CMG Gating Mechanism Significantly Improves Quantum Dynamics Forecasting

Kuo-Chung Peng · hf · 2026-08-11

To address the high cost of processing long contexts in quantum-inspired sequence learning, researchers introduced Self-Modulating QKAN-based Fast Weight Programmers (FWPs).

Traditional scalar gating forces all parameters to share a single memory timescale. The proposed Complementary Matrix Gating (CMG) replaces this by using one sigmoid matrix gate to retain the old state and its complement to write the new proposal. This enables coordinate-wise memory control while preserving the bounded convex update and affine prefix-scan structure.

In multi-step forecasting of Jaynes-Cummings and transmon-resonator dynamics, CMG models maintained mean-squared errors (MSE) of 0.001 or lower across 4, 8, and 16-step horizons, improving upon scalar-gated counterparts by at least 91.2%.

Original post →

More from Research

Research channel →