Data is growing faster than processor speed. The value of data to government and enterprise users is skyrocketing, as knowledge contained within the data can be put to valuable use - if you can extract it.
Treeminer uniquely organizes data specifically for knowledge extraction. Our novel approach organizes data in vertical strips, enabling breakthrough performance increases - data mining results in minutes instead of hours. Why sample your data when all you need to do is organize it correctly?
Class-driven Classification
Modern image classification systems, whether supervised (where an expert has given the system some examples of each class you are looking for) or unsupervised (where the system makes a guess as to the classes present in the data set), require a point by point traversal of all the pixels in the image to make a determination as to its class (for example, is it grass? is it a road? is it a good spot to dig for a mineral?). This approach works, but has a shortcoming that as images increase in resolution, the time it takes to classify the image grows as well. We call this a point-driven classification system. Treeminer is pioneering a new approach to classification - class-driven classification. Typically, while the image may increase in resolution, the number of classes in that image is significantly smaller than the image size, and constant. By classifying the image class by class rather than point by point, dramatic performance increases are possible.
|
The Data ContinuumOver the years, data storage methods have evolved to meet the needs of the moment. As relational databases emerged in the early 1970's, query languages such as SQL evolved to make it efficient to retrieve data when you know what you are looking for. Today, technologies such as hadoop are bringing unstructured data into play, driving dramatic growth in the amount of data available to organizations.What fundamentally hasn't changed is how data is organized for processing by data mining algorithms that look for patterns in the data or attempt to make predictions based on new data still need to structure data in order to execute.
|
