AI companies plant the seeds for quantum machine learning

Quantum computing promises to accelerate analytics faster than the speed of light, but it still feels slightly unreal, in spite of the first signs of broad commercialization

AI companies plant the seeds for quantum machine learning

Quantum isn’t the next big thing in advanced computing so much as a futuristic approach that could potentially be the biggest thing of all.

Considering the theoretical possibility of quantum fabrics that enable seemingly magical, astronomically parallel, unbreakably encrypted, and faster-than-light subatomic computations, this could be the omega architecture in the evolution of AI (artificial intelligence).

No one doubts that the IT industry is making impressive progress in developing and commercializing quantum technologies. But this mania is also shaping up to be the hype that ends all hype. It will take time for quantum technology to prove itself a worthy successor to computing’s traditional von Neumann architecture.

Though the splashy headlines boast of quantum supremacy, which refers to claims that programmable quantum devices can solve problems beyond the reach of von Neumann architectures, there has been far less focus on quantum practicality. In other words, there is still little evidence that quantum computers are being applied to real-world use cases in AI, ML (machine learning), and other advanced analytics. 

Quantum’s long ramp into mainstream AI and ML

AI has been singled out as a quantum killer app for quite some time, but quantum so far has minimal presence in the commercial data analytics arena.

We need to ask ourselves whether all the recent industry activity is setting ourselves up for the dreaded “quantum winter,” analogous to the long AI winter period of backlash against the early hype for that technology. It’s one thing to talk about the mind-blowing potential of quantum analytics that executes across parallel universes as if it were the holy grail. It’s quite another to point to a mature technology with clear killer apps that make a huge difference in our lives today.

Nevertheless, there is a growing sense among researchers and even among analytics professionals that ML could become the core use case for quantum in our lives.

This is not a recent revelation. An MIT professor declared in 2013 that the “first quantum application” is ML. Specifically, the professor, Seth Lloyd, proposed a “q-app” that “encodes Google-like queries with q-bits that enable quantum computers to not only perform real-time searches through even the most gigantic databases, but which also ensures their absolute privacy, since attempts to eavesdrop on the query by the search engine provider would disturb the delicate q-bit’s superposition of states.”

More significantly, recent product launches and other announcements by Amazon Web Services, Microsoft, IBM, and Honeywell in the quantum computing space address AI and ML use cases to varying degrees. None of these announcements pertains to a generally available quantum product or service that’s solving practical business problems. However, most of these announcements include hooks for programmers to develop such solutions on quantum hardware platforms or cloud services.

In November 2019, Microsoft announced Azure Quantum. This quantum-computing cloud service is currently in private preview and expected to become generally available later this year. It comes with a Microsoft open source Quantum Development Kit for the Microsoft-developed quantum-oriented Q# language as well as Python, C#, and other languages. The kit includes libraries for development of quantum apps in ML, cryptography, optimization, and other domains.

A month later, AWS announced the Amazon Braket service. Still in preview, this is a fully managed AWS service that enables scientists, researchers, and developers to begin experimenting with computers from quantum hardware providers (including D-Wave, IonQ, and Rigetti) in a single place. It provides a single development environment to build quantum algorithms—including ML—and test them on simulated quantum computers. Developers can run ML and other quantum programs on a range of different hardware architectures. And it allows users to design quantum algorithms using the Amazon Braket developer toolkit and use familiar tools, such as Jupyter notebooks.

Then in January, IBM announced the expansion of its Q Network, in which more than 200,000 users are running hundreds of billions of executions on IBM’s quantum systems and simulators through the IBM Cloud. Participants in the network have access to IBM’s quantum expertise and resources, open source Qiskit software and developer tools, as well as cloud-based access to the IBM Quantum Computation Center. Many of the workloads being run include ML, as well as real-time simulations of quantum computing architectures.

Less than two weeks ago, Honeywell announced that its high-capacity quantum computer will be generally available within three months. It also announced that Honeywell Ventures is making investments in Cambridge Quantum Computing and Zapata Computing. Both companies have expertise in ML and other cross-vertical algorithms and software for quantum computing applications.

Quantum tools have come into the dominant AI/ML development framework

The most important announcement, coming just a few days ago, was Google’s launch of TensorFlow Quantum. This new software-only stack extends the widely adopted TensorFlow open-source ML library and modeling framework to support building and training of ML models to be processed on quantum computing platforms.

