Designed for product developers, researchers, and manufacturers, the Movidius Neural Compute module seeks to reduce current barriers to the development, tuning, and deployment of AI applications by offering the high processing power of a dedicated deep neural network in a compact module.
As more developers adopt advanced machine learning approaches to create innovative applications and solutions, Intel remains committed to providing the most comprehensive development tools and resources to ensure developers are equipped for an AI-driven digital economy. Whether training artificial neural networks in the Intel® Nervana™ cloud, optimizing for emerging workloads such as artificial intelligence, virtual and augmented reality, or autonomous driving using Intel® Xeon® Scalable processors, or pushing the boundaries of AI with the Movidius vision processing unit (VPU), Intel offers a broad portfolio of AI tools, as well as options for training and deploying the next generation of AI-based products and services.
The development of artificial intelligence consists of two fundamental stages: (1) training an algorithm using large datasets with modern machine learning techniques, and (2) using the algorithm in end-user applications that interpret real-world data. This second stage is known as "inference," and applying inference externally or natively on the device offers numerous benefits in terms of latency, resource consumption, and privacy:
Compile: automatically convert a pre-trained Caffe-based convolutional neural network (CNN) into an integrated neural network optimized to operate directly on the Movidius Myriad 2 VPU.
Tune: performance metrics per layer for both standard and custom-designed neural networks allow for effective tuning to achieve optimal performance with real-world data and ultra-low resource consumption. Validation scripts allow developers to compare the accuracy of the optimized models on the device with that of the original PC-based models.
Accelerate: The Movidius Neural Compute module is the only one capable of acting as a discrete neural network accelerator by adding dedicated deep learning inference capability to current computing platforms to improve performance and efficiency.
