DARPA CriticalMAAS Project
Artificial Intelligence for Critical Mineral Assessment Competition
The Defense Advanced Research Projects Agency (DARPA) and the United States Geological Survey (USGS) have joined hands in a groundbreaking endeavor to harness the potential of Machine Learning (ML) and Artificial Intelligence (AI) for bolstering the efficiency of critical mineral assessments.
Note: The competition has concluded. DARPA's collaboration with USGS is still active under the Critical Mineral Assessments with AI Support (CriticalMAAS) project.
For more details, visit sam.gov.
About Our Effort
Mapping advanced argillic alteration at Cuprite, Nevada using imaging spectroscopy.
Read the full publication
The US's manufacturing and national security sectors heavily rely on a diverse range of non-fuel raw materials termed as "critical minerals." Both the Energy Act of 2020 and the Bipartisan Infrastructure Law have directed the USGS to carry out comprehensive assessments of these vital resources. These studies aim to pinpoint potential mineral resources from existing mines – both historical and active, offering insights into opportunities for sustainable resource development. Given the urgency of addressing modern-day supply chain needs, traditional methods are time-inefficient. This collaboration between DARPA and USGS is an attempt to supercharge these assessments using ML and AI.
Repository Status: Ongoing development. Current files primarily consist of scripts for creating and training models focused on map segmentation.
Repo Contents
| File Name | Description |
|---|---|
GAN_model.py |
Training GAN model for POC |
README.md |
Documentation of the project |
VAE-unet.py |
Create and train VAE U-net model for map segmentation |
VAE-unet_updated.py |
Create and train VAE U-net model for map segmentation with updated loss function |
create_prediction_map.py |
Data loader class script for input processing |
data_util.py |
Utility functions for dataloader that read and process data from TIFF and JSON files |
eval_gan.py |
Evaluation script for GAN model |
eval_vauner.py |
Evaluation script for VAE-Unet model |
train_model.py |
Training script for the attention model with processed legends |
train_model_u_legends.py |
Training script for the attention model with unprocess legends |
unet_util.py |
Utility functions for the attention U-net model |
For more information on specific files or collaboration, please refer to the contribution guidelines or reach out to the repository maintainers.
To-Do
- Evaluate VAE Unet model
- Incorporate intermediate input data
- Test color jittering for legends
- Legend augmentation
Reference: Link to Paper