Algorithms · Quantum machine learning
Quantum kernel anomaly detection for credit card fraud
Detect anomalous, potentially fraudulent, transactions in a credit-card dataset, framed as an anomaly-detection task and compared against classical kernel-based benchmarks such as one-class support vector machines.
Attested & literatureUpdated
- Record typeLiterature reference
- Reference implementationcredit_card_fraud
- Reported costNot stated by the sources read
- Papers1
Detect anomalous, potentially fraudulent, transactions in a credit-card dataset, framed as an anomaly-detection task and compared against classical kernel-based benchmarks such as one-class support vector machines. Kyriienko and Magnusson develop quantum protocols for anomaly detection and apply them to credit card fraud detection. They first establish classical benchmarks from supervised and unsupervised machine learning methods, choosing average precision as their metric for detecting anomalous data, and focus on kernel-based approaches — basing their unsupervised modelling on one-class support vector machines (OC-SVM) — for ease of direct comparison to the quantum protocols. They then employ quantum kernels of different type for the same anomaly-detection task, and report that quantum fraud detection can challenge equivalent classical protocols as the number of features grows, where the number of features equals the number of qubits used for data embedding. Running simulations with registers up to 20 qubits, they find that quantum kernels with a re-uploading structure give better average precision than the classical benchmarks, with the advantage increasing with system size, reaching a quantum-classical separation of 15% in average precision at 20 qubits. The authors discuss the prospects of fraud detection on near- and mid-term quantum hardware and describe possible future improvements.
TakesNothingWhat joins here
No input port at this edge: the record publishes no gate sequence and no register, so there is nothing here to read one off — and unlike a declared hole, nothing has been recorded about what belongs here.
Nothing in the Atlas meets this end.
ReturnsNothingWhat joins here
No output port at this edge: the record publishes no gate sequence and no register, so there is nothing here to read one off — and unlike a declared hole, nothing has been recorded about what belongs here.
Nothing in the Atlas meets this end.
This record publishes no gate sequence and no register, so there is nothing here to read an interface off. Absent rather than empty. See all 152 →
Kyriienko and Magnusson develop quantum protocols for anomaly detection and apply them to credit card fraud detection. They first establish classical benchmarks from supervised and unsupervised machine learning methods, choosing average precision as their metric for detecting anomalous data, and focus on kernel-based approaches — basing their unsupervised modelling on one-class support vector machines (OC-SVM) — for ease of direct comparison to the quantum protocols. They then employ quantum kernels of different type for the same anomaly-detection task, and report that quantum fraud detection can challenge equivalent classical protocols as the number of features grows, where the number of features equals the number of qubits used for data embedding. Running simulations with registers up to 20 qubits, they find that quantum kernels with a re-uploading structure give better average precision than the classical benchmarks, with the advantage increasing with system size, reaching a quantum-classical separation of 15% in average precision at 20 qubits. The authors discuss the prospects of fraud detection on near- and mid-term quantum hardware and describe possible future improvements. The sources read state no complexity bound for this record (The abstract of arXiv:2208.01203, the only source read for this record, states no computational complexity bound anywhere in it — no query count, no gate count, no circuit depth, and no asymptotic scaling in the number of qubits or features. Its quantitative content is a simulated statistical-performance result, not a complexity bound: "Performing simulations with registers up to 20 qubits, we find that quantum kernels with re-uploading demonstrate better average precision, with the advantage increasing with system size." The same abstract adds, "Specifically, at 20 qubits we reach the quantum-classical separation of average precision being equal to 15%." Average precision is a classification-quality metric, not a running-time or resource bound, so it is not recorded in the complexity field. The abstract's only complexity-flavored remark is qualitative and unquantified: it observes that quantum FD "can challenge equivalent classical protocols at increasing number of features (equal to the number of qubits for data embedding)" without stating a rate, an exponent, or any other scaling expression. The demonstration this record covers states no bound either. The complexity field is therefore left empty on purpose.).
TakesNothingWhat joins here
No input port at this edge: the record publishes no gate sequence and no register, so there is nothing here to read one off — and unlike a declared hole, nothing has been recorded about what belongs here.
Nothing in the Atlas meets this end.
ReturnsNothingWhat joins here
No output port at this edge: the record publishes no gate sequence and no register, so there is nothing here to read one off — and unlike a declared hole, nothing has been recorded about what belongs here.
Nothing in the Atlas meets this end.
