Leveraging sensor data, Qeexo is the first company to automate end-to-end machine learning for edge devices (e.g. Cortex M0-M4 class MCUs). Its Qeexo AutoML platform provides an intuitive UI that allows users to collect, clean, and visualize sensor data and automatically build machine learning models using different algorithms. A selected model can be deployed to target embedded hardware with just one click. Delivering high performance, “tinyML” solutions built with Qeexo AutoML are optimized to have ultra-low latency and power consumption, and an incredibly small memory footprint – so tiny it can run on a Cortex-M0+!
[Sensors Supported]
Qeexo AutoML can support any time-series sensor data, including but not limited to IMU sensors (e.g. accel/gyro), acoustic sensors (e.g. microphone & ultrasonic), and environmental sensors (e.g. temp/humidity & air quality).
[Machine Learning Models]
Many equate machine learning with deep learning (neural networks). However, deep learning is often not the best answer, and definitely not the only answer, when it comes to running machine learning on constrained environments at the Edge. Qeexo AutoML can apply 17 (and counting) different machine learning algorithms, including both neural networks and non-neural-networks, to the same dataset, while generating metrics for each (accuracy, memory size, and latency), so that users can pick the model that best fits their data and application.
Libraries built with Qeexo AutoML run locally on edge devices without having to go to the Cloud, resulting in millisecond-latency, high availability, privacy, and security, as well as reduced power, bandwidth and infrastructure costs.
[Target Markets]
Qeexo AutoML can be applied to any market that can benefit from running tinyML (which is pretty any industry). Our current focus is on the industrial space, including quality assurance, condition monitoring, and predictive maintenance of mechanical systems. We have had proven success in the mobile market, having commercialized our ML models on over 400 million devices worldwide. Other markets that are a great fit for Qeexo AutoML include IoT, wearables, smart home, smart cities, and automotive.
Qeexo AutoML is:
- Tiny. Qeexo is one of a handful of companies who are working on automated machine learning for embedded devices (tinyML). We believe that we are the first to automatically build machine learning solutions that can run on something as small as an Arm Cortex-M0/M0+ device.
- Versatile. Qeexo AutoML supports a wide variety of machine learning algorithms, including: SVM, GBM, XGBoost, Random Forest, Gaussian Naive Bayes, Logistic Regression, Decision Tree, CNN, RNN, CRNN, ANN, Local Outlier Factor, and Isolation Forest. This gives the users more flexibility and choice to solve the particular ML problem at hand, to arrive at the highest accuracy given memory, processor, and other constraints. Most competing products only focuses on neural network models.
- Simple. Machine learning is moving to embedded processors on edge devices – improving privacy, latency, and availability. But given limited computation power, memory size, and battery life, building machine learning solutions for edge devices is challenging. Achieving commercial-grade performance requires a team of machine learning engineers, testing algorithms and tuning models. Even for experts, this is a lengthy, error prone, and repetitive process. Qeexo AutoML greatly simplifies the machine learning model building process, with its one-click, fully automated workflow, eliminating room for errors. Qeexo AutoML helps companies make sense of their sensor data without having to invest in building expensive in-house machine learning teams, resulting in huge time and cost savings. Competing products often require coding and machine learning expertise, and are not as intuitive to use.
- Proven. Unlike many other machine learning platform offerings, Qeexo AutoML is a proven solution that has been used internally by Qeexo engineers to build numerous commercial applications over a period of more than five years before making it publicly available. Over 400 million devices worldwide are using machine learning libraries built with Qeexo AutoML.
[Qeexo AutoML Differentiating Features]
- Automates the complex and labor-intensive processes of a typical ML workflow – no coding or machine learning expertise required; generates commercial-level ML models without writing a single line of code!
- Is an end-to-end solution that embeds the data science, machine learning, signal processing, optimization, and embedded engineering needed to deliver AI algorithms for endpoint/edge devices – no need to switch between complicated tools
- Enables a wide range of machine learning methods, including: SVM, GBM, XGBoost, Random Forest, Gaussian Naive Bayes, Logistic Regression, Decision Tree, CNN, RNN, CRNN, ANN, Local Outlier Factor, and Isolation Forest
- Augmented with an easy-to-use interface for labeling, recording, validating, and visualizing time-series sensor data
- Automatically performs feature extraction and selection
- Built-in quantization and model compression to reduce size of models
- Translates machine learning models into C code for ease of deployment to target device
- Out-of-the-box data visualization and model performance evaluation
- Models generated by Qeexo AutoML perform inference on-device and are optimized for constrained environments: low latency, low power consumption, small footprint
- Supports Arm® Cortex™- M0 to M4 class MCUs and ST's MLC sensors usually considered too small to run machine learning