Facebook to Hire AI Chip Head From Google

Company believes that own AI chips will save the company from the need to buy them from other companies
16 July 2018   196

Facebook hired former Google employee Shahriar Rabii to lead the group of artificial intelligence chips developers. They will be used to assess user behavior, TechCrunch reports.

In the corporation, Rabil developed chips for consumer devices, in particular, the Visual Core chip in Pixel 2.

Own chips will save the company from the need to buy them from Qualcomm and will allow money to be used to build their own servers. Thus, Facebook will gain an advantage in the device market - for example, in the work on its "smart" column, the newspaper said.

In May 2018, Facebook announced the development of a chip for video processing. According to the engineers of the company, traditional methods of solving this problem have become inefficient due to high resource and energy intensity.

Oracle to Open GraphPipe Source Code

GraphPipe is a tool that simplifies the maintenance of machine learning models
17 August 2018   136

Oracle has opened the source code of the GraphPipe tool to simplify the maintenance of machine learning models. It supports projects based on the TensorFlow, MXNet, Caffe2 and PyTorch libraries. They are intended for use in IoT-devices, custom web-services and corporate AI-platforms.

The tool eliminates the need for developers to create custom APIs. Also, it eliminates confusion when using multiple frameworks and prevents memory copying during deserialization. The developers hope that GraphPipe will become a standard tool for deploying models.

GraphPipe is free and available on GitHub. It consists of open source tools designed to work with artificial intelligence. For example, the TensorFlow framework and the Open Neural Network Exchange (ONNX) project for creating portable neural networks are among them.

In September 2017, Microsoft introduced own tools for operating with machine learning. At the same time, the company released utilities for using Visual Studio Code when creating models based on the CNTK and Keras frameworks.