{"id":51789958,"url":"https://github.com/jeragilo/noise-robust-hqnn-framework","last_synced_at":"2026-07-20T21:32:11.071Z","repository":{"id":326787489,"uuid":"1106362380","full_name":"jeragilo/noise-robust-hqnn-framework","owner":"jeragilo","description":"Master’s thesis research framework for noise-robust hybrid quantum–classical neural networks (HQNNs), evaluating reliability, architecture design, and deployment strategies on NISQ hardware.","archived":false,"fork":false,"pushed_at":"2026-05-07T18:53:16.000Z","size":775,"stargazers_count":0,"open_issues_count":0,"forks_count":0,"subscribers_count":0,"default_branch":"main","last_synced_at":"2026-05-07T20:36:55.257Z","etag":null,"topics":["cirq","hqnn","hybrid-ai","hybrid-models","hybrid-quantum-classical-algorithms","masters-thesis","nisq","noise-robust-learning","pennylane","qiskit","quantum-computing","quantum-machine-learning","quantum-neural-networks","research","research-framework","variational-algorithms"],"latest_commit_sha":null,"homepage":"","language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":null,"status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/jeragilo.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":null,"code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":null,"dco":null,"cla":null}},"created_at":"2025-11-29T05:10:35.000Z","updated_at":"2026-05-07T18:53:20.000Z","dependencies_parsed_at":null,"dependency_job_id":null,"html_url":"https://github.com/jeragilo/noise-robust-hqnn-framework","commit_stats":null,"previous_names":["jeragilo/quantum-hqnn-thesis","jeragilo/noise-robust-hqnn-framework"],"tags_count":1,"template":false,"template_full_name":null,"purl":"pkg:github/jeragilo/noise-robust-hqnn-framework","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/jeragilo%2Fnoise-robust-hqnn-framework","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/jeragilo%2Fnoise-robust-hqnn-framework/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/jeragilo%2Fnoise-robust-hqnn-framework/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/jeragilo%2Fnoise-robust-hqnn-framework/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/jeragilo","download_url":"https://codeload.github.com/jeragilo/noise-robust-hqnn-framework/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/jeragilo%2Fnoise-robust-hqnn-framework/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":35700994,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-07-20T02:08:10.276Z","status":"ssl_error","status_checked_at":"2026-07-20T02:08:09.736Z","response_time":111,"last_error":"SSL_read: unexpected eof while reading","robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":false,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["cirq","hqnn","hybrid-ai","hybrid-models","hybrid-quantum-classical-algorithms","masters-thesis","nisq","noise-robust-learning","pennylane","qiskit","quantum-computing","quantum-machine-learning","quantum-neural-networks","research","research-framework","variational-algorithms"],"created_at":"2026-07-20T21:32:10.425Z","updated_at":"2026-07-20T21:32:11.065Z","avatar_url":"https://github.com/jeragilo.png","language":"Python","funding_links":[],"categories":[],"sub_categories":[],"readme":"# Noise-Robust Hybrid Quantum Neural Networks Framework\n\nThis repository contains the experimental codebase for my Master’s thesis:\n\n**Noise-Robust Hybrid Quantum Neural Networks: A Framework for Scalable Quantum AI in the NISQ Era**\n\nThe project investigates the reliability, robustness, and practical limitations of hybrid quantum–classical neural networks under simulated NISQ-era noise.\n\nThis repository is structured as a reusable benchmarking and optimization framework, not only as a collection of independent demos.\n\n---\n\n## Flagship Result: Stability-Regularized Multi-Observable HQNN\n\nThe strongest configuration combines:\n\n- Stability-regularized noise-aware training\n- Multi-observable quantum feature extraction\n- Linear HQNN architecture\n- Random Forest learned classical readout\n\nResult under depolarizing noise:\n\n| Metric | Value |\n|---|---:|\n| Clean accuracy | 0.9600 |\n| Noisy accuracy | 0.9600 |\n| Accuracy drop | 0.0000 |\n| Robustness score | 1.0000 |\n| Gain over random parity baseline | +0.5067 |\n\nThis supports the main thesis claim that HQNN robustness improves when noise awareness is embedded into the optimization process and richer quantum measurement information is preserved through learned hybrid readouts.