arXiv:2610.10756 · Quantum Science and Technology
Quantum-Enhanced Inference of Conditional Future Probabilities with Reduced Memory Cost
Jianjun Chen, Akshat Gupta, Chengran Yang, Ariel Neufeld, Jayne Thompson, Thomas J. Elliott, Mile Gu
Abstract
Many potential future events have devastating consequences, from the onset of earthquakes to financial black swan events. Accurate predictions of their probability of occurrence are vital. Yet as the stochastic systems underlying such events become increasingly complex, these probabilities can depend on ever-growing amounts of past data, which require ever more memory to track. The accuracy of our estimates is ultimately limited by both computational and memory constraints; the former to reduce sampling error, the latter to avoid sampling bias. Traditional quantum algorithms -- such as amplitude estimation -- address only the former, and become untenable especially in near-future quantum devices where quantum memory is a premium. Here, we offer quantum-algorithmic means to reduce systematic bias while maintaining a quadratic speed-up. Our results thus advance the timeline for the practical deployment of quantum technologies, and illustrate their additional benefits when simulating processes of growing complexity.
AI Summary
The paper asks whether conditional future-event probabilities in stochastic processes can be estimated with both quantum amplitude estimation’s sampling advantage and lower bias under a fixed memory constraint. It combines existing quantum predictive-model compression with amplitude estimation: a classical model is converted into a quantum simulator with nonorthogonal memory states, optionally compressed through variational infinite-matrix-product-state approximation, and run coherently for $L$ steps. Amplitude estimation accommodates the residual memory–output entanglement by including both registers in state preparation, inversion, and the zero-state reflection. The resulting estimator retains $O(1/\epsilon)$ state-preparation queries for additive sampling error $\epsilon$, separate from model bias. Generic state-preparation synthesis costs $O(L\tilde D^2|\mathcal X|^2)$ CNOTs for compressed memory dimension $\tilde D$. Numerical demonstrations compare compressed quantum models with classical causal-state clusterings optimized for KL-divergence rate. For eight-state ring diffusion compressed to four memory states, the average RMSE over 36 events with probabilities $10^{-9}$–$10^{-8}$ falls below the classical model’s approximately $10^{-6}$ bias floor after roughly $2\times10^4$ queries. A sixteen-state example crosses its classical baseline after approximately 3,000 queries. The demonstrations use small models, and compression introduces irreducible bias. Existing bounds control fidelity divergence rate, rather than directly bounding conditional-event error; bounded entropic quantum memory does not imply bounded exact memory dimension. Compression preprocessing depends on convergence and spectral properties, while quoted overall gate scaling assumes an inexpensive event oracle. A noise simulation additionally tests reduced gate-error accumulation, but does not demonstrate noisy amplitude estimation.
Central claim may not hold: The cyclic-walk figures label the error of a KL-rate-optimal causal-state clustering as a classical lower bound without establishing optimality for the tested conditional events or over all memory-constrained classical models. The target-probability additive-error complexity omits compression bias, although QAE's O(1/epsilon) guarantee directly concerns the compressed model's probability. The continuum-limit fidelity-divergence-rate bound is not converted into a guarantee for finite-horizon probabilities conditioned on a particular past. The supplementary AR(2) derivation replaces continuous past values with their signs and therefore does not represent the claimed sign-thresholded Gaussian AR(2) process. The printed reduced classical cyclic-walk transition formula sums two normalized transition distributions without averaging weights, producing total outgoing probability 2 rather than 1.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.