Filtered by:

Clear All
Date and time to be announced
A Balanced Approach to Inference Implementation at the Edge Markus Levy, NXP Semiconductors show more Description The IoT led to an explosion of data. Though it was once thought the cloud would handle this data, the industry soon realized this was impractical. As a result, machine learning on the edge has increased in popularity and usefulness. One of the most fascinating things about machine learning, or edge computing, is that it can be accomplished on a wide range of device categories, from MCUs with Arm Cortex-M4 and M7 cores to complex SoCs with high-end A-class cores, GPUs, and ML accelerators. This presentation describes the cost and performance tradeoffs for these device categories, using Vision, Voice, and Vibration as examples. Session Type Conference Session Pass Type All-Access Pass Speakers Markus Levy NXP Semiconductors Analyzing Machine Learning Inference with Arm Performance Tools Stephen Barton, Arm show more Description Machine learning workloads can be complex for performance analysis and tuning, especially when the workload utilizes a mix of CPU, GPU and dedicated accelerators within the same system. In this talk, we will describe the performance analysis challenges faced by the different classes of ML users, and show how Arm has extended its performance analysis tools with a deep understanding of the ArmNN framework and ML processor. These new capabilities make it possible to right-size your machine learning algorithms for the chosen hardware platform, or to verify that a platform’s hardware capabilities are being exploited to the fullest extent possible. Session Type Conference Session Pass Type All-Access Pass Speakers Stephen Barton Arm Arm-Synopsys Collaboration to Enable Edge-to-Cloud Computing Rahul Deokar, Synopsys; David Koenen, Network SoC IP / Arm show more Description Arm and Synopsys are collaborating to enable rapid SoC innovation from hyperscale data centers and HPC all the way out to the network and provider edge. These all demand high-performance, efficient, scalable, heterogenous compute solutions. Learn how Arm Cortex-A processors, Arm CoreLink Coherent Mesh Network and System Guidance for Infrastructure (SGI), provide a scalable, energy-efficient solution and, in combination with Synopsys’s solutions (Design Platform with Fusion Technology, QuickStart Implementation Kits (QIKs), Verification Continuum Platform, and DesignWare Interface IP), enable designers to speed creation of their infrastructure products from power-constrained to peak performance on a common software platform. Session Type Conference Session Pass Type All-Access Pass Speakers Rahul Deokar Synopsys David Koenen Network SoC IP / Arm Building Edge Applications on Microcontrollers Cheng-Fu Tan, Arm show more Description Many have believed that machine learning was computationally infeasible on microcontrollers (MCUs). Recent works in algorithm and software engineering have reduced the computational requirement for deep learning significantly. This enables deep learning models to be deployed on affordable and simple systems, the MCUs. Intelligence on the edge will allow us to build smarter devices and serve new applications. In this talk, we will look at how to train a simple model in Tensorflow, then use uTensor and CMSIS-NN to deploy it on Mbed. Session Type Conference Session Pass Type All-Access Pass Speakers Cheng-Fu Tan Arm CCIX: Seamless Data Movement for Accelerated Applications Jon Masters; Milind Matel, Xilinx; Jeff Defilippi, Arm show more Description During Arm TechCon 2017, members of the CCIX consortium provided an introduction to the chip-to-chip interconnect architecture, which was created to solve the performance and efficiency challenges of emerging acceleration applications, such as machine learning, network processing, storage/memory expansion, and analytics that combine processing and acceleration. At this year’s talk, CCIX members will not only focus the advancements to the hardware and software architecture. This talk will also detail the use cases that benefit from the cache coherent, shared virtual memory paradigm and seamless data movement between processors and accelerators, including FPGAs, GPUs, network/storage adapters, intelligent networks, and custom ASICs. Session Type Conference Session Pass Type All-Access Pass Speakers Jon Masters Milind Matel Xilinx Jeff Defilippi Arm Computer Vision at the Edge and in the Cloud: Understanding the Tradeoffs Jeff Bier, Embedded Vision Alliance show more Description Computer vision is rapidly becoming ubiquitous. From devices that prevent automobile accidents, to smart cameras that measure the flow of customers in stores, as well as home assistants that monitor the health of elders, vision is showing up everywhere. A key architectural choice underlies this ubiquity: should vision processing be done at the edge, in the cloud, or a combination of the two? In this presentation, Jeff