A Quantum Leap In AI: IonQ Aims To Create Quantum Machine Learning Models At The Level Of General Human Intelligence

May 17, 2023

IonQ Aria's miniaturized "deck of cards" trap and vacuum chamber package.

Classical machine learning (ML) is a powerful subset of artificial intelligence. Machine learning has advanced from simple pattern recognition in the 1960s to today's advanced use of massive datasets for training and the generation of highly accurate predictions.

Meanwhile, between 2010 and 2020, global data usage increased from 1.2 trillion gigabytes to almost 60 trillion gigabytes. At some point, quantum systems will more easily handle the ongoing exponential growth in data compared to classical computers, which may struggle to keep up. Theoretically, at some point in the not-too-distant future, only quantum computers can handle such massive scale and complexity. Applying this same insight to the realm of ML, it only makes sense that at some point, the real breakthroughs will be coming from quantum machine learning (QML) rather than classical approaches.

IonQ

IonQ roadmap for applications and Algorithmic Qubits (AQ)

Although other quantum computing companies are exploring QML, there are several reasons I have focused on advanced QML research being done at IonQ ($IONQ).

One, IonQ's CEO, Peter Chapman, has a rich background in machine learning when he worked with Ray Kurzweil at Kurzweil Technologies. Chapman played a crucial role in developing a pioneering character recognition system that generated text characters from scanned images.urzweil Technologies eventually used that approach to build a comprehensive digital library for the blind and visually impaired.

Two, Chapman is optimistic about the future of QML. He believes that QML will eventually be as significant as the large language models used by OpenAI's ChatGPT and other generative AI systems. For that reason,QML is built into IonQ's long-term quantum product roadmap.

And three, IonQ collaborates with leading companies in the field of AI and machine learning, such as Amazon, Dell, Microsoft, and NVIDIA. These partnerships combine IonQ's expertise in quantum technology with the partner's AI knowledge of their partners.

IonQ hardware and #AQ

IonQ's primary focus is not just on qubit quantity but more comprehensively on the quality of the qubits and how they operate as a system. This quality—also called qubit fidelity—is a critical differentiator for efficiently completing quantum computations, one that IonQ measures with an application-oriented benchmark that it calls algorithmic qubits or #AQ.

#AQ is based on work pioneered by the Quantum Economic Development Consortium, an independent industry group that evaluates quantum computer utility in real-world settings. Here is how #AQ is computed.

IonQ quantum processors

IonQ has created three trapped-ion quantum computers: IonQ Harmony, IonQ Aria and its latest model, a software-defined quantum computer called IonQ Forte.

There are two Arias online. According to Chapman, the second Aria machine was needed to handle increased customer demand and to improve the company's redundancy, capacity and order processing speed.

Additionally, IonQ is working hard to make the IonQ Forte commercially available

IonQ Aria and IonQ Harmony are cloud accessible via Google, Amazon Braket, Microsoft Azure and IonQ Quantum Cloud. According to the company, cloud access for IonQ Forte will be announced later. Let's take a deeper look at the different quantum computers that IonQ has built:

Forte recently demonstrated a record 29 AQ, which puts it seven months ahead of IonQ's original AQ goal for 2023.

Note: IonQ's next major technical milestone is achieving 35 AQ. At the 35 AQ level, using classical hardware to simulate quantum algorithms can become very challenging and costly. At that point, IonQ believes it will be easier and less expensive for some customers to run models on actual quantum machines rather than attempting to simulate them classically.

ML + QC = QML

Even though quantum computing is still being carried out by mid-stage prototypes, it has the potential, perhaps within this decade, to solve problems far beyond the capability of classical supercomputers. Meanwhile, as quantum computing prototypes are getting closer to becoming operationally sound, scaled versions of classical ML models are already being used in hundreds of thousands of applications across almost every industry. These range from personalized recommendations on shopping sites to critical healthcare diagnostics, such as analyzing X-rays and MRI scans to detect diseases more accurately than humans can.

QML is a still-developing field that uses quantum computers for challenging ML tasks, even though at this point quantum machines are less practical than classical computers. Combining ML and quantum computing (QC) to produce QML creates a technology that should soon be even more powerful than classical machine learning.

According to Peter Chapman, much of today's QML is created by converting classical machine learning algorithms into quantum algorithms. QML is not without challenges. It has many of the same problems as those associated with current quantum computers, the most prevalent being susceptibility to errors caused by environmental noise and decoherence due to prototype hardware limitations.

"Look at the past research we've done with Fidelity, GE, Hyundai and a few others," Chapman said. "All those projects started with regular machine learning algorithms before we converted them to quantum algorithms."

