Quantum Circuit Learning - Mitarai et al. 2018

Overview

Reproduction of the Quantum Circuit Learning paper - one of the foundational works in variational quantum machine learning, showing that quantum circuits can be trained as universal function approximators.

Property	Value
Category	Research
Difficulty	Advanced
Framework	PennyLane
Qubits	4
Paper	Physical Review A 98, 032309 (2018)

Key Contributions

This paper established:

Gradient computation via parameter-shift rule
Universal approximation with quantum circuits
Training framework for variational quantum models

The QCL Architecture

Input Encoding

CODE
|0⟩ → RY(arctan(x_i)) → ...

Each feature maps to a qubit rotation.

Trainable Layer

CODE
┌──────┐┌──────┐
┤RY(θ₀)├┤RZ(φ₀)├──●───────
├──────┤├──────┤┌─┴─┐
┤RY(θ₁)├┤RZ(φ₁)├┤ X ├──●──
├──────┤├──────┤└───┘┌─┴─┐
┤RY(θ₂)├┤RZ(φ₂)├─────┤ X ├
└──────┘└──────┘     └───┘

Output

Expectation value ⟨Z⟩ on first qubit.

Running the Circuit

PYTHON
from circuit import run_circuit

# Classification task
result = run_circuit(
    task='classification',
    n_qubits=4,
    n_layers=2,
    n_train=30,
    n_epochs=50
)

print(f"Accuracy: {result['classification']['train_accuracy']:.1%}")

# Function approximation
result = run_circuit(
    task='function',
    n_qubits=4,
    n_layers=3,
    n_train=30,
    n_epochs=100
)

print(f"MSE: {result['function_approximation']['mse']:.4f}")

Parameter-Shift Rule

The gradient of a gate with parameter θ:

CODE
∂⟨O⟩/∂θ = (⟨O⟩_{θ+π/2} - ⟨O⟩_{θ-π/2}) / 2

This enables:

Exact gradients (no approximation)
Hardware-compatible training
Automatic differentiation in simulators

Expected Results

Classification (Synthetic Data)

Layers	Parameters	Accuracy
1	8	~70%
2	16	~85%
3	24	~95%

Function Approximation (sin(x))

Layers	MSE
1	~0.3
2	~0.1
3	~0.02

Universal Approximation

The paper proves that with sufficient:

Qubits: Encode input dimensionality
Layers: Control approximation accuracy
Entanglement: Capture correlations

QCL circuits can approximate any continuous function.

Comparison with Classical

Aspect	QCL	Neural Network
Parameters	O(nL)	O(n²L)
Gradient	Exact	Backprop
Expressivity	Universal	Universal
Training	Harder	Easier

Implementation Notes

Encoding Strategy

PYTHON
# Paper uses arctan encoding
RY(arctan(x))

# Maps x ∈ (-∞, ∞) to angle ∈ (-π/2, π/2)

Ring Entanglement

PYTHON
for i in range(n_qubits):
    CNOT(i, (i+1) % n_qubits)

Creates periodic boundary conditions.

Historical Impact

This paper:

Introduced parameter-shift gradient rule
Proved quantum circuits as function approximators
Inspired PennyLane and other QML frameworks
Foundation for NISQ machine learning

Paper Citation

BIBTEX
@article{mitarai2018quantum,
  title={Quantum circuit learning},
  author={Mitarai, Kosuke and others},
  journal={Physical Review A},
  volume={98},
  pages={032309},
  year={2018}
}

Quantum Circuit Learning (Mitarai 2018)