This record publishes no gate sequence and no register, so there is nothing here to read an interface off. Absent rather than empty. See all 152 →
- Record type
- Literature reference
- Reference implementation
- credit_card_fraud
- Reported cost
- Not stated by the sources read
- Complexity basis
- The abstract of arXiv:2208.01203, the only source read for this record, states no computational complexity bound anywhere in it — no query count, no gate count, no circuit depth, and no asymptotic scaling in the number of qubits or features. Its quantitative content is a simulated statistical-performance result, not a complexity bound: "Performing simulations with registers up to 20 qubits, we find that quantum kernels with re-uploading demonstrate better average precision, with the advantage increasing with system size." The same abstract adds, "Specifically, at 20 qubits we reach the quantum-classical separation of average precision being equal to 15%." Average precision is a classification-quality metric, not a running-time or resource bound, so it is not recorded in the complexity field. The abstract's only complexity-flavored remark is qualitative and unquantified: it observes that quantum FD "can challenge equivalent classical protocols at increasing number of features (equal to the number of qubits for data embedding)" without stating a rate, an exponent, or any other scaling expression. The demonstration this record covers states no bound either. The complexity field is therefore left empty on purpose.
- Circuit
- Not supplied
This record is not a concrete circuit.
Quantum vs classical
Classical baseline
Compare Quantum machine learning with the strongest classical method for the same instance, input budget, and output metric.
Quantum claim
This reference exposes a quantum circuit pattern; it does not imply an application-level speedup without a matched benchmark.
How to compare
Report input loading, circuit depth, repetitions, classical preprocessing, post-processing, and wall-clock time together.
Nobody has reviewed this record for gaps yet.
ALGORITHM: Quantum kernel anomaly detection for credit card fraud
PROBLEM: Detect anomalous, potentially fraudulent, transactions in a credit-card dataset, framed as an anomaly-detection task and compared against classical kernel-based benchmarks such as one-class support vector machines.
IDEA: Kyriienko and Magnusson develop quantum protocols for anomaly detection and apply them to credit card fraud detection. They first establish classical benchmarks from supervised and unsupervised machine learning methods, choosing average precision as their metric for detecting anomalous data, and focus on kernel-based approaches — basing their unsupervised modelling on one-class support vector machines (OC-SVM) — for ease of direct comparison to the quantum protocols. They then employ quantum kernels of different type for the same anomaly-detection task, and report that quantum fraud detection can challenge equivalent classical protocols as the number of features grows, where the number of features equals the number of qubits used for data embedding. Running simulations with registers up to 20 qubits, they find that quantum kernels with a re-uploading structure give better average precision than the classical benchmarks, with the advantage increasing with system size, reaching a quantum-classical separation of 15% in average precision at 20 qubits. The authors discuss the prospects of fraud detection on near- and mid-term quantum hardware and describe possible future improvements.
REPORTED COST: Not stated by the sources read
BASIS: The abstract of arXiv:2208.01203, the only source read for this record, states no computational complexity bound anywhere in it — no query count, no gate count, no circuit depth, and no asymptotic scaling in the number of qubits or features. Its quantitative content is a simulated statistical-performance result, not a complexity bound: "Performing simulations with registers up to 20 qubits, we find that quantum kernels with re-uploading demonstrate better average precision, with the advantage increasing with system size." The same abstract adds, "Specifically, at 20 qubits we reach the quantum-classical separation of average precision being equal to 15%." Average precision is a classification-quality metric, not a running-time or resource bound, so it is not recorded in the complexity field. The abstract's only complexity-flavored remark is qualitative and unquantified: it observes that quantum FD "can challenge equivalent classical protocols at increasing number of features (equal to the number of qubits for data embedding)" without stating a rate, an exponent, or any other scaling expression. The demonstration this record covers states no bound either. The complexity field is therefore left empty on purpose.
PRIMARY SOURCE: Oleksandr Kyriienko, Einar B. Magnusson (2022), Unsupervised quantum machine learning for fraud detection — https://arxiv.org/abs/2208.01203
This is a literature reference record, not an executable circuit.A reference record, not runnable source. Leona cannot execute it, so it cannot be saved to your Library as a circuit.
Related entries
- Quantum kernel support-vector classifier
Algorithms
- Phase feature-map benchmark · 16 qubits
Basic circuits
- Option pricing by amplitude estimation
Algorithms
References
- Unsupervised quantum machine learning for fraud detection
Oleksandr Kyriienko, Einar B. Magnusson · 2022
Primary source: it describes the classical benchmarks (supervised and unsupervised methods including OC-SVM), the quantum-kernel protocols including re-uploading, and reports the simulated average-precision separation of 15% at 20 qubits. Consult it for the dataset used, the specific quantum kernels compared, the circuit constructions, and any results beyond 20 simulated qubits, none of which the abstract states.