Bier, founder of the Embedded Vision Alliance, will discuss the benefits and trade-offs of edge, cloud, and hybrid approaches, and when you should consider each option. Session Type Conference Session Pass Type All-Access Pass Speakers Jeff Bier Embedded Vision Alliance Designing Intelligent Systems Using Resource Constrained Edge Devices Jacob Beningo, Beningo Embedded Group show more Description Traditional embedded software engineers often think that machine learning and intelligent systems are outside the realm of microcontroller-based systems and, therefore, outside their realm of expertise. Advances in microcontroller technology have made designing intelligent systems using these resource-constrained devices a reality. In this session, we will examine the tools and capabilities that are available to microcontroller designers to start using machine learning and adding a new level of intelligence to their devices. Developers will walk away understanding that machine learning and AI are not just for big data and the cloud. Session Type Conference Session Pass Type All-Access Pass Speakers Jacob Beningo Beningo Embedded Group Enabling Augmented Reality with SLAM Sylwester Bala, Arm Ltd. show more Description Recent advances in AR and VR have led to some exciting developments in use cases and applications based on Arm technology. In particular Augmented Reality enables a wide range of new use cases and new businesses opportunities in the mobile segment. In this session you will learn about the key Arm based technologies and solutions for Augmented Reality using SLAM (Simultaneous Location and Mapping) as a case study. SLAM is the basis for use cases that range from tracking camera pose in mobile AR and VR to more complex high-level understanding of the real world seen through a camera. It is in the heart of AR headsets, AR smartphones and solutions for self-driving cars, unmanned drones, planetary rovers and a lot more other use cases. You will learn about the importance and challenges associated with the technology on current and future platforms. An insight on the system pipeline and what to take into consideration while building AR and VR solutions to achieve best in class end user experience. Session Type Sponsored Session Pass Type All-Access Pass,Arm Mbed Connect Pass,Expo Floor Pass Speakers Sylwester Bala Arm Ltd. Enabling Real-time Machine Vision and Deep Learning on Small Devices Laurent Itti, JeVois Inc show more Description The next generation of consumer devices will require more natural interfaces, stronger situation awareness, and better perceptual capabilities, both to inform users and to interact with them. Computer vision and deep learning are crucial enablers, yet they have remained difficult to implement on embedded systems. I will survey machine vision and deep learning frameworks and technologies available for deployment today, and demonstrate real-time implementations on $5 Arm Cortex-A7 processors of algorithms ranging from barcode reading, to face recognition, and object recognition using deep neural networks. Session Type Conference Session Pass Type All-Access Pass Speakers Laurent Itti JeVois Inc How Do I Select IP to Use for My Machine Learning System?  Helena Zheng, Arm show more Description Machine learning (ML) processing requirements vary significantly according to the network and workload; there is no “one-size-fits-all” solution. Examining use cases, workloads, and performance data from real networks, this talk will give examples to help you choose the right Project Trillium IP from Arm for your application. Examples will include MCUs for cost- and power-constrained embedded IoT systems through CPUs for moderate performance with general-purpose programmability. Other examples include GPUs for faster performance with graphics-intensive applications to NPUs, such as with the Arm ML processor, for intensive ML processing, giving the highest available performance and efficiency.    Session Type Conference Session Pass Type All-Access Pass Speakers Helena Zheng Arm Low Power Neural Networks on Cortex-M7 CPUs Using the OpenMV Cam H7 Kwabena Agyeman, OpenMV, LLC show more Description In March 2018, Arm released the Arm CMSIS NN library which lets you run neural networks trained with desktop tools like Caffe and TensorFlow on low power Cortex-M7 microcontrollers for edge device IoT computing. With the OpenMV Cam H7 powered by a 400 MHz Cortex-M7 STM32H743VIT6 chip, we demonstrate how to turn desktop NNs into binary files that can be dynamically loaded and run on images. Session Type Conference Session Pass Type All-Access Pass Speakers Kwabena Agyeman OpenMV, LLC Optimized Edge Cloud Use Case with Arm-based uCPE Raanan Tzemach, Telco Systems show more Description Universal CPE (uCPE) is one of the first commercial edge computing use cases. A key challenge is effective deployment of multiple VNFs on a power-efficient device. We will discuss utilizing Arm TrustZone to allow multiple edge