 |
Center for Nonlinear Studies |
Getting all of the data in the world does not in and of itself solve complex problems or answer intelligence questions. One of the key demands on the practice of modern intelligence analysis is the identification of relevant data, and the immediate evaluation of new data coming into the system. Most data elements are not important, but occasionally key data points may be collected that complete a puzzle, connect otherwise disparate data points, or demand a reevaluation of fundamental assumptions. As new data feeds come online, separating the critical from the irrelevant becomes increasingly important. Placing new information in context is vital in a field where resources wasted on false trails places National Security at risk. Realtime processing brings new meaning to the term Current Intelligence.
Dr. Goggins will present an example using Tensor Flow to systematically integrate numerous computational models trained to identify anomalies and changes in both individual and multiple data flows. The example focuses on detection of regional changes in the use of racial slurs on Twitter and the identification of news and scraped discussion forum activity at the local, regional and national level that are candidate drivers of a sudden change. Analysts are presented with events and candidate drivers. Each new event detected by an analyst is evaluated and modeled individually. If a particular anomaly detection is demonstrated to be accurate, the embedding matrices used for this category of event are updated. In this way the judgement of analysts becomes systematically integrated into perpetually evolving computational models for predicting civil unrest or other potential threats before they occur. The technology and process has potential application far beyond the example.
Host: Sara Y Del Valle