Zapata Generates High-resolution Digits Using IonQ’s Quantum Computer

Zapata Generates High-resolution Digits Using IonQ’s Quantum Computer


Zapata Computing has released high-resolution images generated by a hybrid quantum-classical algorithm, achieving outstanding results on current noisy intermediate-scale quantum (NISQ) hardware: specifically, IonQ’s ion trap quantum device. The generation of images is an area of machine learning that has already made incredible strides with classical algorithms and is expected to grow further with quantum resources.
In a paper released today, Zapata scientists detail how this algorithm improves over previous methods for learning from the MNIST dataset. We use this method to generate, for the first time with a small gate-based quantum computer, high-quality handwritten digits.  


The core of this work is replacing one part of the widely-used Generative Adversarial Network (GAN) model. The hybrid quantum-classical framework proposed allows the algorithm to maximize the quantum processor’s impact, using quantum effects only available on quantum hardware, while still utilizing the major advances from state-of-the-art machine learning techniques.  More details on the exact processes used are provided in the paper. 

“Generating data in the form of images or video is one of the most challenging tasks in machine learning. We are excited to continue pushing the boundaries of what’s possible with quantum machine learning and with quantum generative models, leveraging quantum technology where we believe it will have the most impact in the field of AI,” says Dr. Alejandro Perdomo-Ortiz, Associate Director of Quantum AI and the technical lead of the published work. 

“This work is an exciting practical application highlighting the potential of quantum generative algorithms, and it is great to see it realized with actual hardware. We look forward to seeing larger and more complex datasets on our new 32 qubit device that we are rolling out,” said Jungsang Kim, Co-founder and Chief Strategy Officer of IonQ.  

The error rates on IonQ’s machine were extremely low, allowing for the excellent results in the paper. IonQ’s device also has very high throughput which makes it effective for practical implementation of real-life applications, such as the algorithm described in this paper. 

This work and these results will help drive the field of quantum machine learning forward and give researchers a baseline to work off of and compare to. Alán Aspuru-Guzik, Chief Science Officer and co-founder of Zapata, said about the results in this paper, “This is, to date, the best quantum rendition of the canonical machine learning problem of the MNIST digits. It sets a benchmark for future machine learning researchers to improve upon and sets the path to further growth in this frontier area of quantum computing applications” 

As the field progresses, both on algorithms and hardware development, we at Zapata are excited to continue pushing the boundaries in quantum machine learning, accelerating the quantum revolution.


The research will be presented at Q2B on December 8  →