In space, there is less space than ever. Growing numbers of new satellites are entering an atmosphere already crowded with old space debris, creating greater complexity and a rapidly increasing amount of data for satellite, space, and defense systems to process and analyze.
Machine learning (ML) technology is already being used in these sectors to filter, de-duplicate and analyze large amounts of raw collected data to create more manageable datasets. ML models can apply techniques like pattern recognition to generate insights and even initiate decisions to autonomously act on those insights.
All of that is of tremendous value for satellites, aircraft and ground-based systems that sometimes need to deliver real-time analysis of space separation events or other incidents, as well as processing of satellite images.
But, the task will only continue to get bigger. The European Space Agency has said there could be 100,000 satellites in orbit by 2030 as mega-constellation operators ramp up their ambitions. At the same time, more space-based sensors are being packed onto more Earth Observation satellites, enabling them to generate much more data, including hyperspectral images and other high-resolution image data; temperature, weather and related environmental data; and geo-location data critical to military and defense operations and other use cases.
Given these challenges, even the advanced machine learning capabilities being used now could use a boost. That boost could be found in the realm of quantum computing, where nascent quantum machine learning technology has shown in early demonstrations that it can enable improved capability to process and manage data sets, and provide faster and more accurate data analysis.
“The satellite and space industry is an interesting and emerging market for quantum computing because of the growing complexity of its unique mission challenges.” -Murray Thom, D-Wave Quantum
“The satellite and space industry is an interesting and emerging market for quantum computing because of the growing complexity of its unique mission challenges,” said Murray Thom, vice president of quantum technology evangelism at D-Wave Quantum, a Palo Alto, Calif.-based, company whose quantum annealing systems represent a particular form of quantum computing targeted at large, multi-variable optimization problems.
What is QML?
Quantum machine learning works in a fundamentally different way than classical computing ML. Classical ML algorithms rely on linear algebra. Classical bits encode information as either 0s or 1s to perform calculations to determine an exact answer. QML algorithms employ the probabilistic nature of quantum mechanics through principles such as superposition and entanglement. In a quantum computer, superposition allows bits (or in this case, qubits) to exist simultaneously as either 0 or 1, or both, enabling parallel processing of a variety of different computational paths. Entanglement expands this effect by linking multiple qubits into a single, collective quantum state to enable faster, greater processing power.
A QML algorithm can allow a computer to execute multiple computing tasks at the same time, including highly complex computations involving many variables.
What all this translates to is that a QML algorithm can allow a computer to execute multiple computing tasks at the same time, including highly complex computations involving many variables, and generate more accurate predictions in a much shorter amount of time than a classical ML algorithm would in tackling computing tasks one at a time.
That gives quantum computers an ability to tackle certain complex problems that may be intractable for classical computers to take on. So, it’s not just a matter of speed.
Abhishek Chopra, founder and chief executive and scientific officer at BQP (formerly BosonQ Psi), whose model-based systems engineering software platform allows “quantum-inspired” simulations, told Constellations, “I like to give this analogy, which I’ve heard from some of the [quantum computer] hardware companies: Don’t think of quantum as a car versus fast car analogy, but think of as a car versus boat analogy in that with quantum, you are able to get to some place which was not possible before, and do so very efficiently.”
A Layer of Quantum
The next thing to know about quantum machine learning is that it does not necessarily require a quantum computer. QML algorithms are designed to be run on quantum computers, already have been successfully demonstrated on quantum computers, and will run most optimally on quantum computers in the future.
For example, quantum software start-up Artificial Brain has run its Planck QML software (named after quantum theorist Max Planck) on quantum computing systems from D-Wave and others, said Dana Linnet, chief strategy advisor at Artificial Brain. In fact, the company last year won the Prototype Track in the Deep Tech Category of the myEUspace competition, organized by the European Union Agency for the Space Programme (EUSPA), for a QML algorithm designed to optimize real-time scheduling for multiple Earth Observation satellites.
