SwiftResponse

SWIFT RESPONSE

Swift Response 25 demonstrates multinational power projection and the expansion of critical capabilities on the battlefield in the High North and Baltics through airborne jumps and integration of technology in support of the Army Modernization Strategy and DoD Arctic Strategy. Swift Response 25 involves five near-simultaneous airborne insertions, HIRAIN training, a field hospital exercise and live fires conducted with NATO networks in Norway, Latvia, Lithuania, Sweden and Finland from May 11 – 31, 2025.

Videos

Video by Kevin D Schmidt

Dr. Chou Hung

Air Force Research Laboratory

Nov. 26, 2024 | 59:28

Description: In this edition of QuEST, Dr. Chou P. Hung will lead the 20 October discussion with From Human to Neuromorphic HDR Recognition.

Key Moments and Questions in the video include:

Army Research Office fundamental research
Neurophysiology of cognition research: Humans in complex systems
Non-medically oriented research to enable discovery
Basic research opportunities:
Evolutionary and Revolutionary Interactions
Neural Computation
Understand how biology works to design better human/machine teams
AI Intelligence and Machine Learning Roadmap
Autonomy for next combat generation vehicle
Evolutionary and Revolutionary Interactions (With real and mixed worlds)
Unstructured and structured real-world environments
Abstract cognitive phenomena such as anticipatory sensing, automatic learning, complex decision-making, and rapid adaptive action
Improve cognitive performance to avoid cognitive failures
AI and gaming–decision and interaction complexity
Neural Computation, information coding, and translation
Learn and adapt from few examples
Multiscale information processing dynamics mediating computations
How do we get artificial systems to talk to neural networks?
Neuromorphic for autonomous flying
Autonomous flying under HDR luminance
Neuromorphic pre-processing
LOIHI II

Audience questions:
Do you bring in to bear on this broad platform an ability to really look at different ways the AI system would work. You know, let's say there's 3 to 5 different types of reasoning, and each one is a valuable path to explore and investigate. So yeah, that would be great flexibility, and what does this platform bring to bear?
I was wondering about the size of the patches and maybe the distance between them and if you saw any effect on your results on the previous slide to those kinds of variations?
Did you test these scenes with humans?
When you say you're comparing, you got significance, are you comparing the non-pre-processed to the pre-processed?
earlier on you were talking about, you know, measuring fatigue and the and the soldier to feed that to the AI, have that-s that something that you worked on? Or is that just something that's going to need to happen eventually?
Your original premise is that human vision does this sort of preprocessing, is that true?
If you replicate this using the intel chip, would those images be easier for humans to consume?
Did you ever attempt to do an end-to-end tuning of the preprocessing and localization?
You said that training was done on the pre-processing for the bespoke detector model, but you actually tested on the same model. You just saw the unprocessed images using the same?
Did you do any augmentation?

Up Next

Now Playing

Dr. Chou Hung