More processes and decisions are being handed over to machines, as some machines make better – and more accurate – decisions than humans. Yet, machine learning is still in its infancy. The field has been around for just over 70 years, and we still have a lot to learn, even as state of the art machine learning models become increasingly complex. However, complexity arises out of the composition of many simpler, fundamental operations rather than the atomic components themselves. For example, deep neural networks are rarely built from more than operations such as addition, multiplication and non-linear transformation.
Now, SMART scholar and SEED Grant recipient Matthew Klawonn, Ph.D., from the Air Force Research Laboratory, is setting out to understand the component structures to create the tools and language to describe system properties. With little pre-existing research, Matthew is beginning fundamental research using a category theoretic perspective to understand simple machine learning algorithms. Once the building blocks of the models are understood, this three-year effort will seek to understand their composition. Follow on efforts will aim to create a toolbox that allows practitioners to specify behavior models that automatically infer what architecture and training procedure is needed.
As the Department of Defense (DoD) becomes more reliant on machine learning and artificial intelligence, an increase understanding is vital for adapting current academic and industry models. Traditionally, academia has large data sets with low risks from which the model can learn; however, the opposite is true for DoD applications. Incorrect predictions can pose dire circumstances. Thus, off the shelf models are not usually appropriate for DoD needs. Matthew’s research will allow DoD researchers to understand the principal components in much larger, more expressive machine learning systems. Doing so will enable future design of more complex machine learning models without sacrificing interpretability and trustworthiness.
Matthew sought SEED Grant funding due to the novelty of his research. Currently, few papers exist on the subject matter. In contrast to other funding sources, SEED encourages the development of SMART scholars’ skills as principal investigators and exploratory research. Matt is currently collaborating with James Fairbanks, Ph.D., from the University of Florida and his student, Tyler Hanks. Fairbanks is an expert in computational category theory. Additionally, Tyler will intern at the Air Force Research Laboratory, Information Directorate to assist Matthew on his research,
“Framing Machine Learning Methods as Database Queries.”
The SMART Scholar SEED Grant Program is sponsored by the SMART Program Office and the Laboratories and Personnel Office under the Office of the Undersecretary of Defense for Research and Engineering. SEED Grant recipients receive research grants up to $100 thousand per year for up to a maximum of three years to help support promising SMART scholars establish a foundational research or engineering effort in their area of expertise as they transition from the pursuit of their Ph.D. to a DoD professional. To foster relationships between SEED Grant recipients and established members of the DoD technical workforce, mentors of SEED Grant recipients are eligible for an additional $25 thousand per year to support close engagement and collaboration with their SEED Grant mentee.
| Date Taken: | 04.13.2022 |
| Date Posted: | 04.13.2022 09:48 |
| Story ID: | 418404 |
| Location: | VA, US |
| Web Views: | 12 |
| Downloads: | 0 |
PUBLIC DOMAIN 
This work, Stepping into the Future with Machine, must comply with the restrictions shown on https://www.dvidshub.net/about/copyright.
"machine" - Google News
April 13, 2022 at 08:48PM
https://ift.tt/f6vL0xB
News - Stepping into the Future with Machine - DVIDS
"machine" - Google News
https://ift.tt/Xb9jsFg
https://ift.tt/WiBgN4O
Bagikan Berita Ini
0 Response to "News - Stepping into the Future with Machine - DVIDS"
Post a Comment