Part
01
of one
Part
01
Deep Learning Uses: US Department of Commerce
The U.S. Department of Commerce deploys deep learning capabilities to various functions including weather forecast research, lane marking detection improvements, spectrum sharing research, among others. The technology is currently in use by several other agencies under the U.S. Department of Commerce such as the Economic Development Administration, and Minority Business Development Agency. A full list is included at the end.
RESEARCH STRATEGY
We commenced the research by first locating the U.S. Department of Commerce and its agencies to identify those deploying deep learning. We found out that presently, deep learning is being used by three primary agencies, namely the National Institute of Standards and Technology and the National Oceanic and Atmospheric Administration. Importantly, based on the research findings, we concluded that the Department of Commerce is currently deploying deep learning specifically for research or testing capacity. Further advancements in the use of deep learning by the department are on standby because the data sets required are still in the development phase. Read of for elaborate illustrations regarding the use of deep learning by the Department of Commerce and find a detailed list of all agencies under the U.S. Department of Commerce likely to benefit from deep learning capabilities.
1. SPECTRUM SHARING RESEARCH
The NIST (National Institute of Standards and Technology) is currently collaborating with the Federal Communications Commission (FCC) to develop a strategy for the U.S. Navy and commercial wireless providers. These efforts are aimed at creating technologies to allow U.S. Navy and commercial wireless vendors to share an extensive section of the 150-megahertz (MHz) radio frequency (R.F.) spectrum for communications.
According to NIST researchers, deep learning algorithms offer sophisticated functionalities capable of detecting operating offshore radars compared to existing technologies in that domain. The researchers managed to develop eight deep learning algorithms capable of detecting offshore radar signals. The algorithms were then compared to energy detectors to establish the best performing technology. It turned out that deep learning algorithms outperformed energy detectors. Overall, NIST continues to improve the algorithms with more data to ensure improved and optimal performance in the future.
2. WEATHER FORECASTING RESEARCH
NIST researchers during a February 2019 conference proposed the creation of a deep learning weather forecast system. The researchers analyzed real weather data via the Keras Sequential model (deep learning tool) to determine and try to predict the weather. Overall, the experiment showed great potential based on data results drawn from five different U.S. regions.
3. NANOSCALE MICROSCOPY DEVELOPMENT
NIST physicists are combining both AI (artificial intelligence) and deep learning capabilities to enhance the quality of “novel nanoscale microscopy techniques.” Existing imaging solutions are incapable of comparing images effectively; thus, experimental data is used. However, it is anticipated that deep machine learning capabilities hold the key to this problem and offer flexible and powerful solutions to develop a multi-dimensional dataset. Also, the unavailability robust training data hinders further development of these solutions, which undermine advancements in microscopy techniques.
4. LANE MARKING DETECTION IMPROVEMENTS
A NIST-based researcher, along with other experts, recently developed deep machine learning dataset that significantly enhances the accuracy required in detecting faces, hands, pedestrians, and other objects in photographs or video. Dubbed “The Patch Proposal Network (PPN)” the solution has a small field of identifying objects then outputs them using pixels with different scales. Presently, the PPN is deployed along with the KITTI-ROAD dataset (an automobiles lane detection system) and other traffic scene lane marking data points.
5. HUMPBACK WHALES STUDIES
Current technologies being used at the National Oceanic and Atmospheric Administration (NOAA), the Fisheries Pacific Islands, and the Fisheries Science Center are incapable of detect Humpback whales that produce varying vocalizations that also change fast. Researchers in this field are deploying high-frequency acoustic recording packages (HARPs) equipped with underwater microphones to record sounds from whales, dolphins and other artificial or environmental noises.
However, Google’s sophisticated AI and deep learning technologies powering their voice recognition services were faster at detecting Humpback whale vocalizations. Nevertheless, while more training for the Google AI dataset is still required, researchers and analysts alike have expressed optimism that those new data sets will significantly improve studies on whale vocalizations.
6. MARINE BIRD STUDIES
A study commissioned by the NOAA (National Centers for Coastal Oceans Science) built maps of marine birds and how the different bird species are distributed across the Atlantic Outer Continental Shelf waters. These maps shall be used for future research efforts and planning. The study included the deployment of an ensemble machine-learning solution to detect bird species and their habitats. Overall, the researchers cautioned that their maps are only predictive, and further field studies and analyses are required to ascertain the accuracy of the data the maps predict.
U.S. DEPARTMENT OF COMMERCE AGENCIES
- Bureau of Economic Analysis
- Bureau of Industry and Security
- Committee for the Implementation of Textile Agreements
- Economics and Statistics Administration
- Economic Development Administration
- International Trade Administration
- Minority Business Development Agency
- National Institute of Standards and Technology
- National Oceanic and Atmospheric Administration
- National Technical Information Service
- National Telecommunications and Information Administration
- U.S. Census Bureau