Artificial intelligence is taking a quantum leap
On June 1, IonQ, an industry leader in quantum computing, published the results of an early study of its application of quantum computers to simulate human cognition. The paper describes the world's first publicly available approach: the research team has run a basic human cognition model on quantum hardware, which paves the way for the potential development of improved decision models that mimic the human mindset.
Classical machine learning (ML) is a powerful subset of artificial intelligence. Machine learning has evolved from simple pattern recognition in the 1960s to today's advanced use of massive data sets for training and producing highly accurate predictions.
At the same time, global data usage increased from 1.2 trillion gigabytes to nearly 60 trillion gigabytes from 2010 to 2020. at some point, quantum systems will be able to handle the continued exponential growth of data more easily than classical computers, which may need to struggle to keep up. Theoretically, at some point in the near future, only quantum computers will be able to handle such large scale and complexity.
This observation also applies to the ML field, where at some point the real breakthroughs will come from quantum machine learning (QML) rather than classical methods - it only makes sense.
The potential value of quantum computers capable of simulating human decision-making processes cannot be overstated, as such a future is getting closer to reality," said Peter Chapman, CEO and President of IonQ. This breakthrough offers tremendous potential for fields such as generative artificial intelligence to create complex AI systems that produce highly realistic and creative output. With the unparalleled computational power of quantum computing, this research provides an important foundation for developing complex associative networks that will also drive the pursuit of future innovation."
Indeed, while other quantum computing companies are exploring QML, IonQ's advanced QML research is particularly "fascinating."
IonQ's CEO Peter Chapman had an extensive background in machine learning when he worked with Ray Kurzweil at Kurzweil Technologies: Chapman played a key role in developing a groundbreaking character recognition system that generated text characters from scanned images. Urzweil Technologies eventually used this approach to create a comprehensive digital library for the blind and visually impaired.
And, Chapman is optimistic about the future of QML. He believes that QML will eventually be as important as OpenAI's ChatGPT and the large language models used by other generative AI systems. For this reason, QML has long been included in IonQ's long-term quantum product roadmap.
The next major technical milestone for IonQ is the implementation of 35 AQ.
Today, IonQ is working with many of the leading companies in artificial intelligence and machine learning, which includes Amazon, Dell, Microsoft, and Nvidia. These partnerships combine IonQ's quantum technology expertise with the AI knowledge of its partners.
IonQ's primary focus is not just on the number of quantum bits, but more fully on the quality of quantum bits and how they operate as a system. This quality (also known as the fidelity of the quantum bits) is a key differentiator for effectively accomplishing quantum computing, which IonQ measures with an application-oriented benchmark it calls the Algorithmic Quantum Bit (or #AQ).
To date, IonQ has created three capture-ion quantum computers: IonQ Harmony, IonQ Aria, and its newest model, a software-defined quantum computer called IonQ Forte.
- IonQ Harmony is the first commercial quantum computer launched by IonQ. It can be dynamically reconfigured in software to use up to 11 quantum bits, #AQ is 9. All quantum bits are fully connected, meaning customers can run a two-quantum bit gate between any pair of quantum bits. It is an excellent processor, has an efficient back-end, is suitable for small-scale proof-of-concept work, and is compatible with most quantum SDKs.
- IonQ Aria is the fifth generation of IonQ's quantum machine. It is available on IonQ Quantum Cloud and all public clouds. With a #AQ of 25, a higher quantum bit count and high gate fidelity, IonQ Aria is capable of performing calculations on more complex problems. Its #AQ of 25 also means that there is less noise in the quantum system. With less noise, even the most complex problems require fewer iterations, thus saving valuable time and money.
The IonQ Aria's miniaturized trap and vacuum chamber.
- The IonQ Forte is IonQ's latest quantum computer. It offers enhanced flexibility, accuracy and performance. IonQ Forte is equipped with a highly specialized acousto-optical deflector (AOD) that points the laser beam at individual quantum bits in the ion chain, and, applies logic gates between the quantum bits. The processor, like the IonQ Aria, has a capacity of up to 32 quantum bits, which can be further expanded in software.
There are currently two Aria's online. According to Chapman, a second Aria machine is needed to handle increased customer demand and to increase the company's redundancy, capacity and order processing speed.
In addition, IonQ is working to commercialize IonQ Forte, and IonQ Aria and IonQ Harmony are already available in the cloud through Google, Amazon Braket, Microsoft Azure and IonQ Quantum Cloud. Cloud access for IonQ Forte will be announced at a later date, according to the company.
Forte recently demonstrated a record 29 AQs, which puts it seven months ahead of IonQ's original 2023 AQ target.
Although quantum computing is still being performed by mid-stage prototypes, it has the potential to solve problems far beyond the capabilities of classical supercomputers, perhaps within this decade. Meanwhile, scaled versions of the classical ML model are already being used in hundreds of thousands of applications in virtually every industry as quantum computing prototypes get closer to operational soundness. These applications range from personalized recommendations on shopping sites to critical medical diagnostics (such as analyzing X-rays and MRI scans) to enable more accurate detection of disease than humans.
QML is a still-developing field that uses quantum computers to perform challenging ML tasks, although quantum machines are less practical than classical computers at this point. Combining ML and quantum computing (QC) has produced QML to create a technique that should soon be more powerful than classical machine learning.