Developed by Google’s X R&D unit, TensorFlow Quantum enables data scientists to use Python code to develop quantum ML models through standard Keras functions. It provides a library of quantum circuit simulators and quantum computing primitives that are compatible with existing TensorFlow APIs.

TensorFlow Quantum’s release is no big surprise, coming several months after Google declared  “quantum supremacy,” which refers to its achievement of a quantum computing feat that would have been impossible on traditional computing architecture.

In addition to providing a full AI/ML software stack into which quantum processing can now be hybridized, Google intends to expand the range of more traditional chip architectures on which TensorFlow Quantum can simulate quantum ML. It has announced plans to expand the range of custom, quantum-simulation hardware platforms supported by the tool to include graphics processing units from various vendors as well as its own Tensor Processing Unit AI-accelerator hardware platforms.

Recognizing that quantum computing is not yet mature enough to process the full range of ML workloads with sufficient accuracy, Google has wisely designed its new open source tool to support the many AI use cases with one foot in traditional computing architectures. TensorFlow Quantum enables developers to rapidly prototype ML models that hybridize the execution of quantum and classic processors in parallel on learning tasks. Using the tool, developers can build both classical and quantum datasets, with the classical data natively processed by TensorFlow, and the quantum extensions processing quantum data, which consists of both quantum circuits and quantum operators.

Developers can use TensorFlow Quantum for supervised learning on such ML use cases as quantum classification, quantum control, and quantum approximate optimization. They can also execute advanced quantum learning tasks such as meta-learning, Hamiltonian learning, and sampling thermal states.

In addition, Google has designed TensorFlow Quantum to support the growing range of AI use cases, such as “deepfakes” that do video, voice, and image generation with a high degree of verisimilitude. Google ML developers can use TensorFlow Quantum to train hybrid quantum/classical models to handle both the discriminative and generative workloads at the heart of the generative adversarial networks used in such applications.

Strategically, Google’s likely next move will be to combine TensorFlow Quantum with its pre-existing Quantum Computing Playground into a full-featured, managed quantum ML service. Considering the fact that Google’s top public-cloud rivals (Microsoft, AWS, and IBM) all have such services either on the market or in preview, it would be shocking if the Mountain View, Calif.-based company doesn’t try to one-up them with a quantum ML service of its own.

Quantum AI/ML researchers lay down track before their speeding train

Building “write once run anywhere” quantum ML will be tricky for quite some time, even in TensorFlow Quantum. This is because quantum researchers are experimenting with a wide range of alternative architectures even while they try to build practical applications.

Here are some of the principal ways the leading quantum ML vendors are supporting these more fundamental quantum R&D requirements:

  • Google designed TensorFlow Quantum to support advanced research in alternative quantum computing architectures and algorithms for processing ML models. This makes the new offering an invaluable research tool for computer scientists who are experimenting with different quantum and hybrid processing architectures optimized for ML workloads. To this end, TensorFlow Quantum incorporates Cirq, an open source Python library for programming quantum computes. It supports programmatic creation, editing, and invoking of the quantum gates that constitute the Noisy Intermediate-Scale Quantum (NISQ) circuits characteristic of today’s quantum systems. It also enables developer-specified quantum computations to be executed in simulations or on real hardware.
  • Microsoft’s Azure Quantum includes libraries for simulation of alternative quantum circuit processing scenarios and prediction of likely program performance in these environments.
  • Amazon Braket allows users to explore, evaluate, and experiment with quantum computing hardware, design quantum algorithms, and simulate performance of their programs’ execution either on low-level quantum circuits or fully managed hybrid algorithms.
  • IBM Q Network supports real-time simulations of alternative quantum computing architectures.

Mainstreaming Quantum ML in the ’20s

Quantum computing has been in wait-and-see mode for so long that we tend to overlook the fact that it’s being rapidly put to practical uses.

Even as quantum computing platform vendors experiment with new materials, methodologies, and architectures, researchers around the world have been demonstrating that quantum processing of ML models is in fact feasible. We can expect that AI/ML researchers at Google and elsewhere will probably use TensorFlow Quantum to do some fairly amazing things that were never feasible on traditional AI-accelerator hardware platforms.

It’s clear from all these recent industry announcements that we’ll not only see commercialized quantum ML in our lifetime but that it has already begun to emerge and will gain steady adoption in this decade.

Copyright © 2020 IDG Communications, Inc.