Wildfire Response

Latest News

National Guard Members Continue LA Wildfire Response
January 21, 2025 — LOS ANGELES – U.S. Army Sgt. Ricardo Hernandez watched from a cross street as sporadic traffic... MORE

U.S. Army Sgt. Bryce Carter, an infantryman with C Company, 1st Battalion, 160th Infantry Regiment, California Army National Guard, sharpens the blade of a hoe to clear brush and other debris as part of remediation efforts along the Mulholland Trail near Tarzana, California, in the aftermath of the Palisades Fire, Jan. 18, 2025. Carter and other members of his unit were assisting CALFIRE in mop-up efforts, which included clearing brush and backfilling firebreaks and other areas to prevent mudslides and reduce the impact of firefighting efforts.

California Guardsman Helps Battle Wildfires in His Community
January 16, 2025 — CHANNEL ISLANDS AIR NATIONAL GUARD STATION, Calif. - Amid the wildfire crisis in California, an Air... MORE

Master Sgt. Alan Franklin, a commander's support Airman with the 146th Airlift Wing, speaks to 1st Lt. Aiden Flores about the Modular Airborne Fire Fighting System mission on the flightline at Channel Islands Air National Guard Station, Port Hueneme, California, Jan. 13, 2025. MAFFS aircraft from the Air National Guard’s 153rd Airlift Wing, Cheyenne, Wyoming, the 152nd Airlift Wing, Reno, Nevada, the 146th Airlift Wing, Port Hueneme, California, and Air Force Reserve Command’s 302 AW, Peterson Air Force Base, Colorado, are working together to combat fires in the Los Angeles area.

National Guard Bureau Chief Thanks Firefighting Guardsmen
January 14, 2025 — CHANNEL ISLANDS AIR NATIONAL GUARD STATION, Calif. – The National Guard is part of a multi-agency... MORE

Air Force Gen. Steve Nordhaus, chief, National Guard Bureau, and Army Senior Enlisted Advisor John Raines, SEA to the CNGB, visit National Guard members supporting wildland firefighting in Southern California, Channel Islands Air National Guard Station, Calif., Jan. 11, 2025. Thousands of National Guardsmen are involved in multiple air and ground firefighting in the Los Angeles Basin and Southern California.

Wyoming, Nevada Guard Aircrews Assist California Firefighters
January 13, 2025 — CHEYENNE, Wyo. – Three Wyoming Air National Guard C-130s equipped with the U.S. Forest Service’s... MORE

U.S. Air Force Airmen assigned to the 153rd Airlift Wing load and install the Modular Airborne Fire Fighting Systems onto a C-130H Hercules aircraft in Cheyenne, Wyoming, Jan. 10, 2025, in preparation to support firefighting efforts in the Los Angeles area.

California, Nevada, Wyoming Guard Join Firefighting Battle
January 10, 2025 — MOFFETT AIR NATIONAL GUARD BASE, Calif. - Hundreds of National Guard members are now involved in... MORE

U.S. Air Force Airmen with the 129th Rescue Wing, California Air National Guard, at Moffett Air National Guard Base, Calif., prepare an HH-60G Pave Hawk helicopter to help battle the Palisades Fire Jan. 9, 2025.

Latest Photos

Latest Videos

Video by Kenneth M McNulty, Kevin D Schmidt

Michael Robinson - Topological Features in Large Language Models (and beyond?)

Air Force Research Laboratory

Oct. 11, 2024 | 01:00:53

In this edition of QuEST, Michael Robinson will discuss topological features in large language models

Key Moments and Questions in the video include:
Acknowledgement of colleagues from DARPA and Galois
Manifolds in machine learning
LLM token space is higher dimensional
Manifold spaces tend to be negatively curved
LLM turn text into vectors
Transformers turn vectors into new text
How do we turn the text into vectors?
We think of LLM as being trained on all human language, but they have not
GPT2 Open source LLM as the source for model
ChatGPT2 used as the example
Tokens have topology and geometry
Words are a categorical variable
Vectors are a numerical variable
Mixing data types can lead to some problems
Why care about the token space?
Not all tokens correspond to a valid vector
Estimating dimensions
Volume of a sphere
Log of Volume vs log of radius curves
Ricci scalar curvature
Stratifications are visible
GPT2 uses a state space that is not a manifold
Dollar sign shown different in GPT2 because the $ is used in code where other currency symbols are not
GPT2’s 768 dimensions unwrapped using tSNE
Tokens with leading spaces
Beginnings of words show up in separate piece of low dimension
Visual similarity to hyperbolic plane
LLEMMA7B dimensions
Plotting dimension
Dark space are non-printing characters
Thinking about how neural activation patterns work
We have been thinking about manifold learning out of mathematical convenience
State spaces are not manifolds
Open presentation to conversation