AI processors will be launched in the near future, the company states, but for now customers can use the Chameleon simulator to create synthetic vision datasets for training neural networks and the tools, provide a data management framework for deep learning systems.
Mindtech is on a path to introduce integrated AI processors for visual processing in a variety of target markets. This includes what Mindtech describes as a “third-generation” AI engine at the core of the chips with a fourth-generation under development. Target markets include automotive driving, unsupervised machines, retail and security.
It remains unclear what the first two generations of AI technology are, whether Mindtech is making use of licensed AI technology, or is simply characterizing its technology as being of industry-wide third and fourth generations. The company was founded in August 2017.
“Two generations have already seen silicon; the third generation is currently being integrated into our family of processors targeted for launch in the near future. Our integrated processor road map delivers solutions covering IoT sensor fusion, high performance visual processing for client side products and edge servers,” the company states on its website.
For the purpose of training AI processors real-world data sets have a number of limitations including bias, modelling of edge cases, and privacy and GDPR issues. The use of synthetic data improves the accuracy of neural networks, can actively reduce bias and reduce the amount of data required, Mindtech asserts.
Chameleon AI tools manage the merging, verification and augmentation of datasets for use in industry standard frameworks such as TensorFlow and Caffe2. The tools report and visualize relevant statistics for results analysis.
Related links and articles: