Boon Logic has created patented software, Boon Nano, that can perform Unsupervised Machine Learning unlike ever before. Due to the unique architecture of Boon Logic’s algorithms, machine learning can now be conducted with a fraction of the compute resources. The software’s low compute requirement enables Boon Logic to perform real-time unsupervised machine learning at the edge and at an unrivaled degree of precision and power. Every data point collected is used without downsampling, providing a smarter and more comprehensive model. In addition, since Boon Logic can operate at the edge, actions can be taken quicker with mission-critical solutions.
The precision and overall effectiveness of a model, supervised or unsupervised, is determined by two factors: the quality of data that is used to train the model and the speed in which the data is given. The biggest impediment behind these factors is distance, more specifically the distance between where the data is gathered and transmitted to where the data is modeled and analyzed. Other machine learning solutions use complicated algorithms that require immense computing resources—driving such solutions to the cloud and further away from where the data was originally collected. Large amounts of insightful information are then lost during downsampling, a practice required for the transmission of data to the cloud, lowering the quality of data and impeding the model’s precision. Transmitting data to the cloud, utilizing it in a model, then transmitting it back to an actionable platform on the ground takes time and additional compute resources—potentially compromising mission-critical solutions.
The goal of Machine Learning is to automate and imitate human learning and reasoning through pattern recognition, anomaly detection, and prediction so that valuable and actionable insights can be provided on received information.
Rule-based Machine Learning determines what output to give for a given input by using a set of defined and unchanging rules. Learning at this level occurs exclusively during initial programming.
Supervised Machine Learning is more dynamic than Rule-based Machine Learning as learning can take place after the program is initially created. A labeled training set of data is used to create a model that predicts an output variable (y) with a given input variable (x). Human supervision is then used to adjust the model for increased accuracy.
The most powerful form of machine learning is Unsupervised Machine Learning. Unsupervised Machine Learning uses clustering and association techniques to discover underlying themes in a data set and its structure. The structure of a data set can then be used to identify past inferences that are anomalous and to predict future inferences with a high degree of certainty. During Unsupervised Machine Learning, learning occurs without human intervention and continuously with every new data point.
"Unsupervised Learning is the biggest challenge to developing true AI that can learn without the need for labeled data." - Quoc Le, Google Research Scientist