He explained, however, that IonQ's research has shown QML performance to be superior to many of its classical ML counterparts. "Our QML versions beat comparable classical ML versions," he said. "Sometimes the results show that the QML model did a better job capturing the signal in the data, or sometimes the number of iterations needed to go through the data was substantially less. And sometimes, as our most recent research indicates, the data needed for QML was about 8,000 times less than a classical model needs."

Why QML performs better than classical ML

QML uses superposition and entanglement, two principles of quantum mechanics, to develop new machine learning algorithms. Quantum superposition allows qubits to be in multiple states simultaneously, whereas quantum entanglement allows many qubits to share the same state. This is in contrast to classical physics, where a bit can be in only one state at a time, and where connectivity between bits is only possible by physical means. The relevant quantum properties allow developers to create QML algorithms to solve problems that are intractable using classical computers.

It is important to note that QML is still in its early stages of development. It is not yet powerful enough to solve very large and very complex machine learning problems. Still, QML has the potential to revolutionize classical machine learning by training models faster, providing greater accuracy and opening the door for newer and even more powerful algorithms.

Quantum artificial intelligence

Quantum AI is even newer than QML. About a year ago, IonQ started exploring quantum AI. Its first research effort produced a paper on modeling human cognition that was published in the peer-reviewed scientific journal Entropy. The paper shows that human decision making can be tested on quantum computers. Since the 1960s, researchers have found that people don't always follow the rules of classical probability when making decisions. For instance, the sequence in which people are asked questions can influence their answers. Quantum probability helps clarify that oddity.

The research paper doesn't say that the brain explicitly operates on using quantum mechanics. Instead, it applies the same mathematical structures to both fields, which adds to the intrigue of using quantum computers to simulate human cognition.

"We are excited by the potential for quantum to not only add power to machine learning but to artificial general intelligence or AGI as well," Chapman said. "AGI is the point at which AI is strong enough to accomplish any task that a human can. Some things are almost impossible to model on a classical computer but are possible on a quantum computer. And, I think that AGI will likely be where these kinds of problem sets will be done."

Wrapping up

Quantum Machine Learning is still an emerging field. It is the intersection where techniques from quantum information processing, machine learning and optimization come together to solve problems faster and more accurately than classical machine learning.

It is possible to use classical machine learning algorithms and convert them to quantum machine learning. IonQ has done this successfully several times. These QML models often outperform the original ML models.

QML offers several advantages over traditional machine learning thanks to quantum mechanics in the form of superposition and entanglement. QML can complement the growing trend of using ML models for many classification tasks, from image recognition to NLP.

Analyst's notes:

Here are a few IonQ QML-related research papers I found interesting:

January 2023 — Quantum natural language processing (QNLP) is a subfield of machine learning that focuses on developing algorithms that can process and understand natural language (i.e., the languages spoken by humans). IonQ researchers demonstrated that statistically meaningful results can be obtained using real datasets, even though it is much more difficult to predict than with easier artificial language examples used previously in developing quantum NLP systems. Other approaches to quantum NLP are compared, partly with respect to contemporary issues including informal language, fluency and truthfulness.

January 2023 — Research by IonQ focused on text classification with QNLP. This research demonstrated that an amplitude encoded feature map combined with a quantum support vector machine can achieve 62% average accuracy predicting sentiment using a dataset of 50 actual movie reviews. This is small, but considerably larger than previously-reported results using quantum NLP.

November 2022 — This joint research by IonQ, the Fidelity Center for Applied Technology (FCAT) and Fidelity Investments focuses on generative quantum learning of joint probability distribution functions—GANs, QGANs and QCBMs—all of which use machine learning to learn from data and make predictions. The research demonstrates that a relationship between two or more variables can be represented by a quantum state of multiple particles. This is important because it shows that quantum computers can be used to model and understand complex relationships between variables.

November 2021 — IonQ and Zapata Computing developed the first practical and experimental implementation of a hybrid quantum-classical QML algorithm that can generate high-resolution images of handwritten digits. The results outperformed comparable classical generative adversarial networks (GANs) trained on the same database. GAN is a machine learning model with two neural networks that compete against each other to produce the most accurate prediction.

September 2021 — Researchers from IonQ and FCAT developed a proof-of-concept QML model to analyze numerical relationships in the daily returns of Apple and Microsoft stock from 2010 to 2018. Daily returns are the price of a stock at the daily closure compared to its price at the previous day's closure. The metric measures daily stock performance. The model demonstrated that quantum computers can be used to generate correlations which cannot be efficiently reproduced by classical means, such as probability distribution.

