Neuton AutoML Overview
Neuton is a revolutionary no-code TinyML AutoML platform designed to make edge devices intelligent. Leveraging a patented Neural Network Framework, Neuton automates the process of building extremely compact and accurate machine learning models without requiring any coding. You can generate models that are significantly smaller than traditional methods, often under 5 KB, and embed them directly into microcontrollers and sensors, even those with limited 8-bit capabilities. Neuton is now part of Nordic Semiconductor, accelerating their shared vision of empowering developers worldwide to create smarter, faster, and more efficient AI-powered devices.
Neuton AutoML Key Features
- Automated Tiny Model Creation: Build incredibly small and accurate ML models automatically, without extensive coding or manual parameter tuning.
- Extremely Small Model Sizes: Achieve model sizes of less than 5 KB on average, with some as small as 3 KB, enabling deployment on highly constrained devices.
- No-Code Platform: Simplify the ML development process with a user-friendly interface that requires minimal user intervention.
- Embeddable on MCUs: Seamlessly embed generated models into 8, 16, and 32-bit microcontrollers and sensors without the need for additional compression techniques.
- No Loss in Accuracy: Neuton's unique neural network framework builds models without compromising accuracy, even at minimal sizes.
- Patented Neural Network Framework: Utilizes a novel algorithm that avoids error backpropagation and stochastic gradient descent, leading to more efficient and smaller models.
- Selective Connections: The framework establishes only necessary neuron connections, preventing random growth and optimizing model structure.
- Automatic Structure Growth: Models are grown neuron by neuron, learning from general to specific features to find the optimal size and accuracy.
- Constant Cross-Validation: Each neuron addition is cross-validated, enhancing the model's generalizing capabilities.
- Compatibility: Neuton's silicon-agnostic models can be deployed on a wide range of microcontrollers and small computing devices with challenging memory and processor constraints.