Stable Self-Modulating Quantum Fast-Weight Programmers with Bounded Memory Gates
📄 arXiv:2607.02363 · 📥 PDF · 2026-07-02 · quant-ph
Authors: Kuo-Chung Peng [arXiv · scholar] , Jiun-Cheng Jiang [arXiv · scholar] , Chun-Hua Lin [arXiv · scholar] , Yifeng Peng [arXiv · scholar] , Junghoon Justin Park [arXiv · scholar] , Huan-Hsin Tseng [arXiv · scholar] , Hsin-Yi Lin [arXiv · scholar] , Kuan-Cheng Chen [arXiv · scholar] , Chen-Yu Liu [arXiv · scholar] , Shinjae Yoo [arXiv · scholar] , Samuel Yen-Chi Chen [arXiv · scholar]
🕰 Orloj analysis
Paper navrhuje stabilizační strategii pro Self-Modulating Quantum Fast-Weight Programmers (QFWPs) řešící divergenci v dlouhých sekvencích. Zavádí omezenou modulační bránu pro starý stav paměti, čímž zlepšuje robustnost a konzistenci algoritmu.
💡 Jde o cenné, inkrementální vylepšení existujícího kvantového algoritmu pro sekvenční modelování, které zvyšuje jeho praktickou použitelnost a stabilitu.
✓ falsifiable, modest_claims
📄 Abstract
Quantum Fast-Weight Programmers (QFWPs) store temporal information in dynamically programmed variational-circuit parameters rather than in nonlinear recurrent hidden states, offering a practical route to quantum sequence modeling. Self-Modulating QFWP improves this framework by using input-dependent gates for both new fast-weight updates and the accumulated fast-weight state, but its unbounded old-state multiplier can diverge in long-sequence regimes. We propose a bounded old-state modulation rule that applies a sign-preserving tanh gate only to the recurrent memory branch while leaving the additive update and new-update modulation unchanged. We evaluate standard QFWP, full Self-Modulating QFWP, Only-New, and Only-Old variants on two CUDA-Q quantum-dynamics forecasting tasks and on Milan SMS telecommunication activity prediction. The quantum-dynamics results show that old-state modulation is the most consistent source of improvement over Standard QFWP, and that bounding the old-state gate removes long-sequence divergence while improving aggregate robustness. On Milan SMS forecasting, the original unbounded Self-Modulating QFWP converges across the tested grid and shows its clearest gains at longer input windows, with behavior close to the Only-Old ablation. These findings identify accumulated-memory modulation as the key mechanism of Self-Modulating QFWP and bounded old-state gating as a targeted stabilization strategy.