applications to share DPI information and other metadata. The talk will discuss the software and SoC aspects of creating an Arm-based uCPE solution. It will compare approaches by NXP/Marvell/Cavium and discuss performance figures. Session Type Conference Session Pass Type All-Access Pass Speakers Raanan Tzemach Telco Systems Practical Challenges and Proposed Solutions for Implementing Embedded AI on Arm Cortex A/M Devices Ganesh Balamitran, Renesas Electronics America Inc. show more Description With the rapid growth in AI and machine vision, the industry’s focus is quickly turning to deploying neural networks for inference on end-point devices, like Arm Cortex A microprocessors (MPUs) and even Cortex M MPUs. In this session, we will share our learnings on the challenges of using 1) various implementations of popular networks on full frameworks like TensorFlow and Caffe, as well as 2) customized/optimized code better suited for embedded MPUs. We will also explore some of the new tool approaches and new tools that are addressing these challenges. Session Type Conference Session Pass Type All-Access Pass Speakers Ganesh Balamitran Renesas Electronics America Inc. Preserving Proprietary Innovation in an Open Source World Mark Orvek, Linaro show more Description Uniquely in the Arm ecosystem, there is the collision of two worlds: the well-established embedded world of often bespoke platforms, stacks, and tools, and the arriving world of open-source software frameworks, often originating on commodity hardware. This session looks at this collision and asks if this meeting of open source and proprietary innovation has to be a zero-sum game. Session Type Conference Session Pass Type All-Access Pass Speakers Mark Orvek Linaro Rise of Accelerated Edge Compute to Service 1 Trillion Connected Devices Jeff Defilippi, Arm show more Description Distributed cloud and network infrastructures, the fundamental building block in our hyper-connected world, rely upon heterogeneous computing to efficiently analyze and service vast amounts of real-time data at its source with standard virtualization and container software stacks. This talk will discuss the evolution of smart acceleration solutions, emerging system requirements, and detail how Arm processors, system IP, custom accelerators, and architectures, such as AMBA, PCIe, and CCIX, can be combined to customize accelerated compute solutions for hyperscale performance at the power constrained edge. Session Type Conference Session Pass Type All-Access Pass Speakers Jeff Defilippi Arm Speed Up Your AI Designs with Dedicated Arm Machine Learning Hardware Ian Forsyth, Arm show more Description Discover the features and benefits of Arm’s Project Trillium's hardware processors: Machine Learning (ML) and Object Detection (OD) processors, their software support, and applicability for different markets and the options for incorporating them in differentiating SoC designs. This talk will describe our strategy and plans for the highly scalable, ground-up designed ML architecture, the markets it will target, and future product iterations. It will also include a comparison with other Arm solutions, enabling you to choose the best software and hardware combination to address your specific needs.   Session Type Conference Session Pass Type All-Access Pass Speakers Ian Forsyth Arm The latest high-performance CPU for laptop-class performance LIONEL BELNET, Arm show more Description The newest high-performance Cortex-A76 CPU from Arm brings laptop-class performance within the smartphone power envelope, enabling even longer battery life, increasing productivity and bringing more compute intelligence to the edge. This talk will explore the applicability of this performance for premium devices ranging from small screens to large. Come take a technical deep-dive to learn more about Cortex-A76 and how it redefines the mobile experience. Session Type Sponsored Session Pass Type All-Access Pass,Arm Mbed Connect Pass,Expo Floor Pass Speakers LIONEL BELNET Arm There Is No One-Size-Fits-All in Machine Learning at the Edge Jim McGregor, TIRIAS Research show more Description Many silicon and system architectures are emerging for edge computing. These solutions vary from using standard logic solutions to dedicated neural processing units (NPUs) and in-memory processing units. While all will work as inference engines, there are tradeoffs between performance, power consumption, manufacturing complexity, cost, and form factor size. The choice is also dependent upon the machine learning task(s) to be performed. As a result, the software model has a significant impact on the choice of machine learning solution. This presentation will discuss the different approaches and the most appropriate use by application and system requirements. Session Type Conference Session Pass Type All-Access Pass Speakers Jim McGregor TIRIAS Research
Close Filters

Session Type


Pass Type

30135 PassType
27253 SecondaryTrack
30821 Date