Project management,installation and application of third party software, data analysis and model generation using CogNova proprietary systems, and summary and analysis of results.
CogNova has considerable knowledge and experience in a relatively new but increasingly important area of machine learning called Transfer Learning. Transfer learning is the use of prior data or models from one or more source tasks when learning a new but related target task. Transfer learning is of particular value when you have a data mining problem you wish to solve but have insufficient data from which to develop a predictive or classificaton model.
CogNova has applied transfer learning in areas such as medical diagnosis and environmental prediction. For example, we have shown that Transfer Learning can be used to create a more accurate model of heart disease from a small number of patients from a new hospital. An abundance of data for the same diagnostic problem from several hospitals located in three different parts of the world can be used to transfer knowledge to a model for the new hospital. A second example, is the case where an environmental scientists wishes to predict the water level in a stream, pond, or acquiver one or more days from now based on recent weather. To do this using only data from the speciifc water source will require several years worth of weather and water level data. We have shown that with tranfer learning from prior related models, accuracte models can be developed from only one year of data.
We have developed several transfer learning systems and have been able to use them effectively on difficult data mining problems. Cognova is now soliciting for interesting projects that require Transfer Learning.
A number of papers on Transfer Learning are referenced in the Publications section of this website.