MediaTek is now officially part of the neural network conglomerate known as the Open Neural Network Exchange (ONNX), according to an announcement made February 22. The general goal of ONNX, for those who may not already be aware, is to establish standards for interoperability as it pertains to transferring deep learning models between frameworks. The standards allow those working with A.I. to more easily match the tools and frameworks they need, with the ability to implement those during any stage of a project. Mediatek's involvement in the group, which was previously only comprised of Microsoft, Amazon, and Facebook, is intended to help evolve the edge computing segment of ONNX's A.I. platform further.
Beyond that, combining participation with ONNX with its current involvement in the Android Neural Network project will allow MediaTek expand its own portfolio of A.I.-enabled technologies. Specifically, it will help the company to build on its already robust edge A.I.-heavy NeuroPilot A.I. Platform and should help the hardware and software associated with that to be more inclusive and open. The platform already supports a wide array of frameworks and neural processing SDKs – including Google's TensorFlow, Caffe, Amazon MXNet, and Sony NNabla, to name a few. For those who don't already know, NeuroPilot is comprised of a dedicated SDK and A.I. processing unit that supports mobile, smart home implementations, and the burgeoning automotive technology industry. The expansiveness of MediaTek's efforts in A.I. and its experience with those will also undoubtedly be extremely useful as a supplement to ONNX's current efforts.
Interoperability is a going to be an increasingly important component of A.I. as the technology continues advancing and the IoT. This move should give MediaTek a substantial boost with regards to developing its platform for edge-based A.I. computing solutions in support of nearly every facet of the technology industry. Moreover, MediaTek's own ties in the industry stretch substantially further than the other companies involved in the partnership. That may provide the group with exactly the knowledge needed to get an edge over any competing efforts to create a standardization for the interoperability that has been missing from many IoT products.