The ONNX format, otherwise known as Open Neural Network Exchange format, is meant to give a common way of representing the data that’s used by neural networks. Most frameworks have their own particular model format which works only with models from other frameworks through a conversion tool. Moreover, the format enables models to be freely swapped between frameworks with no need for the conversion process. A model trained in a single framework could be used for inference by another framework.
ONNX FORMAT ADVANTAGES
The ONNX format offers advantages above and beyond no need for converting between model formats. For example, it enables developers to choose frameworks, which reflect the workflow and job at hand, as each and every framework tends to be optimized for various use scenarios, including fast training, supporting flexible network architectures and inference on mobile devices among others.
Facebook notes that little key frameworks already are on board to begin supporting ONXX. PyTorch and Caffe2, which are Facebook’s projects, and Cognitive toolkit, which is a project of Microsoft, provided support in September. This enables models trained in one of the frameworks to be exported in another inference. It’s not clear right away how the ONNX model sizes shape up against those that are commonly used already. The Core ML format of Apple, for example, was designed by Apple so small but accurate models can be deployed to and served from devices of end-users, such as the iPhone. Core ML, however, is proprietary. One of the long-term goals of ONNX is to make it easier to deliver inference models to a lot of targets.
ONNX is an integral part Facebook’s AI team’s deep learning approach. They continuously try to push the AI frontier and build better algorithms for learning. With ONNX, the organization is more focused to bring the worlds of artificial intelligence research and products closer together for faster innovation and deployment of intelligence.
NEW VIRTUAL REALITY TOOLS AND SOLUTIONS
Oculus announced various new virtual reality solutions and tools at the Oculus Connect conference. The company announced the Oculus Go launch, a new VR wireless headset, which does not have to be connected to a phone or personal computer. Also, they announced updates to Project Santa Cruz, a tracked headset, which includes controllers that enable hand presence in virtual reality. Moreover, the company also introduced additional features and tools for developers, like the upgrade to Oculus Avatars as well as dedicated tools for enabling in-app blocking and reporting. Also, they made MultiView available for mobile, enabling developers to build more complex scenarios in virtual reality.
BOX SKILLS NEW FRAMEWORK
Box announces the development of a new framework, the Box Skills. Box, a cloud content management provider announced the development of the new framework. This framework applies machine learning tools for content stored in Box. It enables creating transcripts from video and audio files as well as detect topics and people from image and video files. Also, it announced the Box Skills Kit and Box Graph. The kit could be used by developers that are considering building custom Box Skills. Box Graph is a network of relationships, content, and activity. This allows Box to power new services and experiences for both individual users and organizations.
The power of cloud content management is that there is one central and secure place for all business content in the cloud. Box Graph and Box Skills represent a truly intelligence app for the enterprise, making sure that customers could realize amazing value from each and every piece of content in Box.
MICROSOFT SUPPORT FOR ONNX
Microsoft, the giant tech company, announces preview support for both Facebook and Microsoft’s ONNX format in Cognitive Toolkit, their deep learning toolkit. They hope that the new format will make it easier for developers to build and deploy Artificial Intelligence.
Deep learning and artificial intelligence have gone mainstream, a huge array of companies announced that they will bring compatible products to market. ONNX is described as a standard, which enables developers to move neural networks from framework to framework, provided both adhered to the standard of the system. Organizations should choose the framework that they are going to use for their model before they begin to develop it. However, the framework that provides best options for testing and tweaking a neural network are not necessarily the frameworks with features one wants when bringing a product to market.
It’s exciting that these organizations decided to join in the mission of ONNX support in their ecosystem. Enabling interoperability between various frameworks and simplifying the path from research towards production would help boost the innovation speed in the artificial intelligence community.
ONNX has acquired the support of numerous technology systems and organizations. These organizations offer support of the system in their ecosystem. This simplifies the path from research into production with speed in innovation. ONNX offers a common way of data representation.