Machine Learning and Quantum Mechanics

Nothing has captured the imagination of our collective technoculture more in the past few years than machine learning. Capable of seemingly magical feats of comprehension, such as differentiating photos of dogs from cats, we have embraced this technology as a paradigm of our information infrastructure, and are eagerly looking towards a future where intelligent algorithms shape every facet of our lives.

If these learning algorithms are the future of software, then what will be the corresponding disruptive innovation in hardware? While rumors of the death of Moore's law may or may not be greatly exaggerated, it is clear that our best engineers are worried about the end of real-world exponential scaling of computing power. However, as Joshua Fryman, system architect for extreme scale research and development at Intel states, "There is nothing on the horizon today to replace CMOS. Nothing is maturing, nothing is ready today. We will continue with CMOS. We have no choice. Not just Intel, but everybody."

Then, if processor architecture is destined to be mired in silicon for the foreseeable future, to what hardware disruption will we look forward to in our lifetime? The seemingly obvious answer, borne abundantly by an accelerating online interest, is quantum computers. These mystical devices, perplexingly either 4 years away or 40, promise to revolutionize everything from cryptography to simulation and more (for a complete list, visit the zoo). Uniquely, quantum computing hardware seems to exist in a simultaneous state of being both totally successful, and not even started.

The reality is that quantum computers may be in an even more infantile stage than the other, (perhaps) less exotic replacements proposed for silicon CMOS in conventional "classical" processors. The community of physicists and engineers have yet to even agree on the basic physical implementation of a quantum bit -- although Google, Microsoft, IBM, and others have made recent bets on superconducting quantum circuits. In any case, the obvious marriage of modern machine learning algorithms to quantum hardware seems stalled in the starting gate. This would seem to mean that we will have to wait for the massively powerful conscious intelligence that some people (Andy Rubin among them) purport could be the spawn of this union.

Nonetheless, beyond the blog headlines, there is a serious effort quietly emerging in the corridors and on the blackboards of academia and industry on this nascent field of quantum machine learning. The field is still in the preliminary stages of defining itself. To some, it is the (obvious) plan to implement machine learning on future quantum hardware, or at least, to explore whether any advantage exists in doing so. However, to others it is using existing machine learning algorithms to attack the myriad of ultra-complex problems encountered on a daily basis by physicists studying quantum systems using conventional computers. Finally, to some it is the opportunity to take strategies developed during years of struggle on these ultra-complex problems, and to use this experience to improve machine learning algorithms for the larger community of data scientists.

Whatever it means, it is an exciting time to be involved in quantum machine learning, as we watch the field take shape. It is our hope that this blog will play a small role in of building a community, and to serve as a space for the sharing of ideas about the relationship between quantum mechanics and machine learning.