Complementary Matrix-Gated QKAN Fast-Weight Programmers for Quantum Dynamics Forecasting
CMG lets each fast-weight coordinate keep or overwrite memory on its own schedule.
The paper replaces scalar gating in QKAN fast-weight programmers with low-rank, element-wise modulation. Its Complementary Matrix Gating uses one sigmoid matrix to retain old state and the complement to write the new proposal, keeping the update bounded and prefix-scan friendly. Across the reported benchmarks, CMG is the most consistent gain when the fast programmer includes a QKAN module. In multi-step Jaynes-Cummings and transmon-resonator forecasts, CMG keeps MSE at about 0.001 or below and improves on scalar-gated versions by at least 91.2%. HF Daily Papers' note
The paper replaces scalar gating in QKAN fast-weight programmers with low-rank, element-wise modulation. Its Complementary Matrix Gating uses one sigmoid matrix to retain old state and the complement to write the new proposal, keeping the update bounded and prefix-scan friendly. Across the reported benchmarks, CMG is the most consistent gain when the fast programmer includes a QKAN module. In multi-step Jaynes-Cummings and transmon-resonator forecasts, CMG keeps MSE at about 0.001 or below and improves on scalar-gated versions by at least 91.2%. HF Daily Papers' note
score 3