Hardware-Trained Quantum Eigensolver for 70-Qubit Systems
November 11, 2026
Transformers are trained to generate state-preparation circuits for sample-based quantum diagonalization using the IBM Heron processor. The training includes a reward penalty to prevent collapse and achieves results on organotin photoresist models up to 70 qubits.
HOW THIS AFFECTS YOU
●
researcherYou can train generative policies directly on quantum hardware to reach regimes beyond classical simulation.
●
healthThis method could eventually help model complex molecules used in specialized manufacturing and medicine.