quantum-enhanced sampling
**Quantum-Enhanced Sampling** refers to the use of quantum computing techniques to accelerate sampling from complex probability distributions, leveraging quantum phenomena—superposition, entanglement, tunneling, and interference—to explore energy landscapes and probability spaces more efficiently than classical Markov chain Monte Carlo (MCMC) or other sampling methods. Quantum-enhanced sampling aims to overcome the slow mixing and mode-trapping problems that plague classical samplers.
**Why Quantum-Enhanced Sampling Matters in AI/ML:**
Quantum-enhanced sampling addresses the **fundamental bottleneck of classical MCMC**—slow mixing in multimodal distributions and rugged energy landscapes—potentially providing polynomial or exponential speedups for Bayesian inference, generative modeling, and optimization problems central to machine learning.
• **Quantum annealing** — D-Wave quantum annealers sample from the ground state of Ising models by slowly transitioning from a transverse-field Hamiltonian (easy ground state) to a problem Hamiltonian; quantum tunneling allows traversal of energy barriers that trap classical simulated annealing
• **Quantum walk sampling** — Quantum walks on graphs mix faster than classical random walks for certain graph structures, achieving quadratic speedups in mixing time; this accelerates sampling from Gibbs distributions and Markov random fields
• **Variational quantum sampling** — Parameterized quantum circuits trained to approximate target distributions (Born machines) can generate independent samples without the autocorrelation issues of MCMC chains, potentially providing faster effective sampling rates
• **Quantum Metropolis algorithm** — A quantum generalization of Metropolis-Hastings that proposes moves using quantum operations, accepting/rejecting based on quantum phase estimation of energy differences; provides sampling from thermal states of quantum Hamiltonians
• **Quantum-inspired classical methods** — Tensor network methods and quantum-inspired MCMC algorithms (simulated quantum annealing, population annealing) bring some quantum sampling benefits to classical hardware, improving mixing in multimodal distributions
| Method | Platform | Advantage Over Classical | Best Application |
|--------|---------|------------------------|-----------------|
| Quantum Annealing | D-Wave | Tunneling through barriers | Combinatorial optimization |
| Quantum Walk Sampling | Gate-based | Quadratic mixing speedup | Graph-structured distributions |
| Born Machine Sampling | Gate-based | No autocorrelation | Independent sample generation |
| Quantum Metropolis | Gate-based | Quantum thermal states | Quantum simulation |
| Quantum-Inspired TN | Classical | Improved mixing | Multimodal distributions |
| Simulated QA | Classical | Better barrier crossing | Rugged landscapes |
**Quantum-enhanced sampling leverages quantum mechanical phenomena to overcome the fundamental limitations of classical sampling methods, offering faster mixing through quantum tunneling and interference, autocorrelation-free sampling through Born machines, and quadratic speedups through quantum walks, with broad implications for Bayesian ML, generative modeling, and combinatorial optimization.**