Introduction

Variational Quantum Circuits (VQCs) are quantum circuits with trainable parameters. Inspired by classical neural networks, they are the backbone of many quantum machine learning models, allowing quantum computers to learn from data. These circuits are especially valuable in the NISQ era, where fully fault-tolerant quantum computers are not yet available.

---

What Are VQCs?

A VQC consists of:

1. Parameterized quantum gates (e.g., RX, RY, RZ with variable angles).
2. Entangling layers (e.g., CNOT gates) that connect multiple qubits.
3. A cost function, typically based on the measurement outcome.
4. A classical optimizer that updates the parameters based on gradients or heuristics.

Together, this forms a hybrid quantum-classical feedback loop where the quantum circuit evaluates and the classical computer updates parameters.

---

VQCs vs Classical Models

Feature	Classical ML	VQCs
Parameters	Weights/Biases	Rotation angles (θ)
Computation Space	Real-valued space	Complex Hilbert space
Hardware	CPUs/GPUs	Quantum processors (QPU)
Optimization	Gradient descent, etc.	Same (with quantum-aware tweaks)

---

Structure of a Variational Quantum Circuit

Visual: Structure of a Variational Quantum Circuit

        Classical Data (x)
               │
               ▼
     ┌─────────────────────┐
     │  Data Encoding Layer│ ◄─── Embed classical data into qubit states           (e.g., AngleEmbedding)
     └─────────────────────┘
               │
               ▼
     ┌─────────────────────┐
     │ Variational Layers  │ ◄─── Parameterized quantum gates (RX, RY, RZ)
     │   + Entanglement    │      and entangling gates (CNOTs)
     └─────────────────────┘
               │
               ▼
     ┌─────────────────────┐
     │    Measurement      │ ◄─── Readout: e.g., ⟨Z⟩ expectation value
     └─────────────────────┘
               │
               ▼
     ┌─────────────────────┐
     │   Cost Function     │ ◄─── Compare output with target (e.g., MSE, cross-entropy)
     └─────────────────────┘
               │
               ▼
     ┌─────────────────────┐
     │ Classical Optimizer │ ◄─── Update weights (θ) to reduce error
     └─────────────────────┘
               │
               └──── Feedback Loop ────► Repeat

Flow Description:

1. Classical Data (x): Start with input features.
2. Data Encoding Layer: Encode classical data into quantum states.
3. Variational Layers + Entanglement: Apply parameterized quantum gates and entangling operations.
4. Measurement: Observe qubits to gather quantum outcomes.
5. Cost Function: Calculate performance based on task.
6. Classical Optimizer: Update parameters based on feedback.
7. Repeat: Iterate until the model converges.

---

VQC Code Example (PennyLane)

import pennylane as qml
from pennylane import numpy as np

dev = qml.device("default.qubit", wires=2)

@qml.qnode(dev)
def circuit(weights, x=None):
    qml.templates.AngleEmbedding(x, wires=[0,1])
    qml.templates.StronglyEntanglingLayers(weights, wires=[0,1])
    return qml.expval(qml.PauliZ(0))

weights = np.random.randn(3, 2, 3)
x = [0.5, -0.3]
print(circuit(weights, x=x))

---

Applications of VQCs in Quantum AI

• Quantum classifiers (binary/multiclass)
• Quantum generative models
• Anomaly detection
• Quantum reinforcement learning
• Quantum kernel methods (via variational distance)

---

Strengths & Challenges

✅ Benefits

• Trainable: Adapts to data like neural networks.
• Hardware-friendly: Runs on today’s NISQ devices.
• Hybrid-ready: Easily integrates with classical systems.

⚠️ Limitations

• Optimization is prone to barren plateaus (vanishing gradients).
• Performance depends on circuit architecture and data encoding.
• Requires many iterations and good initialization.

---

🔗 What’s Next?

➡️ Next: Quantum k-Nearest Neighbors (Quantum k-NN)