CANN for Edge AI Deployment

• Experience with AI model development or deployment workflows
• Basic knowledge of embedded systems, Linux, and Python
• Familiarity with deep learning frameworks such as TensorFlow or PyTorch

Audience

• IoT solution developers
• Embedded AI engineers
• Edge system integrators and AI deployment specialists

Huawei's Ascend CANN toolkit enables powerful AI inference on edge devices such as the Ascend 310. CANN provides essential tools for compiling, optimizing, and deploying models where compute and memory are constrained.

This instructor-led, live training (online or onsite) is aimed at intermediate-level AI developers and integrators who wish to deploy and optimize models on Ascend edge devices using the CANN toolchain.

By the end of this training, participants will be able to:

• Prepare and convert AI models for Ascend 310 using CANN tools.
• Build lightweight inference pipelines using MindSpore Lite and AscendCL.
• Optimize model performance for limited compute and memory environments.
• Deploy and monitor AI applications in real-world edge use cases.

Format of the Course

• Interactive lecture and demonstration.
• Hands-on lab work with edge-specific models and scenarios.
• Live deployment examples on virtual or physical edge hardware.

Course Customization Options

• To request a customized training for this course, please contact us to arrange.

Introduction to Edge AI and Ascend 310

• Overview of Edge AI: trends, constraints, and applications
• Huawei Ascend 310 chip architecture and supported toolchain
• Positioning CANN within the edge AI deployment stack

Model Preparation and Conversion

• Exporting trained models from TensorFlow, PyTorch, and MindSpore
• Using ATC to convert models to OM format for Ascend devices
• Handling unsupported ops and lightweight conversion strategies

Developing Inference Pipelines with AscendCL

• Using the AscendCL API to run OM models on Ascend 310
• Input/output preprocessing, memory handling, and device control
• Deploying within embedded containers or lightweight runtime environments

Optimization for Edge Constraints

• Reducing model size, precision tuning (FP16, INT8)
• Using the CANN profiler to identify bottlenecks
• Managing memory layout and data streaming for performance

Deploying with MindSpore Lite

• Using MindSpore Lite runtime for mobile and embedded targets
• Comparing MindSpore Lite with raw AscendCL pipeline
• Packaging inference models for device-specific deployment

Edge Deployment Scenarios and Case Studies

• Case study: smart camera with object detection model on Ascend 310
• Case study: real-time classification in an IoT sensor hub
• Monitoring and updating deployed models at the edge

Summary and Next Steps