nisqoptml 0.1.1

NISQ-Optimized Machine Learning: Noise-resilient QML library with Federated and DP Support

NISQOptML

NISQ-Optimized Machine Learning: A noise-resilient Quantum Machine Learning library with Federated Learning and Differential Privacy support.

Overview

This library provides a framework for building and training quantum neural networks on NISQ (Noisy Intermediate-Scale Quantum) devices. It addresses challenges in quantum machine learning including noise mitigation, distributed execution, and privacy-preserving federated learning.

Key Features

• Quantum Neural Networks (QNN) - Build and train variational quantum circuits
• Noise Mitigation - Automatic error mitigation and zero-noise extrapolation techniques
• Federated Learning - Train models across distributed nodes with FedAvg
• Differential Privacy - Add Gaussian noise for privacy-preserving training
• Explainability - Sensitivity analysis and parameter visualization tools
• Distributed Execution - MPI-based distributed quantum circuit execution

Installation

pip install nisqoptml

Quick Start

Basic QNN Training

import nisqoptml as nq
import numpy as np

# Create a QNN model
model = nq.QNN(layers=2, qubits=4)

# Prepare training data
X = np.random.rand(10, 4)
y = np.random.rand(10, 4)

# Train the model
model.fit(X, y, epochs=10)

With Error Mitigation

# Enable automatic error mitigation
model = nq.QNN(layers=2, qubits=4, mitigation='auto')
model.fit(X, y, epochs=10)

# Or use zero-noise extrapolation
model = nq.QNN(layers=2, qubits=4, mitigation='zne')
model.fit(X, y, epochs=10)

Federated Learning with Differential Privacy

# Create federated QNN with DP noise
model = nq.QNN(
    layers=2, 
    qubits=4, 
    federated=True, 
    dp_sigma=0.01,  # DP noise standard deviation
    mitigation='auto'
)

# Train with federated learning
local_X = np.random.rand(5, 4)
local_y = np.random.rand(5, 4)
model.federated_fit(local_X, local_y, rounds=5, local_epochs=2)

Distributed Execution

# Enable distributed circuit execution (requires MPI)
model = nq.QNN(layers=2, qubits=4, distributed=True)
model.fit(X, y, epochs=10, shots=1000)

Explainability

# Analyze parameter sensitivity
model = nq.QNN(layers=2, qubits=4)
model.fit(X, y, epochs=10)

# Generate sensitivity analysis plot
result = model.explain(noise_impact=True)
print(result)  # Saves sensitivity.png

API Reference

QNN Class

QNN(
    layers=2,              # Number of quantum layers
    qubits=4,              # Number of qubits
    mitigation='none',     # 'none', 'auto', or 'zne'
    distributed=False,     # Enable distributed execution
    federated=False,       # Enable federated learning
    dp_sigma=0.0          # DP noise standard deviation
)

Methods

• fit(X, y, epochs=10, shots=1000) - Train the model on provided data
• federated_fit(local_X, local_y, rounds=5, local_epochs=2) - Federated training with DP support
• explain(noise_impact=False) - Generate sensitivity analysis visualization

Requirements

• Python >= 3.8
• PennyLane >= 0.30
• Qiskit >= 1.0
• PyTorch >= 2.0
• NumPy, SymPy, Matplotlib
• MPI4Py (optional, for distributed execution)

Architecture

The library implements a variational quantum circuit architecture with:

• Data Encoding: Rotation gates (RY) for classical data embedding
• Variational Layers: Parameterized rotations (RZ) with entangling gates (CNOT)
• Measurement: Pauli-Z expectation values for all qubits

Error Mitigation

Two mitigation strategies are available:

1. Auto: Neural network-based error prediction and correction
2. ZNE: Zero-noise extrapolation using simple scaling factors

Federated Learning

Implements the Federated Averaging (FedAvg) algorithm:

• Local training on each node
• Optional differential privacy noise injection
• Global parameter aggregation via MPI

License

Apache License 2.0

Citation

If you use this library in your research, please cite:

@software{nisqoptml,
  author = {Venkata Vikhyat Choppa},
  title = {NISQOptML: NISQ-Optimized Machine Learning},
  version = {0.1.0},
  year = {2025},
  url = {https://github.com/VikhyatChoppa18/nisqoptml}
}

Author

Venkata Vikhyat Choppa
Email: [e-mail] Github: https://github.com/vikhyatchoppa18

Acknowledgments

Built with PennyLane, Qiskit, and PyTorch.