Deep Learning from Idea to Embedded Hardware
Scientists and engineers are challenged on how to quickly prototype a complete algorithm—including pre- and post-processing logic around deep neural networks (DNNs) —to get a sense of timing and performance on standalone CPUs and GPUs.
Traditionally, they would hand translate the complete algorithm into C/C++ or CUDA® and compile it with the NVIDIA® toolchain. With MATLAB®, they can short-circuit the standard process by generating CUDA code that compiles on modern NVIDIA GPUs.
MATLAB provides an integrated framework to develop, prototype, and deploy the complete algorithm workflow. It automates labelling of ground truth data and accelerates training models in HPC without changing your code. MATLAB supports the Open Neural Network Exchange (ONNX™) framework that scientists and engineers can use to improve team collaboration. Key capabilities include:
- Test and debug deep neural networks (DNNs) architectures interactively with Deep Network Designer
- Quickly prototype a complete workflow in MATLAB running on desktop GPUs
- Generate code, cross-compile, and run software-in-the-loop (SIL) testing on any modern NVIDIA GPUs, from Tesla® to DRIVE to Jetson AGX Xavier platforms
Load Forecasting System
By connecting distributed devices, the Internet of Things (IoT) creates the possibility to collect large amounts of data from a wide range of sources. This is transforming the way businesses operate in many fields. A prime example is how the IoT helps power system operators accurately predict electricity demand to schedule the dispatch of power plants. Recently, this task has become much more complicated due to the continued integration of many distributed sources. However, using IoT technology, power systems operators now also have unprecedented access to streams of real-time measurement and weather data. Deep Learning can be used to find trends and predict the electricity demand based on the collected data.
In this demo, a neural network is trained on a desktop MATLAB environment to continuously forecast electrical load in the State of New York, and then deployed as a cloud application using the MATLAB Production Server™.
Visit this showcase area to see and discuss how MATLAB is used for:
- End-to-end data analytics workflow
- Multiple predictive modeling techniques, including neural networks and machine learning
- Cloud-based deployment
Mechatronic System Design, Simulation, and Validation
Mechatronic systems require expertise from multiple domain experts. Simulating motors, electronics, and sensors together with mechanical and control systems is critical to optimizing system performance. System engineers need to perform trade-off studies to choose the correct design and implementation based on requirements.
To ensure that testing is efficient, MathWorks offers several ways to easily balance the trade-off of model fidelity and simulation speed, as well as the ability to import data from various sources, from CAD to measurement files as well as technical datasheets. The ability to generate C code from both the controller and the system model enables engineers to test the design in real time.
- Use comprehensive component libraries and industrial examples to model your mechatronic system
- Perform trade-off studies and sensitivity analysis to reach optimal design
- Validate software design early by simulating it with the system model in one unified environment
Motor Control with Systems-on-Chip
From simulation models to system-on-chip (SoC), see the implementation of a motor regulation algorithm on a heterogeneous SoC made of an FPGA and an ARM processor through IP core generation workflows and automatic HDL and C-code generation. This demo is based on a Xilinx® ZedBoard™ using an Analog Devices® motor control FMC board, and it runs a field-oriented controller (FOC) for a permanent magnet synchronous machine (PMSM).
Visit this showcase area to see and discuss how MATLAB is used to:
- Reduce dependency on hardware with simulation
- Perform faster hardware iterations with automatic deployment
- Perform on-target debugging with FPGA data capture
Learn how System Composer™ can help you create architecture models including components, interface definitions, and systems behaviors that meet requirements.
MathWorks Training Services
For MATLAB and Simulink instruction, MathWorks provides an expert team with exclusive product knowledge.
Each course is designed to help participants quickly master necessary skills and workflows. A hands-on approach allows participants to practice, apply, and evaluate their knowledge in the classroom. There are a variety of training paths available for users with specific job roles or application areas, which means new and advanced users of all application areas can find training that suits their needs.
Discover how courses from MathWorks Training Services can accelerate your use of MATLAB and Simulink. Come by the booth to:
- Learn about training offerings—from public training throughout Europe to online self-paced training at a time and location that suits you best, we can tailor training to your needs
- See training materials for various courses
- Interact with online self-paced materials, including the free Onramp trainings, which enable new users to get up to speed in only a few hours