• Proposes a multi-learner architecture that swaps specialists in and out as the data distribution shifts.
• Demonstrates strong performance on streaming classification tasks while providing calibrated uncertainty estimates.