DAHLGREN, Va. –
Hacking is not normally a skill you would highlight on a resume. According to a popular security company, there are around 2,200 cyberattacks a day that could equate to more than 800,000 people being hacked per year. At Naval Surface Warfare Center Dahlgren Division (NSWCDD), hackathons hosted at Dahlgren’s Innovation Lab (iLab) provide the building blocks for positive hacking skills for scientists and engineers.
The fourth hackathon of the year was held Oct. 25 through Oct. 27 at the iLab. The accelerated artificial intelligence (AI) and machine learning (ML) exercises generated with NVIDIA the perfect environment for NSWCDD employees to learn and gain exposure to AI and ML applications.
This hackathon focused on the AI/ML computing environment, building algorithms and applying the environment to image processing and object detection, which is a technically challenging area of expertise. The previous hackathons hosted this year introduced interns and new employees to surface warfare capabilities through a wargaming simulation.
For this hackathon, industry collaborative research and development partner, NVIDIA Corporation – a multinational technology company – provided Jupyter notebook guides that utilized powerful pre-trained models as a basis for Deep Learning. Dahlgren employees used these pre-trained models, in addition to various training strategies and techniques, to train customized models. Models were then exported and deployed to NVIDIA Jetsons which are small form-factor, high-performance computers, built specifically for AI/ML applications.
There was plenty of ground for NSWCDD scientists and engineers across all technical departments to cover during the three-day event. The event’s NSWCDD facilitators were Molly Thomson and Ben Goldman from the Autonomous Weapons and Robotics System Branch, iLab Director Tamara Stuart, and Senior Scientific and Technical Manager for Naval Data Sciences Dr. Jeffrey Solka.
“This is a challenging technical area and any additional training tools or computer power we can provide to our workforce will enable us to more rapidly deliver combat ready capabilities to the warfighter,” said Stuart.
Topics varied from data augmentations, model evaluations and optimizations, automatic speech recognition (ASR) and the NVIDIA train, adapt and optimize (TAO) toolkit, which uses transfer learning to train a model, based on pre-trained models.
“Training the workforce with these technologies will allow us to deploy future AI and ML capabilities to the Navy’s weapons systems in a more efficacious manner,” explained Solka.
During the hackathon, participants researched data augmentation techniques and applied them to the training modules to see the results. They also used model evaluations and optimizations to determine accuracy in their models, along with being introduced to pruning, data augmentations and re-training.
Although the event focused heavily on object detection classifiers and image processing, there was also opportunity to incorporate and show all the TAO capabilities through an ASR demonstration.
“Instead of spending hours and hours training and instead of spending days to months to years labeling data, you get access to a pre-trained model to learn from, which you can then use transfer learning to deploy those onto a Jetson,” said Thomson. “The hope is to have all the technical departments go through this process and understand how to incorporate transfer learning into their own organization to better benefit the warfighter.”
The hackathon was a success in empowering the workforce at NSWCDD and presenting new training, experience and efforts. NSWCDD scientist Adam Hawkins gained a new appreciation for the TAO toolkit during the event. “I enjoyed learning about the latest advances in machine learning and artificial intelligence and their novel applications,” said Hawkins. “The event is informative to the workforce about the power of transfer learning and showing the accuracy of the teaching models can be trained with much less data, time, resources and expertise than when used for the original model. This enhances the accessibility to the workforce.”
“NSWCDD is always looking to build our workforce with exposure and experience to better support the warfighter,” said Goldman.