Researchers Find Resilient Feature Map Maintains Accuracy at 10% Noise

Muhammad Ahsan Shakeel from Management Sciences and colleagues have shown that an amplitude-inspired quantum feature map maintains one hundred per cent test accuracy up to an error probability of 0.10 across depolarizing, bit-flip and phase-flip noise channels. Other tested feature maps experienced performance drops between sixty-five and ninety percent under similar conditions. Maintaining high classification accuracy in Quantum Support Vector Machines was previously challenging on noisy intermediate-scale quantum hardware due to gate errors distorting kernel matrices.

The team identified a key method for preparing quantum computers for machine learning tasks; this new approach to encoding information preserves perfect accuracy even with relatively high levels of error. Specifically, the amplitude-inspired data encoding performed flawlessly, up to ten times better than other tested methods when subjected to simulated hardware errors. These findings offer practical guidance for building functional systems using existing technology by establishing clear standards for selecting appropriate encoding techniques. A quantum feature map acts as a translator, converting classical data into a format understandable by a quantum computer, much like transforming into Morse code. Their work reveals that the amplitude-inspired method maintained one hundred per cent test accuracy up to an error probability of 0.10 across various noise types, while others experienced sharply reduced performance. This resilience offers valuable guidance for building practical machine learning systems using current technology. Amplitude encoding yields resilient performance in noisy quantum support vector machines Ten-fold improvements in noise tolerance emerged; an amplitude-inspired feature map maintained one hundred per cent test accuracy up to an error probability of 0·10, a threshold previously unattainable with other tested architectures which fell between sixty-five and ninety percent.

The team at Lahore 54972 and Management Sciences demonstrated this strong durability across depolarizing, bit-flip and phase-flip noise channels using Quantum Support Vector Machines. This breakthrough establishes data-driven criteria for selecting appropriate quantum feature maps on near-term noisy intermediate-scale quantum hardware, offering a pathway towards more robust machine learning applications. The Z feature map exhibited complete immunity to phase-flip errors due to its fundamental commutation properties; this characteristic was not shared by entangled circuit designs which showed train-test generalisation gaps reaching seventeen point five per cent as noise increased. Zero generalisation gap emerged throughout all tests with amplitude variants, indicating consistent performance on both training and unseen data.

The team at Lahore 54972 and Management Sciences limited experiments to relatively small datasets and qubit numbers raising questions about scalability for larger systems. Simulated noise revealed that the amplitude-inspired encoding method excelled but translating these results directly onto existing hardware presents challenges because real devices exhibit unique imperfections beyond simple depolarisation or bit flips. Despite limitations with larger systems and subtle nuances in real device behaviour, this work offers valuable guidance for those building applications on near-term quantum computers; it establishes a clear hierarchy of encoding methods based on noise durability. Sustaining perfect classification accuracy up to a ten per cent error rate, an amplitude-inspired approach sharply improved performance over other tested methods which experienced substantial drops under similar conditions; resilience demonstrated across multiple noise types including depolarizing, bit-flip and phase-flip errors simulating realistic disturbances found on early quantum computers. The research showed that using an amplitude-inspired feature map maintained one hundred percent test accuracy in Quantum Support Vector Machines even with error probabilities as high as ten percent across three different noise channels. This matters because gate-level noise is known to reduce the effectiveness of these machines on current quantum hardware.

The team at Lahore 54972 and Management Sciences identified this encoding method’s durability, offering data-driven criteria for selecting appropriate feature maps; entangled circuit variants produced generalisation gaps up to seventeen point five per cent under similar noisy conditions. 👉 More information🗞 Noise Resilience of Quantum Support Vector Machine with Selected Feature Maps✍️ Muhammad Ahsan Shakeel, Saad Muzammil, Danyal Tayyub and Muhammad Faryad🧠 ArXiv: https://arxiv.org/abs/2608.17495 More like thisQuantum PhysicsWiring density limits qubit control, Bluefors research showsQuantum PhysicsTwo-dimensional quantum models simulated without full wave functionQuantum Research NewsBonn Researchers Win EU Grants to Study Brains and Quantum SystemsPhysicsITMO’s Faculty of Physics finds electrons ‘twist’ to emit light without magnetsStay currentSee today’s quantum computing news on Quantum Zeitgeist for the latest breakthroughs in qubits, hardware, algorithms, and industry deals. Tags: