Integrated Quantum’s VEIL Achieves up to 99.96% Data Compression with Full Accuracy
ARQQ•Integrated Quantum released a white paper demonstrating its VEIL™ technology compresses sensitive inputs by 95% to 99.96% across benchmarks MNIST, CBIS-DDSM and E2006 financial filings while maintaining or exceeding baseline model accuracy. The paper reports VEIL™ outperforms differential privacy in reconstruction and attribute inference attack resilience under testing conditions.
1. White Paper Release
On May 26, 2026, Integrated Quantum Technologies published a white paper titled "Informationally Compressive Anonymization: Non-Degrading Sensitive Input Protection for Privacy-Preserving Supervised Machine Learning," detailing its VEIL™ product aimed at removing sensitive data from AI pipelines without performance loss.
2. Compression and Performance Results
The paper reports VEIL™ achieved between 95% and 99.96% compression across diverse datasets—including MNIST, Fashion-MNIST, Ames Housing, YearPredictionMSD, Home Credit Default Risk, Default of Credit Card Clients, CBIS-DDSM medical imaging, and the E2006 financial filings dataset—while matching or exceeding baseline model accuracy in every test.
3. Comparative Analysis and Limitations
VEIL™ outperformed differential privacy in simulated reconstruction and attribute inference attacks and incurred lower compute costs than homomorphic encryption, though the company cautions that certain adversarial scenarios and system vulnerabilities could still permit limited data leakage and that real-world results may vary.
