Falkonry has won allowance of its time series deep learning patent application focusing on the use of convolutional variational autoencoder generated encodings for classification and anomaly detection. This patented GPT is now the basis of the new product Falkonry Insights. In three years, we have gone from initial research to patented innovation to commercially supported product to competitive advantage. This patent is a testament to the innovation that the Falkonry team has unleashed on the less loved field of time series data management. This invention is the brainchild of Vukasin Toroman and Dan Kearns among others.
Ali Ghodsi , CEO of Databricks , has taught us that wherever there is lots of data, there will be AI. Nowhere is this more true than with unstructured data such as time series. Manufacturing and process engineers are increasingly managing and reviewing unprecedented amounts of operational data. Thus, it would be helpful to have an improved solution to processing, storing, or visualizing large volumes of data.
The patented approach uses multiple time scales (e.g., hierarchical time ranges) to remove noise while taking into account distribution, sequence and context. This approach is responsive to data compression, oversampling, non-uniform sampling, outliers, noise, transmission delays, limited computational resource, and a variety of other practical limits and idiosyncrasy of industrial operations. This latest innovation builds on the previous patent on panning and zooming large volumes of time series data, which has been commercially implemented into the Falkonry AI Cloud.
P. S. Agatha H. Liu is my favorite patent attorney. I have worked with her for close to 7 years after she started assisting Christopher Palermo at Hickman Palermo. While I still don’t like reading patent applications (I do have close to two dozen patents), when you read this one, you will see the some of the best IP drafting.