\n### Multi-Channel Noise Validation\n\nThe flagship HQNN configuration was also validated across four NISQ-relevant noise channels at eval_noise = 0.05.\n\n| Noise Channel | RF Clean Accuracy | RF Noisy Accuracy | Accuracy Drop | Robustness Score |\n|---|---:|---:|---:|---:|\n| Depolarizing | 0.9600 | 0.9467 | 0.0133 | 0.9861 |\n| Bit flip | 0.9467 | 0.9467 | 0.0000 | 1.0000 |\n| Phase flip | 0.9533 | 0.9533 | 0.0000 | 1.0000 |\n| Amplitude damping | 0.9600 | 0.9267 | 0.0333 | 0.9653 |\n\nThis strengthens the thesis claim by showing that the flagship HQNN method generalizes beyond a single depolarizing-noise condition.\n---\n\n## Main Thesis Claim\n\nNaive HQNN models with fixed parity readout can perform poorly because they compress quantum measurement information into a single decision signal. This framework shows that HQNN performance can be substantially improved by combining:\n\n1. Noise-aware training objectives  \n2. Stability regularization  \n3. Multi-observable quantum feature extraction  \n4. Architecture search  \n5. Learned classical readouts  \n6. Repeated-trial and statistical validation  \n\nThe framework moves the thesis from simple HQNN benchmarking toward a reusable method for discovering noise-robust hybrid quantum-classical learning configurations.\n\n---\n\n## Key Algorithmic Contributions\n\n### 1. Learned Classical Readout HQNN\n\nInstead of using a fixed parity threshold, the circuit measurement distribution is used as a quantum feature representation for a learned classical readout.\n\nBest observed learned-readout result:\n\n| Method | Clean Accuracy | Noisy Accuracy |\n|---|---:|---:|\n| Fixed parity readout | 0.3444 | 0.3667 |\n| Learned logistic readout | 0.8333 | 0.8222 |\n\n---\n\n### 2. Multi-Observable HQNN Readout\n\nThe framework extracts richer quantum-derived features:\n\n- Full bitstring probabilities\n- Single-qubit Z expectations\n- Pairwise ZZ correlations\n- Global parity expectation\n- Probability-distribution statistics\n\nThis produced a 31-dimensional quantum-derived feature vector.\n\nBest observed multi-observable result:\n\n| Method | Clean Accuracy | Noisy Accuracy |\n|---|---:|---:|\n| Multi-observable Logistic Regression | 0.8000 | 0.8200 |\n| Multi-observable Random Forest | 0.8733 | 0.8533 |\n\n---\n\n### 3. Architecture Search for HQNN Robustness\n\nThe framework compares different entanglement structures:\n\n- No entanglement\n- Linear entanglement\n- Ring entanglement\n- Full entanglement\n\nBest architecture-search result:\n\n| Architecture | Readout | Clean Accuracy | Noisy Accuracy |\n|---|---|---:|---:|\n| Linear | Random Forest | 0.8933 | 0.8867 |\n\n---\n\n### 4. Best-Architecture Noise Sweep\n\nThe optimized linear HQNN configuration was tested across increasing depolarizing noise levels.\n\n| Noise Level | Accuracy |\n|---:|---:|\n| 0.00 | 0.8467 |\n| 0.01 | 0.8467 |\n| 0.03 | 0.8467 |\n| 0.05 | 0.8467 |\n| 0.07 | 0.8200 |\n| 0.10 | 0.8133 |\n\nThis shows that the optimized configuration remains stable across moderate noise levels.\n\n---\n\n### 5. Dual-Loss and Stability-Regularized HQNN Training\n\nThe framework implements noise-aware objective functions that combine clean behavior, noisy behavior, and stability under perturbation.\n\nTraining objectives include:\n\n- Standard clean-loss training\n- Noise-aware training\n- Dual-loss training\n- Stability-regularized training\n\nBest observed result:\n\n| Training Mode | Readout | Clean Accuracy | Noisy Accuracy | Accuracy Drop |\n|---|---|---:|---:|---:|\n| Stability-regularized | Multi-observable Random Forest | 0.9600 | 0.9600 | 0.0000 |\n\n---\n\n## Framework Overview\n\nThe framework uses:\n\n- Qiskit\n- Qiskit Aer\n- Qiskit Machine Learning\n- Cirq\n- PennyLane\n- scikit-learn\n- NumPy\n- Matplotlib\n\nThe contribution of this project is the reusable evaluation, optimization, reporting, and robustness-analysis layer built around those libraries.