However, some QML benefits also can be achieved using “quantum-inspired” software and algorithms that run on classical computers but use data compression and other techniques to mimic the quantum effects of superposition and entanglement.
That is good because many quantum computers today cost well over $1 million and often are described as “noisy” and error-prone compared to their classical counterparts. Although, for perspective, “error-prone” in this case might mean being 98.5% accurate rather than 99%, and many current quantum computers and quantum annealing systems like D-Wave’s have demonstrated the ability to excel at specific problems in a variety of industries. Still, fully fault-tolerant quantum computers remain at least five years away if you believe the earliest estimates.
“We are basically adding onto existing machine learning models just a layer of quantum, and that is actually reducing trainable parameters of those models." -Abhishek Chopra, BQP
“Currently we do quantum without quantum,” said BQP’s Chopra. “We are basically adding onto existing machine learning models just a layer of quantum, and that is actually reducing trainable parameters of those models. So, basically compressing the models” so that they can be run faster and more efficiently on classical computing resources like GPUs and CPUs, and provide more accurate and even real-time analysis.
Using QML in Satellite, Space and Defense
The satellite, space, and defense industries are home to numerous examples of how increasing numbers of complex variables can lead to challenges that press the limits of classical machine learning.
“The potential use cases [in satellite, space, and defense] are plentiful,” said D-Wave’s Thorn. “D-Wave and its customers have explored applications such as optimizing satellite image acquisition and optimizing launch pad scheduling. In addition, we are simplifying the process of building and deploying solutions for fast and reliable quantum optimization. For example, D-Wave has an open source demo that shows how to maximize satellites’ constellation observation, and its technology has also been used to optimize manufacturing operations for cost efficiencies.”
Thorn, acknowledging D-Wave’s work with Artificial Brain, said the start-up demonstrated how quantum computing capabilities can enhance “the efficiency and accuracy of EOS mission planning.” He also mentioned GMV, a space and defense technology company in Europe that he said is currently evaluating how quantum can be used in satellite control centers.
“Along with private and public partners, the research team at GMV explored data encoding and problem modeling through quantum machine learning and combinatorial optimization,” he said. “These techniques were then applied in the context of Earth observation mission planning.”
While mission planning is one potential area of application, it probably is not the most urgent one. As Linnet of Artificial Brain explained, space sustainability continues to be a pressing problem.
“There’s so much space junk up there,” she said, “and yet [the space industry is] going to deploy even more satellites. We could see triple the number of satellites in the not-too-distant future. The amount of space junk that is becoming dangerous has proliferated even in the last couple of years.”
In response, Artificial Brain developed a space sustainability solution for satellite scheduling optimization, an application area that has emerged as a key one for machine learning, as it helps satellites draw from large volumes of data to determine how to prioritize the assignment of various tasks.
Linnet said companies in the space market “have to schedule across a lot of different assets,” and have a lot of choices about what kind of satellite data to grab and from where. Adding to the complexity, military organizations, while leveraging their own satellite imagery and data, also draw on images and other data captured by commercial satellites that can help them enhance mapping, targeting, and the tracking of troop movements.
“There are so many variables and so many factors, and we can’t rely solely on AI for that anymore because the answers are going to still take too long.” -Dana Linnet, Artificial Brain
“How do they know what to prioritize, what to look at, how, when, what the capabilities are, what the power loads are, what the time from ground link to space is?” Linnet said. “There are so many variables and so many factors, and we can’t rely solely on AI for that anymore because the answers are going to still take too long.”
She said that adding QML to the mix can help deliver those answers more quickly and efficiently, while allowing for the consideration of variables like a national security organization having a higher priority for access to certain satellite imagery.
Linnet, Thorn and BQP’s Chopra also all cited anomaly detection in satellite imagery as another application for QML.
“There could be a certain event that is happening in space,” Chopra said. “Is it a planned separation event, or is it an adversary knocking out one of our satellites? For that kind of image classification problem, your current classical methods, your data-driven methods, are not enough because you don’t sometimes have enough data to say whether it is an adversary event or it’s a friendly event… but when you bring your data and physics together, your [machine learning] models become very, very large.”
