Page 14 - Demo

The GAM-powered source reconstruction uses batch aggregation to maximize accuracy
• Batch aggregatation GAM (bGAM) method:
• Data is separated into randomized
subsets of data, or “batches”
• A GAM is then trained on each batch independently
• The results are then averaged together, weighted by the relative error
• Testing data is held separately from training batches to determine error
• The reconstruction is highly accurate, as the batch aggregation serves to “average away” systematic and experimental errors
Presenter: Gabriel Player
14