QSBench/QSBench-Thermal-Demo-v1.0.0

🌐 Website | 🤗 Dataset | 🛠️ GitHub | 🚀 Interactive Demo

QSBench Thermal Relaxation Demo v1.0.0

Quantum Machine Learning dataset for realistic noise robustness and sim-to-real research. Includes paired ideal and noisy expectation values under thermal relaxation (T1/T2) noise — the most physically relevant noise model on current quantum hardware.

2048 high-quality synthetic quantum circuits with thermal relaxation noise — demo subset of the QSBench Noise Pack.

Designed for researchers and engineers working on noise-aware quantum ML, decoherence analysis, and error mitigation.

Why this dataset?

Real quantum hardware is dominated by energy relaxation (T1) and dephasing (T2). Thermal relaxation is the standard model used by IBM, Google, IonQ and most superconducting platforms. This dataset allows you to:

• —Study realistic T1/T2 decoherence effects
• —Train models that predict or mitigate relaxation-induced errors
• —Benchmark robustness under physically accurate noise
• —Explore sim-to-real transfer on 8-qubit circuits

Use Cases

• —Noise robustness benchmarking (T1/T2)
• —Error mitigation research (especially relaxation errors)
• —Predicting noisy expectation values
• —Learning error correction models
• —Feature engineering for decoherence-aware quantum ML

Dataset Overview

• —Samples: 2048
• —Qubits: 8
• —Depth: 6
• —Circuit Families: Mixed (HEA, RealAmplitudes, QFT, Efficient SU(2), Random)
• —Entanglement: Full
• —Noise: Thermal Relaxation (T1 = 50 μs, T2 = 30 μs)
• —Observables: Z, X, Y in mixed mode (global + per-qubit)
• —Shots: 1024
• —Splits: Train / Validation / Test — deterministic hash-based

What's Inside Each Sample

Each sample in the Parquet files contains:

• —Raw and transpiled QASM representations
• —Circuit adjacency matrix
• —Gate statistics (CX, H, RX, RY, RZ, etc.)
• —Structural metrics: Gate entropy + Meyer-Wallach entanglement
• —Ideal expectation values
• —Noisy expectation values (after thermal relaxation)
• —Explicit error targets: error_<label> = ideal - noisy
• —Circuit metadata and generation parameters
• —Deterministic split label

Key Learning Signals

For every observable, the dataset provides: ideal_expval_*, noisy_expval_*, error_*, sign_ideal_*, sign_noisy_*. This supports both high-precision regression and classification tasks for realistic noise modeling.

QSBench-Thermal: Realistic Decoherence Modeling

You don't need a PhD in Quantum Physics to use this dataset. This dataset represents the dominant noise channel on today's quantum processors — thermal relaxation. It behaves like gradual "battery drain" of quantum information over time.

The ML Mission: Complex Tabular Regression

Unlike symmetric depolarizing noise, thermal relaxation is biased and time-dependent. Your mission is to build a predictive model that understands how circuit structure (adjacency, gate counts, entanglement) influences T1/T2 decay rates.

Dataset Anatomy (Features & Targets)

Quick Start Idea

Compare feature importances. Does meyer_wallach or two_qubit_gates matter more when predicting thermal relaxation errors?

Load the Dataset

from datasets import load_dataset

# Load the thermal relaxation demo dataset
dataset = load_dataset("QSBench/QSBench-Thermal-Demo-v1.0.0", split="train")

# Inspect the first sample
print(dataset[0])

Repository Structure

The dataset is stored in the main branch and contains only the data files:

QSBench-Thermal-Demo-v1.0.0/
├── README.md
└── data/
    └── shards/
        └── *.parquet
        └── *.csv

All metadata files (meta.json, schema.json, coverage.json, etc.) are in the metadata branch. 👉 browse metadata branch

Related QSBench Datasets

• —Core Pack (2k samples)
• —Depolarizing Pack (5k samples)
• —Amplitude Damping Pack (5k samples)
• —Transpilation Pack (5k samples)
• —Realistic hardware-mimic (2k samples)
• —Readout Error (2k samples)

Part of the QSBench Family

This is a small public demo version. Full-scale datasets (20k–150k+ samples), specialized noisy versions, and custom hardware packs are available on the QSBench website.

Notes

• —Fully synthetic dataset generated with Qiskit Aer

• —No real-world or personal data

• —License: CC BY-NC 4.0 (Personal & Research Use)

Questions or custom requests? Visit QSBench website or open an issue on GitHub.

Email: QSBench@gmail.com

Support QSBench

You can support the project directly on this Giveth page: [https://giveth.io/project/qsbench](https://giveth.io/project/qsbench)

Your donations help us generate larger datasets, cover GPU costs, and continue developing new realistic noise models.

Generated with QSBench Generator v5.1.0