Quantum Computing News

IonQ Reports Quantum AI Developments, Including Improved LLMs and Synthetic Data Creation

Hybrid Quantum

The application of quantum computing to artificial intelligence (AI) and machine learning has advanced significantly, according to IonQ, a leading company in commercial quantum computing and networking. The business described two novel hybrid quantum-classical techniques for producing synthetic materials data and improving large language models (LLMs).

These advancements are seen as proof of viable, near-term commercial quantum applications in artificial intelligence, which are especially helpful for challenging jobs and environments with limited data.

Enhancing Large Language Models via quantum fine-tuning is one area of progress. In a recent study, IonQ presented a hybrid quantum-classical architecture intended to enhance LLM fine-tuning. This allows the model to be repurposed for certain tasks, such comprehending phrase sentiment, by adding a parameterized quantum circuit as a new layer and adding a short batch of training data to a pre-trained LLM.

Using a same number of parameters, this hybrid quantum approach demonstrated a significant gain in accuracy compared to classical-only methods, according to IonQ. Researchers found that as the number of qubits increased, so did the categorisation accuracy. Additionally, as the issue size increases above 46 qubits, they predict significant energy savings for the inference phase when employing the hybrid quantum method in comparison to classical models.

This study raises the possibility of using quantum-enhanced fine-tuning to a variety of AI models, such as those for image processing, natural language processing, and the prediction of scientific properties. “This research demonstrates how quantum computing can be strategically integrated into classical AI workflows, leveraging increased expressivity to enhance LLMs, especially in scenarios with limited data,” said Masako Yamada, Director of Applications Development at IonQ.

Pioneering quantum generative modelling for enhanced material characteristics is the second significant advancement. IonQ and a major automaker worked together to apply quantum-enhanced generative adversarial networks (GANs) to materials science in a different research publication. These quantum-enhanced GANs were trained by researchers to produce artificial images of steel microstructures. This methodology is useful for enhancing traditional imaging techniques where sparse data can make classical model training challenging.

In up to 70% of cases, synthetic microstructure images created using IonQ’s hybrid QGAN method outperformed images created using baseline traditional methods in terms of quality score. Developing industrial AI models to optimise manufacturing processes and material qualities requires the capacity to supplement image data, particularly when using proprietary, data-poor, or unbalanced datasets.

“This work shows how combining IonQ’s quantum computers with classical machine learning can yield impressive results for materials science and manufacturing, producing higher quality images with less data thantraditional approaches,” says IonQ SVP of Product Ariel Braunstein.
These research milestones, according to IonQ, make use of their Forte Enterprise-class quantum computers and come after other recent announcements, including a memorandum of understanding with AIST’s G-QuAT to advance hybrid quantum computing and AI and a quantum simulation tool created in collaboration with Ansys.

ArXiv has the whole technical papers that describe these discoveries. “Quantum Large Language Model Fine-Tuning” is the title of one study.

These developments put IonQ at the forefront of investigating how quantum computing might soon provide noticeable advantages to important AI and machine learning applications.

In summary

The latest demonstrations from IonQ mark a major advancement in proving the usefulness of quantum computing in artificial intelligence. The effective application of quantum methods to generative modelling and LLM fine-tuning demonstrates how quantum computers may be able to overcome the drawbacks of traditional AI, especially in settings with little data. IonQ’s dedication to creating commercially viable quantum applications in AI is further demonstrated by these developments, which are backed by noticeable gains in accuracy and image quality.

News source via source