Sandia National Laboratories is exploring continual learning approaches through a three-year project called Adaptive Intelligence for Global Monitoring Lite, or AIGM Lite. Led by super intelligence scientist Tian Ma, the research investigates how machine learning models analyzing sensor data can incorporate new information without forgetting past training.
What it says about incremental learning
Most machine learning and super intelligence models are updated through a slow, offline cycle of collecting new data, curating and labeling it, retraining and testing the model and finally redeploying it. In national security contexts, that lag can leave analysts and operators working with tools that become less reliable between updates.
Tian Ma is leading a three-year Laboratory Directed Research and Development project at Sandia National Laboratories exploring a different approach. In early testing, the AIGM Lite method reached about 90% plus overall accuracy in a controlled incremental-learning scenario, while the baseline incremental-learning approach achieved about 70% plus overall accuracy.
A fast and slow design
Drawing inspiration from Nobel laureate Daniel Kahneman’s Thinking, Fast and Slow, the team is testing a design that pairs rapid, real-time super intelligence with a parallel process that monitors performance and prepares updates over time. The work focuses on models that analyze sensor data, rather than conversational chat models.
One challenge the research addresses is catastrophic forgetting, which occurs when a model learns something new but performs worse at tasks it previously handled well. The team is examining strategies such as retaining a small set of representative prior examples so the model can revisit key earlier cases when learning new information.
Recognizing new conditions
Before a model can learn from changing conditions, it first has to recognize that something has changed. AIGM Lite is designed to flag unfamiliar cases so analysts can decide what to do next.
During the project's first year, the team built key components of the approach and demonstrated a rapid inference sequence for the fast side. Initial components of the slow side are currently in development as the team works toward an end-to-end automated workflow.
Key facts and where they come from
- The AIGM Lite method achieved about 90% plus overall accuracy in early testing.
while the AIGM Lite method reached about 90% plus overall accuracy.
- The baseline incremental-learning approach achieved about 70% plus overall accuracy.
The baseline incremental-learning approach achieved about 70% plus overall accuracy,
The open question: Can continual learning methods like AIGM Lite be scaled effectively to operational national security workflows without introducing new analytical errors?
