Source-linked AI summary
Machine Learning for Precise Quantum Measurement
Alexander Hentschel, Barry C. Sanders
TL;DR
Single-shot adaptive interferometric phase estimation can use an optimal but naturally hard-to-prepare entangled input state. The paper applies particle swarm optimization to learn adaptive feedback policies and finds photon-loss-robust quantum-enhanced measurements.
Problem
The minimum uncertainty state is optimal for single-shot adaptive interferometric phase estimation but is naturally hard to prepare because of its entanglement.
Method
Particle swarm optimization autonomously learns adaptive feedback strategies by optimizing policies represented as particle positions and their best previously visited positions.
Results
At a photon-loss rate of η = 40%, PSO-generated policies achieve variance scaling as N ∝N −1.307±0.009 in the quantum-enhanced measurement domain.
Takeaways & Limitations
The PSO can learn a good adaptive strategy for alternative, more practical input states and remains applicable to moderately noisy experiments.
Abstract
from arXiv · showhide
Adaptive feedback schemes are promising for quantum-enhanced measurements yet are complicated to design. Machine learning can autonomously generate algorithms in a classical setting. Here we adapt machine learning for quantum information and use our framework to generate autonomous adaptive feedback schemes for quantum measurement. In particular our approach replaces guesswork in quantum measurement by a logical, fully-automatic, programmable routine. We show that our method yields schemes that outperform the best known adaptive scheme for interferometric phase estimation.
A. Interferometer Description
The Mach–Zehnder interferometer combines two input modes, applies phase shifts, and uses a second beam splitter chosen to recover the input when both arm phases match.
- The Mach–Zehnder interferometer uses two input modes and two output modes, with phase shifts applied in its arms.
- The second beam splitter B2 is selected to recover the input when the controllable and unknown phases are equal.
- The interferometer operator is expressed using the beam-splitter transformations and the relative phase parameter θ.
B. Input States
The paper uses permutation-symmetric minimum uncertainty states as input states for adaptive phase estimation, while noting that their entanglement makes preparation difficult.
- Minimum uncertainty states are symmetric under qubit permutations and reduce the Holevo variance of phase estimates.
- These states are represented using two-mode Fock states |n_a, n_b⟩ with N = n_a + n_b.
- The minimum uncertainty state is optimal for single-shot adaptive interferometric phase estimation but is naturally difficult to prepare because of its entanglement.
- The study uses the same input state as the BWB policy for fair comparison, while allowing other practical states for autonomous policy learning.
C. Feedback Technique
The feedback technique updates the interferometer’s phase difference after each detection, while the remaining entangled input state depends on the measurement record and unknown phase.
- The phase parameter θ changes during the experiment because the feedback phase Φ is varied after detections.
- The feedback phase Φ_m is applied after the mth detection and determines the phase difference used for the next input particle.
- After m photodetections, the remaining input is represented by a state |ψ(n_m, ϕ)⟩ conditioned on the record and phase.
- The Kraus operator represents measuring the kth particle with outcome u_k, and the resulting unnormalized state norm gives the probability of record n_m given ϕ.
D. Performance Measure For Policies
Policy performance is evaluated through the sharpness derived from the phase-estimation error distribution and measurement-record probabilities, but its exact computation scales exponentially with N.
- Sharpness is computed from the probability distribution of the policy’s phase-estimation error.
- The measurement-record probability P_ρ(n_N|ϕ) is evaluated under a flat prior P(ϕ) = 1/(2π).
- The sharpness is obtained from the probability distribution using equation (1).
- O(2^N) computation is required because the sharpness sum evaluates every length-N measurement bit string.
E. Optimization Problem
The optimization problem searches the policy space for the policy with maximum sharpness.
- The goal is to find a policy ρmax whose sharpness is at least that of every policy ρ in the policy space P.
F. Details of the employed PSO algorithm
The employed PSO algorithm searches quantum-measurement policy space using interacting particles that evaluate sharpness, share local optima, and update their positions with exploration and exploitation controls. Its settings and learned best policies are specified for photon numbers up to N = 14.
- Particle representation: PSO represents each particle by a policy-space position and its best previously visited position.The swarm contains Ξ particles, each with position ρ(i) ∈ P and personal best ˆρ(i).
- Optimization cycle: Each PSO round evaluates particle sharpness, updates personal bests, communicates neighborhood optima, and changes particle positions.The particle-update equation is applied after these assessment and communication steps.
- Neighborhood communication: Particles communicate their best policies within neighborhoods and select the sharpest neighborhood policy as g(i).Neighborhood relations are arranged using a predefined ring topology with interaction range r.
- Update controls: The update dynamics use damping, exploitation, exploration, random sampling, and a bounded maximum step size.ω is the velocity damping factor; ϕ1 and ϕ2 control attraction to personal and neighborhood best positions, respectively.
- Reported settings: For N = 1, . . . , 14, the authors report successful PSO settings and the best learned policy parameters, while larger N requires larger populations.The settings include swarm size, PSO steps, maximum step size, weights, and interaction range.
G. Noise resistance
The PSO-generated policies are trained on a noiseless interferometer but remain applicable under moderate noise and photon loss. Their robustness is attributed mainly to the highly entangled, permutation-symmetric input state.
- Training assumption: The PSO algorithm is trained using a simulated noiseless Mach-Zehnder interferometer, as is the BWB-policy.The authors relate this modeling choice to experimental evidence that the considered feedback policies are robust against noise.
- Noise applicability: The resulting policies are applicable to moderately noisy experiments despite noisefree training.
- Photon-loss performance: At η = 40% photon loss, the variance scales as N^-1.307±0.009 and remains within the domain of quantum-enhanced measurements.
- Mechanism: Robustness against photon loss is mainly attributed to an input state that is highly entangled and symmetric under qubit permutations.The state remains entangled even when a high percentage of photons are lost.