\n\n---\n\n## Framework Features\n\n### Standardized Dataset Handling\n\n- Synthetic classification data\n- Iris\n- Wisconsin Diagnostic Breast Cancer dataset\n- Quantum-compatible low-dimensional preprocessing\n\n### Noise-Analysis Toolbox\n\nSupported noise models:\n\n- Depolarizing noise\n- Bit-flip noise\n- Phase-flip noise\n- Amplitude damping\n\n### Robustness Metrics\n\nThe framework includes reusable metrics:\n\n- `accuracy_drop`\n- `robustness_score`\n- `degradation_slope`\n- `training_instability`\n- `cross_framework_deviation`\n\n### Benchmark Pipelines\n\nThe framework includes pipelines for:\n\n- Hybrid vs classical comparison\n- Noise robustness\n- Cross-framework validation\n- Learned-readout HQNN evaluation\n- Multi-observable HQNN evaluation\n- Architecture search\n- Best-architecture noise sweep\n- Repeated-trial validation\n- Statistical validation\n- Dual-loss noise-aware training\n- Stability-regularized multi-observable HQNN\n\n### Standardized Outputs\n\nThe framework generates:\n\n- JSON summaries\n- CSV summaries\n- Accuracy plots\n- Noise curves\n- Heatmaps\n- Statistical validation reports\n\n---\n\n## Repository Structure\n\n```text\nframework/\n  datasets.py\n  noise_channels.py\n  robustness_metrics.py\n  reporting.py\n  benchmark_runner.py\n\npipelines/\n  main_hybrid_vs_classical.py\n  main_noise_robustness.py\n  main_cross_framework_validation.py\n  main_full_benchmark_summary.py\n  main_framework_capabilities_report.py\n  main_training_mode_comparison.py\n  main_learned_readout_hqnn.py\n  main_multi_observable_hqnn.py\n  main_architecture_search_hqnn.py\n  main_best_architecture_noise_sweep.py\n  main_best_architecture_repeated_trials.py\n  main_statistical_validation.py\n  main_dual_loss_noise_aware_hqnn.py\n  main_dual_loss_multi_observable_hqnn.py\n\ndemos/\n  core/\n  industry/\n\nresults/\n  framework/\n\nenv/\n  requirements.txt\n\nrun_framework.py\nRunning the Framework\n\nRun the full framework:\n\npython run_framework.py\n\nRun individual framework pipelines:\n\nPYTHONPATH=. python pipelines/main_hybrid_vs_classical.py\nPYTHONPATH=. python pipelines/main_noise_robustness.py\nPYTHONPATH=. python pipelines/main_cross_framework_validation.py\nPYTHONPATH=. python pipelines/main_full_benchmark_summary.py\nPYTHONPATH=. python pipelines/main_framework_capabilities_report.py\n\nRun advanced HQNN optimization pipelines:\n\nPYTHONPATH=. python pipelines/main_learned_readout_hqnn.py\nPYTHONPATH=. python pipelines/main_multi_observable_hqnn.py\nPYTHONPATH=. python pipelines/main_architecture_search_hqnn.py\nPYTHONPATH=. python pipelines/main_best_architecture_noise_sweep.py\nPYTHONPATH=. python pipelines/main_best_architecture_repeated_trials.py\nPYTHONPATH=. python pipelines/main_statistical_validation.py\nPYTHONPATH=. python pipelines/main_dual_loss_noise_aware_hqnn.py\nPYTHONPATH=. python pipelines/main_dual_loss_multi_observable_hqnn.py\nDemonstration Ecosystem\n\nThe project includes a 13-demo experimental ecosystem across Qiskit, Cirq, and PennyLane.\n\nCore Demos\nHQNN Toy Classifier\nVQE Energy Minimization\nQAOA MaxCut\nQSVM Anomaly Detection\nNoise-Robust HQNN\nCross-Framework Noise Benchmark\nCross-Platform Parity Consistency\nHQNN Training Loop with SPSA\nIndustry-Inspired Demos\nMedical Risk Classification\nEnergy Grid Optimization\nCybersecurity Anomaly Detection\nHQNN Explainability\nCross-Noise Robustness Heatmap\n\nRun an individual demo from the repository root:\n\nPYTHONPATH=. python demos/core/demo05_hqnn_noise_robust_qiskit.py\n\nCybersecurity demo:\n\nPYTHONPATH=. python demos/industry/demo11_cyber_anomaly_qiskit.py\nEnvironment Setup\nconda create -n hqnn python=3.11 -y\nconda activate hqnn\npip install -r env/requirements.txt\nThesis-Relevant Interpretation\n\nThe results suggest that HQNN performance is not determined only by the quantum circuit itself. It depends strongly on the full hybrid pipeline:\n\nhow noise is inserted into training,\nhow measurement information is extracted,\nhow the classical readout interprets quantum features,\nhow entanglement architecture is selected,\nand how robustness is validated statistically.\n\nThe strongest result shows that a stability-regularized, multi-observable, learned-readout HQNN can preserve performance under simulated depolarizing noise, achieving 0.9600 clean accuracy and 0.9600 noisy accuracy in the reported experiment.\n\nStatus\n\nThe repository currently includes:\n\nFramework layer\nPipeline layer\nCore demos\nIndustry demos\nBenchmark outputs\nAdvanced HQNN optimization pipelines\nStatistical validation outputs\nThesis-ready figures and JSON summaries\nContact\n\nGitHub: https://github.com/jeragilo/\n\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fjeragilo%2Fnoise-robust-hqnn-framework","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fjeragilo%2Fnoise-robust-hqnn-framework","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fjeragilo%2Fnoise-robust-hqnn-framework/lists"}