According to Chapman, much of today's QML was created by converting classical machine learning algorithms to quantum algorithms. qML is not without its challenges. It has many of the problems associated with current quantum computers, the most prevalent being susceptibility to decoherence caused by environmental noise and prototype hardware limitations.
"Look at the research we've done in the past with Fidelity, GE, Hyundai and a few others," Chapman has said, "All of these projects started with ordinary machine learning algorithms before we converted them to quantum algorithms. "
However, he explained, IonQ's research showed that QML outperformed many of its classical ML counterparts. "Our version of QML beats comparable classical ML versions," he says: "Sometimes the results show that the QML model captures the signal in the data better, or sometimes the number of iterations required to get through the data is significantly less. And sometimes, as our recent research has shown, QML requires about 8,000 times less data than the classical model."
QML uses two principles of quantum mechanics, superposition and entanglement, to develop new machine learning algorithms. Quantum superposition allows quantum bits to be in multiple states at the same time, while quantum entanglement allows many quantum bits to share the same state. This is in contrast to classical physics, where a bit can only be in one state at a time, and connections between bits can only be made by physical means. The associated quantum properties allow developers to create QML algorithms to solve problems that are difficult to solve using classical computers.
It is worth noting that QML is still in the early stages of development; it is not yet powerful enough to solve very large and very complex machine learning problems. Nonetheless, QML still has the potential to revolutionize classical machine learning by training models faster, providing greater accuracy, and opening the door to newer and more powerful algorithms.
Quantum AI is even newer than QML. About a year ago, IonQ began exploring quantum AI. Its first research effort produced a paper on modeling human cognition, published in the peer-reviewed scientific journal Entropy. The paper showed that human decision making can be tested on a quantum computer. Since the 1960s, researchers have found that people do not always follow the rules of classical probability when making decisions. For example, the order in which people are asked questions can affect their answers; quantum probability helps clarify this oddity.
This research paper does not say that the brain operates explicitly using quantum mechanics. Instead, it applies the same mathematical structure to both domains, which adds to the complexity of using quantum computers to model human cognition.
"We're excited about the potential of quantum to add power not only to machine learning, but also to artificial general intelligence or AGI," Chapman said, "AGI is the degree to which artificial intelligence is powerful enough to do any task that a human can do. Some things are almost impossible to model on a classical computer, but possible on a quantum computer. And, I think that AGI will likely be the ultimate place to finish these types of problem sets."
Using classical machine learning algorithms and converting them to quantum machine learning is possible - and IonQ has done this successfully many times:
- January 2023 Quantum Natural Language Processing (QNLP) is a subfield of machine learning focused on developing algorithms that can process and understand natural language (i.e., language used by humans.) IonQ researchers have demonstrated that statistically significant results can be obtained using real data sets, although it is easier than the easier artificial language examples previously used in the development of quantum NLP systems more difficult to predict. The team compared alternative approaches to quantum NLP, in part addressing contemporary issues, including informal language, fluency and authenticity.
- In January 2023, IonQ's research focused on text classification with QNLP. This study showed that amplitude-encoded feature maps combined with quantum support vector machines could use a dataset of 50 actual comments and achieve an average accuracy of 62% to predict sentiment. This is small, but much larger than previously reported results using quantum NLP.
- This joint study by IonQ, Fidelity Center for Applied Technologies (FCAT) and Fidelity Investments in November 2022 focused on generative quantum learning of joint probability distribution functions (GAN, QGAN and QCBM) They all use machine learning to learn data and make predictions. This research shows that the relationship between two or more variables can be represented by the quantum states of multiple particles. This is important because it shows that quantum computers can be used to model and understand complex relationships between variables.
- In November 2021, IonQ and Zapata Computing developed the first practical and experimental hybrid quantum-classical QML algorithm that can generate high-resolution images of handwritten numbers. The results surpass those of comparable classical generative adversarial networks (GANs) trained on the same database.
- In September 2021, researchers from IonQ and FCAT developed a proof-of-concept QML model to analyze the numerical relationships in the daily returns of Apple and Microsoft stocks from 2010 to 2018. The daily return is the stock's price at the close of each day compared to the price at the close of the previous day. The metric measures the daily performance of the stock. The model shows that quantum computers can be used to generate correlations that cannot be effectively reproduced by classical means such as probability distributions.
- In December 2020, in a collaboration between IonQ and QC Ware, classical data was loaded onto quantum states to enable efficient and powerful QML applications. Machine learning achieves the same level of accuracy and runs faster than on a classical computer. The project uses QC Ware's Forge Data Loader technology to transform classical data into quantum states. Quantum algorithms running on IonQ's hardware performed at the same level as classical algorithms, identifying the correct number on average 8 out of 10 times.
This is just a partial snapshot of IonQ's exploration of quantum artificial intelligence. It is worth noting that these QML models often outperform the original ML models. In summary, although quantum machine learning is still an emerging field: it comes from the intersection of quantum information processing, machine learning, and optimization techniques, this field still promises to solve problems faster and more accurately than classical machine learning.
Reference links:
[1] https://www.forbes.com/sites/moorinsights/2023/06/02/a-quantum-leap-in-ai-ionq-aims-to-create-quantum-machine-learning-models- at-the-level-of-general-human-intelligence/?sh=70565b727b88
[2] https://www.mdpi.com/1099-4300/25/4/548
