MoSAIC: Scalable Probabilistic Error Cancellation via Variational Blockwise Noise Aggregation

We present MoSAIC(Modular Spatio-temporal Aggregation for Inverted Channels), a scalable error-mitigation framework that suppresses correlated noise in quantum circuits by combining circuit partitioning with Probabilistic Error Cancellation (PEC). Instead of applying PEC after every circuit layer, MoSAIC performs mitigation on variationally characterized multi-qubit blocks, enabling accurate recovery of local noise channels with substantially reduced sampling overhead. To further improve efficiency, we estimate the control-variate coefficient from pilot runs to determine the required number of samples.

We perform extensive numerical simulations across a variety of circuits, including random circuits, fixed-point Grover search, 2D Ising model, and 1D Transverse-Field Ising Model(TFIM) for systems ranging from 4 to 20 qubits. Results show that MoSAIC achieves the same precision as standard PEC while requiring less than half the number of samples for small-scale circuits. In particular, simulations on the 1D TFIM demonstrate that MoSAIC maintains strong mitigation performance, achieving final observables within 4% accuracy for systems up to 20 qubits, more than twice the standard PEC experimentally demonstrated. These results indicate that MoSAIC can be effectively extended to larger systems on real quantum hardware. We are currently implementing MoSAIC on IBM Heron processor, targeting systems of up to 100 qubits, where we expect similarly robust error-mitigation performance.