Login to the D&D KM-IT.

Not a registered user? click here for U.S. registration or here for international registration.      Forgot your password? Click here

User name:   Password:   Close

Search the D&D KM-IT

Welcome Guest
Try our mobile friendly tool.

Share page:  

printer friendly logo

Tech Talk
Natural Language Processing for Event Discovery, Extraction, and Inferential Forecasting
October 18, 2023

Natural Language Processing for Event Discovery, Extraction, and Inferential Forecasting


On October 18, 2023, FIU featured a Tech Talk with Dr. Thomas Danielson from Savannah River National Laboratory who presented his research on Natural Language Processing for Event Discovery, Extraction, and Inferential Forecasting.

Tech Talk Description
The world wide web is a massive open source dataset that contains broad ranging, ever evolving information in the form of social media posts, news articles, journals, blogs, and public records, to name a few. Within this ever expanding information paradigm, there continues to be a need to scour the open source for information related to many different topics and form logical connections between entities and activities or events across time such that inferential forecasts or audit trails of historical events can be made. Furthermore, manual curation and processing of the data is likely to miss relevant information and/or overlook the connectedness of obscure or seemingly disparate information.

Therefore, starting in 2020, the Department of Energy’s Office of Defense Nuclear Non-Proliferation Research and Development funded researchers from the Savannah River National Laboratory to develop a machine learning based modeling pipeline to extract events of interest from the massive open source and provide inferential forecasting capabilities based on the trajectory of information extracted. Built on a foundation of natural language processing, the pipeline uses word embedding models to identify contextual shifts in key words and phrases across time, which act as indicators of events of interest. Having discovered key points in time for entities, topics, or keywords of interest, the pipeline leverages transformer based language models to extract and/or summarize relevant information. The application of the modeling pipeline to two different topical domains – “plutonium pit production at the Savannah River Site” and “worldwide state-sponsored civil nuclear power” – will be presented, along with potential future use cases and developments.

Flyer    Presentation    Feedback Form    Newsletter Signup

 

Video Archive


Speaker

Dr. Thomas Danielson

Dr. Thomas Danielson


Environmental Sciences and Dosimetry
Savannah River National Laboratory

Dr. Thomas Danielson received his PhD in Materials Science and Engineering from Virginia Tech in the spring of 2016 and has undergraduate degrees in physics and mathematics. Thomas joined the Savannah River National Laboratory in September of 2016 and is currently a Scientist in the Environmental Sciences and Dosimetry group. Dr. Danielson has played a key role in developing and executing deterministic models for groundwater contaminant transport for various activities at the Savannah River Site, including performance assessment at the E-Area Low Level Waste Facility and deactivation and decommissioning of the 235-F facility. For the past several years, Dr. Danielson has been applying artificial intelligence, machine learning, and advanced data analytics in the areas of contaminant transport, meteorological forecasting, and nuclear non-proliferation.


Joining the meeting instructions

This event is being hosted using Microsoft Teams. It is required for every attendee to have this app installed on their desktop or mobile device. You can download this app from the following link or use the links of the sidebar for mobile devices.

https://www.microsoft.com/en-us/microsoft-365/microsoft-teams/download-app

This event is sponsored by The U.S. Department of Energy

Back to Top
More Modules

Download Original    Management of D&D of Oak Ridge Building 3505 | After

Pre cache