KaNN platform provides a seamless tool to developers of high-performance Machine Learning inference on Kalray’s MPPA® Intelligent Processor
Grenoble, France and Brussels, Belgium, November 13, 2018 – Kalray (Euronext Growth Paris : ALKAL), a pioneer in processors for new intelligent systems,today announced the launch of the Kalray Neural Network 3.0 (KaNN), a platform for Artificial Intelligence application development. KaNN allows developers to seamlessly port their AI-based algorithms from well-known machine learning frameworks including Caffe, Torch and TensorFlow onto Kalray’s Massively Parallel Processor Array (MPPA®) intelligent processor.
Artificial Intelligence is at the heart of a growing number of applications such as autonomous vehicles, intelligent storage servers, data centers, robotics, drones, and more. Kalray’s manycore MPPA intelligent processor has been architected from the ground up to meet the incredible performance requirements of such advanced applications. However, it is crucial to provide an easy way for AI developers to seamlessly support the AI networks they develop.
“The Kalray MPPA intelligent processor is capable of processing massive amounts of data and trillions of instructions in parallel and in real time, making it the perfect architecture for Artificial Intelligence applications like self-driving cars and intelligent storage algorithms,” said Stéphane Cordova, Vice-President of Embedded Business Unit at Kalray. “Today, we are pleased to release the third version of the Kalray Neural Network platform to the AI and machine learning community. The KaNN is modular, compatible, easy-to-use and customizable, and it empowers developers to unleash the full potential and performance of AI applications built for Kalray MPPA processors.”
KaNN 3.0 has been designed to be integrated into a more complex system as a module. Intelligent systems are not composed of a single AI function, but rather dozens of highly demanding functions running in parallel. As an example, an autonomous driving system runs dozens of AI-based and non-AI-based algorithms simultaneously. Kalray’s MPPA processors offer a unique capability to integrate, on a single processor, all the critical functions of an intelligent system, without performance degradation and while offering a guarantee of freedom of interference. This makes the MPPA unique to address intelligent system challenges that industry encounters in applications such as next generation vehicles, planes, drones or robots.
The Kalray Neural Network was created to give developers convenient access to the full power of the MPPA, including an easy way to view and modify the processor’s network mapping. In addition to being compatible with any machine learning framework, KaNN imports the trained models from these frameworks to ensure optimal execution of AI algorithms from deep learning networks such as GoogLeNet, ResNet, YOLO and others. Users of KaNN enjoy simplified prototyping, accelerated Convolutional Neural Network (CNN) development, and full visibility in the process of generating built-in “C” code and its deployment.