Brodmann17, an Israeli Tier 2 automotive software supplier, has launched what it calls the world’s first automated deep-learning training platform, specifically designed for automotive-grade advanced driver assist systems and autonomous driving applications (ADAS/AD).
The platform seamlessly trains and deploys deep learning neural network models for ADAS/AD solutions, automating a process that cannot be done at scale manually, according to the company.
The platform protects against human error and lowers associated risks, while reducing time to market and costs. Additionally, the platform is said to solve a major industry pain point by enabling automakers and Tier 1 suppliers to collaborate with AI companies in the process of developing neural networks.
“AI doesn’t need to be a black box,” says Adi Pinhas, co-founder and CEO of Brodmann17. “This platform is meeting the growing interest from our customers to be involved in the training process, creating greater transparency, and fostering collaboration in the industry.”
“We are opening up development of AI to OEMs by giving our customers the ability to automatically upload new training data and deploy new neural networks at scale, with the push of a button.”
Brodmann17 developed this platform to optimize the entire neural network training process: the data selection, parameter tuning, neural network architecture search, and deployment to target embedded processors for runtime benchmark and error analysis. By optimizing the process as a whole, as opposed to optimizing each stage separately, a much better neural network can be achieved to meet customers’ given requirements.
The platform was initially developed as an internal neural network production line to scale Brodmann17’s operations and improve the highly complex process of training and deploying neural networks. The platform now will be available for use by select Brodmann17 customers.
Deployed on different clouds, the automated platform enables OEMs, Tier 1 suppliers and any other Brodmann17 customer to upload their data and run the training process without exposing training videos to the outside. In doing so, the platform protects customers’ private data.
“The cloud-based platform runs hundreds of experiments to achieve the desired result, which – if done manually – would be impractical,” says Amir Alush, co-founder and chief technology officer of Brodmann17. “It performs methodological searches and continuously re-iterates over thousands of servers simultaneously. Because it is cloud-based, the platform is infinitely scalable, and the training process can be completed as quickly as necessary.”