That is where quantum’s ability to help compress machine learning models comes into play. Chopra said BQP has shown in published papers that in such an anomaly detection problem, its quantum-inspired ML approach “can reduce the trainable parameters from 14.6 million, which is the size of a classical deep learning model, to 2,000 in a quantum [machine learning] model.” This reduction by an order of magnitude makes the problem manageable for classical computing resources like GPUs to handle.
BQP’s ability to apply QML to anomaly detection in satellite images led to the being selected for a Space Domain Awareness Tap Lab accelerator program in Colorado Springs. There, BQP showcased how its technology could be used in real-time satellite tracking and space battle management scenarios.
“With classical orbital mechanics and numerical methods, it takes quite some time to get the kind of accuracy you need for these kinds of applications, or sometimes you don’t even get real-time accuracy,” Chopra said. “Our background is modeling and simulation, and the quantum part is the how—how we’re solving the problem of time and accuracy bottlenecks in complex simulations.”
Linnet added that quantum-enhanced anomaly detection could even be used to monitor carbon emissions from cargo ships at sea to ensure compliance with environmental regulations.
Artificial Brain also is promoting QML for use in anomaly detection in hyperspectral images–dense multi-dimensional images that contain sensor-based data from across the electromagnetic spectrum.
“Clients who collect this data often have to manually label it or rely on slower AI solutions,” Linnet said. “Our quantum machine learning vector can auto-label hyperspectral data with just a handful of reference points, unlocking value from billions of images that would otherwise go unused.”
Sparking QML adoption
QML is not in wide usage yet beyond the occasional demonstrations and trials, but private and public space organizations from around the world are starting to take notice of what QML can bring. According to the companies quoted in this story, NASA, the ESA, the UKSA (UK Space Agency, the ISRO (India’s space agency), and other government agencies in different countries all are exploring quantum technology for the kinds of applications mentioned above. For example, Thorn noted that the EU has an open call for applications to identify quantum solutions for air traffic management.
There is something of a quantum “space race” occurring globally right now as countries around the world try to strategically plan for the eventuality for fault-tolerant quantum computers that will impact everything from vertical industries to national cyber security.
There is something of a quantum “space race” occurring globally right now as countries around the world try to strategically plan for the eventuality for fault-tolerant quantum computers that will impact everything from vertical industries to national cyber security. While the US and China have been seen as leaders in developing quantum computing technology, many people in the quantum technology sector believe that China’s coordinated, intensely-driven national strategy is allowing it to move ahead. European countries, individually and together, also have intensified their efforts, as have Japan, Australia, and multiple countries in the Middle East.
While the US is not necessarily falling behind, there have been rumblings that it could lose some of its early-mover advantage without more coherent national policies that include incentives for adoption of quantum technology across the private and public sectors. Portions of the original National Quantum Initiative (NQI) Act passed by Congress in 2018 already have expired, and multiple bills seeking the re-authorization of the NQI Act have languished in Congress through the last year of the Biden administration, and now the first year of the second Trump administration.
As of this writing, it was widely reported that President Trump could be close to issuing an Executive Order related to quantum-safe security, as well as to maintaining US leadership in the quantum and AI sectors. Further details remained unclear, but any more aggressive quantum push at the federal government level likely would help drive awareness and adoption of technologies like QML, among others.
“Deploying quantum would provide a huge value benefit to US competitiveness,” Linnet said. “We need to aggregate capital, talent, and policy support, or risk a ‘quantum Sputnik moment’ where others leap ahead. There’s so much good we could be doing with this technology. The time to embrace it is now.”
Dan O’Shea, a freelance writer and editor, covers the quantum technology sector at Quantum News Nexus. You can find him on X @QuantumNewsGuy.
Explore More:
Satellites Are ‘Key’ to Global Quantum-Safe Communication
Quantum Cyberattacks: A Future or Current Concern for Satellites?
Debris: Examining Altitudes, Anomalies and Attributions
