the training process, keeping data and training videos secure, maintaining customer privacy and adhering to strict GDPR laws in Europe.
Neural networks are a vital part of artificial intelligence. Together with machine learning, neural networks are fed large datasets which they can then crunch, on huge scales, to create algorithms. In an autonomous vehicle situation, many neural networks would then be used to constantly test the algorithms to improve the autonomous functions. If the neural network training can be automated, it removes the need for humans – in essence, computers teaching computers.
The system optimizes the process overall, rather than optimizing each individual stage separately, resulting in stronger neural network. It was originally developed as an internal training tool for Brodmann17’s own data, before the company realized it could be opened to select customers.
