nano
Edge Deployed Unsupervised Machine Learning
PicoAI is a lightweight architecture for resource-constrained microcontrollers that runs without GPUs or cloud resources. Powered by Nano , PicoAI enables edge-native AI with on-device training and real-time inference for pattern recognition and anomaly detection across various industries.
One-Pass
N-dimensional segmentation algorithm that clusters floating-point vectors
Scalable Training
Unsupervised ML models from a blank slate adding new clusters as needed while consuming the training data
Any Platform
Leveraging general purpose processors (from microcontrollers up to server-grade CPUs)
No GPU
Trains and runs without GPU or cloud support
PicoAI is transforming industries globally
Energy
Consumer
Wearables
$800k Compressor Failure Avoided
88% Fewer Falsely Accepted Defects
1,000x Faster than K-Means
What is unsupervised ML?
Unsupervised machine learning
In this type of machine learning, the model is trained from unlabeled data, and its objective is to discover patterns, structures, or relationships within the data.
Best use cases:
- Anomaly detection
- Predictive analytics
- Clustering/segmentation
Supervised machine learning
In this type of machine learning, the model is trained from a labeled dataset to create a mapping from input data to output classifications.
Best use cases:
- Classifiers
- Object recognition
- Natural Language Processing (NLP)
How it works
PicoAI brings powerful self-learning capabilities directly to embedded devices, training on real operational data without relying on the cloud or pre-trained models. By learning what "normal" looks like locally, PicoAI identifies anomalies and usage patterns 1,000 times faster than traditional methods – with precision, even in dynamic and unpredictable environments.
At its core, PicoAI uses an extremely efficient clustering algorithm that supports up to thousands of clusters per model, far outperforming conventional edge AI solutions that typically manage only a few dozen. This makes PicoAI one of the most accurate and scalable pattern recognition engines available for embedded systems.
Superior speed
If your project demands real-time insight from high-speed data or the ability to train and evaluate models quickly, PicoAI is built for the job. It processes massive datasets while keeping inference times steady, even when feature counts are high. Designed for demanding, real-time environments, PicoAI can be deployed entirely on-premise or in the cloud.
Built for complexity
If you’re working with datasets containing hundreds of features and intricate relationships, PicoAI is built to handle that level of complexity. It supports real-time training and inference with models containing up to 5,000 clusters across a 500-dimensional space, making it ideal for high-dimensional, relational data.
High accuracy
If your application demands exceptional accuracy, PicoAI is designed to deliver it. By forming thousands of clusters to represent the data, it achieves highly precise segmentation and anomaly detection. You can train individualized models for each distinct data source, resulting in identification and detection with outstanding specificity and sensitivity.
Implementation
BoonLogic has partnered with MicroEJ to make implementation a snap. MicroEJ has developed the platform MICROEJ VEE as a virtual execution environment which is a container for embedded systems supporting MCU’s, MPU’s and SoC’s.
MICROEJ VEE acts as a software container that runs on any OS/RTOS commonly used in embedded systems (FreeRTOS, Zephyr Project,QP/C, ucOS, ThreadX, mBed OS, Mbed OS, VxWorks, PikeOS, Integrity, Linux,…) and can also run without RTOS (bare-metal) or proprietary RTOS.
MICROEJ VEE is powered by the small virtual processor MEJ32 (a 32-bit virtual core), along with a wide range of free libraries.