Explore advanced anomaly detection techniques for embedded machine learning in this tinyML talk. Learn to implement custom DSP blocks for IoT data analysis, leverage feature importance to focus on key frequency bands, and optimize anomaly detection thresholds. Discover how to create effective models for classifying anomalous sensor readings using Edge Impulse's powerful features. Gain insights into data-driven engineering for dataset creation, and understand various applications from cold chain monitoring to fault detection in industrial machinery and satellites. Dive into topics such as impulse design, neural network classification, live classification, and deployment options. Master the use of tools like Feature Explorer, Anomaly Explorer, and EON Tuner to enhance your anomaly detection capabilities on constrained always-on devices.
Overview
Syllabus
Introduction
What is Edge Impulse
Advanced Anomaly Detection
Features with DSP Blocks
Advanced Anomaly Detection Use Cases
Additional Resources
Questions
Blog
Project Dashboard
Adding Data
Impulse Design
Feature Explorer
Neural Network Classifier
Calculating Feature Importance
Live Classification
Anomaly Explorer
EON Tuner
EON Tuner Demo
Model Testing Demo
Deployment Options
Versioning
Importing CSV
Cloud Application
Feature Not Important
Digital Signal Processing
Anomaly Detection
Performance Metrics
Proof of Concept
Will it change
Why add a DSP block
The DSP dashboard
Other features
Other feature outputs
Paid vs free version
Whats next
Strategic Partners
Taught by
tinyML
