White Paper

WHITE PAPER Accelerating Big Data Processing and AI at the Edge mrcy.com 6 ENGAGE WITH MERCURY TO EVALUATE DEPLOYABLE AI OPTIONS Mercur y is committed to driving for ward GPU-based AI capabilities, making them available as solutions for deployment on aerospace and mission-critical systems at the edge. We see close cooperation with public sector system visionaries, program managers and engineering teams as key to recognizing maximum value from all of our solutions. Our technology exper ts work to understand individual applications, then collaborate to develop optimized system solutions. Engage with Mercur y 's embedded exper ts to explore how moving AI to the edge on 5G networks can deliver power ful new capabilities to your programs. Many NVIDIA SDKs suppor t the P ython programming model, so customer application development teams can work with the SDKs using a familiar, commonly used language. Developing AI Applications for the Real World The last step in suppor ting AI at the edge is developing functional and flexible AI application software that can effectively deal with real-world situations in real time. AI software applications can be extremely complex. While P ython libraries and GitHub make developing cer tain applications from scratch easier, others remain labor- intensive, time-consuming projects with schedules that are difficult to predict—not characteristics a program manager is looking for. For tunately, new AI software for specific applications can be created by building on a base of existing software. NVIDIA is the leader in offering building-block software modules for GPUs, modules that can be efficiently customized by an application development team. For example, many AI applications use deep learning techniques to recognize and identify images. NVIDIA has modules that are pre-trained to recognize cer tain types of images, e.g., common vehicles. Customers can then star t with those modules and add training for additional vehicle types, developing deep learning software optimized for their application ver y quickly. This use of pre-trained modules, referred to as transfer learning, could likewise be utilized for speech recognition or gesture recognition. NVIDIA offers developers over 800 SDKs, including software stacks and function libraries optimized for GPU per formance. NVIDIA SDKs suppor t AI application development across a host of functions beyond image recognition, including augmented reality (AR), vir tual reality (VR), and speech recognition. Other SDKs streamline the communications component for applications, such as the Aerial™ framework for building and deploying GPU-accelerated 5G networks. Building-Block Software Modules NVIDIA offers pre-trained starting modules that can be efficiently customized.

• Cover