Established tech giants, including Google and IBM, as well as startups such as California-based Rigetti and Maryland-based IonQ, are investigating the potential of quantum machine learning. Pictured is an artistic rendering of increasing the understanding of particle collisions through quantum machine learning. *Nature**
Machine learning-based artificial intelligence and quantum computers are two of the most popular areas of research in the technology world. Together, they form a "dream team" known to scientists as quantum machine learning. In a recent report, the British journal Nature** pointed out that scientists are exploring the potential of this future computing alliance and trying to gain insight into the extent to which it will change or reshape the face of science.
It has attracted the attention of technology companies from all walks of life
Established tech giants, including Google and IBM, as well as startups such as California-based Rigetti and Maryland-based IonQ, are investigating the potential of quantum machine learning.
Scientists working in academic research are also very interested. Scientists at CERN are pioneers in the field of quantum machine learning. They have used machine learning to look for "clues" to some of the subatomic particles in the data generated by the LHC. Sophia Valekosa, a physicist and head of CERN's quantum computing and machine learning research group, said they want to use quantum computers to speed up or improve classical machine learning models.
Scientists are trying to answer the big question: Does quantum machine learning have an advantage over classical machine learning in some cases? Theories suggest that quantum computers can speed up computations for tasks such as simulating molecules or finding prime numbers for large integers. But researchers still lack enough evidence to prove that machine learning can do the same. However, some scientists have pointed out that quantum machine learning can uncover certain patterns that classical computers miss, even without increasing computational speed. Other researchers are focusing on applying quantum machine learning algorithms to certain quantum phenomena.
Alam Hartan, a physicist at the Massachusetts Institute of Technology in the United States, said that this is "an area with a fairly clear quantum advantage" in all the proposed applications of quantum machine learning.
Quantum algorithms are not a panacea
Over the past 20 years, quantum computing researchers have developed a large number of quantum algorithms that theoretically improve the efficiency of machine learning. In 2008, Hartan et al. co-invented a quantum algorithm that is several times faster than classical computers at solving large systems of linear equations.
But in some cases, quantum algorithms don't perform as well. In 2018, 18-year-old computer scientist Yiwen Tang invented a new recommendation algorithm that can run and complete calculations on a traditional computer. This algorithm achieves exponential speedup over previous recommendation algorithms and beats a quantum machine learning algorithm designed in 2016.
Ms. Tang said she was "very skeptical" of any claims that quantum algorithms could accelerate machine learning.
However, computational speed is not the only criterion for judging the quality of quantum algorithms. There are indications that quantum AI systems powered by machine learning can learn to recognize patterns in data that classical AI systems miss. Carl Jensen, of the Particle Physics Laboratory at the German Electron Synchrotron Institute (DESY), explains that this may be because quantum entanglement between qubits, allowing the data to establish associations that are difficult for classical algorithms to detect.
How to work better
How to make quantum machine learning work better? The solution that scientists have come up with is to use quantum machine learning algorithms on data in a quantum state, which can avoid the process of converting classical data into a quantum state.
Scientists load these quantum states directly onto the qubits of a quantum computer and then use quantum machine learning to discover patterns without intersecting with classical systems.
Physicists at the Massachusetts Institute of Technology (MIT) conducted proof-of-principle experiments on Google's Sycamore quantum computer. They simulate the behavior of an abstract material with a few qubits, and another part of the processor then takes information from those qubits and analyzes it using quantum machine learning. The study found that this technology measures and analyzes data much faster than traditional methods.
The researchers note that adequate collection and analysis of quantum data allows physicists to solve problems that classical measurements can only answer indirectly. For example, whether a material is in a specific quantum state, making it a superconductor.
Particle physicists are also looking at using quantum sensors to process data generated by future particle colliders, Jensen noted. Distant observatories can also use quantum sensors to collect data and transmit it to the ** laboratory via a future "quantum internet" for processing on a quantum computer.
If this application of quantum sensing proves successful, quantum machine learning could play a role in the measurement of these experiments, as well as in the analysis of the resulting quantum data.
*:Technology**.