December 2020 — In a partnership between IonQ and QC Ware, classical data was loaded onto quantum states to allow efficient and robust QML applications. Machine learning achieved the same level of accuracy and ran faster than on classical computers. The project used QC Ware's Forge Data Loader technology to transform classical data onto quantum states. The quantum algorithm, running on IonQ's hardware, performed at the same level as the classical algorithm, identifying the correct digits eight out of 10 times on average.

Paul Smith-Goodson is the Vice President and Principal Analyst covering AI and quantum for Moor Insights & Strategy. He is currently working on several personal research projects, one of which is a unique method of using machine learning and ionospheric data collected from a national network of HF transceivers for highly accurate prediction of real-time and future global propagation of HF radio signals.

For current information on these subjects, you can follow him on Twitter.

Moor Insights & Strategy provides or has provided paid services to technology companies like all research and tech industry analyst firms. These services include research, analysis, advising, consulting, benchmarking, acquisition matchmaking, and video and speaking sponsorships. The company has had or currently has paid business relationships with 8×8, Accenture, A10 Networks, Advanced Micro Devices, Amazon, Amazon Web Services, Ambient Scientific, Ampere Computing, Anuta Networks, Applied Brain Research, Applied Micro, Apstra, Arm, Aruba Networks (now HPE), Atom Computing, AT&T, Aura, Automation Anywhere, AWS, A-10 Strategies, Bitfusion, Blaize, Box, Broadcom, C3.AI, Calix, Cadence Systems, Campfire, Cisco Systems, Clear Software, Cloudera, Clumio, Cohesity, Cognitive Systems, CompuCom, Cradlepoint, CyberArk, Dell, Dell EMC, Dell Technologies, Diablo Technologies, Dialogue Group, Digital Optics, Dreamium Labs, D-Wave, Echelon, Ericsson, Extreme Networks, Five9, Flex, Foundries.io, Foxconn, Frame (now VMware), Fujitsu, Gen Z Consortium, Glue Networks, GlobalFoundries, Revolve (now Google), Google Cloud, Graphcore, Groq, Hiregenics, Hotwire Global, HP Inc., Hewlett Packard Enterprise, Honeywell, Huawei Technologies, HYCU, IBM, Infinidat, Infoblox, Infosys, Inseego, IonQ, IonVR, Inseego, Infosys, Infiot, Intel, Interdigital, Jabil Circuit, Juniper Networks, Keysight, Konica Minolta, Lattice Semiconductor, Lenovo, Linux Foundation, Lightbits Labs, LogicMonitor, LoRa Alliance, Luminar, MapBox, Marvell Technology, Mavenir, Marseille Inc, Mayfair Equity, Meraki (Cisco), Merck KGaA, Mesophere, Micron Technology, Microsoft, MiTEL, Mojo Networks, MongoDB, Multefire Alliance, National Instruments, Neat, NetApp, Nightwatch, NOKIA, Nortek, Novumind, NVIDIA, Nutanix, Nuvia (now Qualcomm), NXP, onsemi, ONUG, OpenStack Foundation, Oracle, Palo Alto Networks, Panasas, Peraso, Pexip, Pixelworks, Plume Design, PlusAI, Poly (formerly Plantronics), Portworx, Pure Storage, Qualcomm, Quantinuum, Rackspace, Rambus, Rayvolt E-Bikes, Red Hat, Renesas, Residio, Samsung Electronics, Samsung Semi, SAP, SAS, Scale Computing, Schneider Electric, SiFive, Silver Peak (now Aruba-HPE), SkyWorks, SONY Optical Storage, Splunk, Springpath (now Cisco), Spirent, Splunk, Sprint (now T-Mobile), Stratus Technologies, Symantec, Synaptics, Syniverse, Synopsys, Tanium, Telesign,TE Connectivity, TensTorrent, Tobii Technology, Teradata,T-Mobile, Treasure Data, Twitter, Unity Technologies, UiPath, Verizon Communications, VAST Data, Ventana Micro Systems, Vidyo, VMware, Wave Computing, Wellsmith, Xilinx, Zayo, Zebra, Zededa, Zendesk, Zoho, Zoom, and Zscaler. Moor Insights & Strategy founder, CEO, and Chief Analyst Patrick Moorhead is an investor in dMY Technology Group Inc. VI, Fivestone Partners, Frore Systems, Groq, MemryX, Movandi, and Ventana Micro., MemryX, Movandi, and Ventana Micro.