Opendata, web and dolomites


Teaser, summary, work performed and final results

Periodic Reporting for period 1 - Q-ANNTENNA (Quantum Artificial Neural Networks with Tensor Network Algorithmus)


During the last decades, quantum information processing has gained tremendous momentum: encoding information in quantum particles has opened a vast array of possibilities and, although the vision of building a scalable universal quantum computer is drawing nearer, many quantum...


During the last decades, quantum information processing has gained tremendous momentum: encoding information in quantum particles has opened a vast array of possibilities and, although the vision of building a scalable universal quantum computer is drawing nearer, many quantum technologies have already reached remarkable degrees of maturity, some of which have already percolated to the industry. On the other hand, our society is producing vast amounts of ever-increasing data, growing at an astronomical rate, rendering traditional data processing methods obsolete and leading to the advent of modern machine learning, resulting in key advances for the processing and interpretation of these data sets. This has triggered a revolution in areas so diverse as computer vision, medical diagnosis, voice recognition, spam filtering or search engines, each having a direct impact onto our society and quality of life.

A novel field, quantum machine learning, has emerged from the intersection of the two disciplines, generating a promising symbiosis: quantum resources have the potential to provide the speedup needed to take machine learning to the next level, and ideas from quantum information can in turn be applied to obtain better classical, quantum-inspired algorithms. Interestingly, much of our current understanding of the techniques that underlie the last revolution of modern machine learning have their roots in insights gained from condensed matter and statistical physics and quantum information.

The overall objectives of Q-ANNTENNA are to provide a better understanding among these different fields, bringing together the insights in quantum information that the fellow gained during his PhD and the world leading expertise in many-body physics and quantum computation of the host group, envisioning their application in the novel exciting field of quantum machine learning.

Work performed

The work carried out in Q-ANNTENNA can be divided into four objectives:

The first objective is to deepen our understanding of the relationship between machine learning processes and tensor networks, which provide a faithful, efficient representation of some physically-relevant many-body low-energy quantum states. Thus, minimizing the energy of a many-body system is a key component in machine-learning methods, as well as in general optimization tasks. Here we have provided a new quantum algorithm to prepare ground states of quantum Hamiltonians using less qubits. Focusing on the most immediate short term, we have developed an heuristic version of it within the paradigm of the so-called NISQ (noisy, intermediate-scale quantum) devices. For this part, the tensor network library maintained at the host group has been crucial to the bench-marking of the algorithm. The results obtained have been exploited and disseminated in one publication, a code release, one invited talk in Canada, two contributed talks and two seminars.

The second objective of Q-ANNTENNA is to import physical insights from many-body quantum systems and quantum information, into machine learning processes. This has crystallized into several scientific results mostly in the line of efficient certification of quantum properties (Bell nonlocality, entanglement, correlations/depth quantification, self-testing), which have been disseminated through nine publications, two invited talks, one colloquium, two contributed talks and three seminars.

The third objective of Q-ANNTENNA is to study the renormalization procedure. In here we have developed several methods based on reinforcement learning to assist on the efficient certification of quantum properties. We have also put forward an efficient test to detect Bell correlations in many-body system. The scaling here is constant with the system size, therefore the method describes essentially the same object at different levels resolutions. The results obtained have been disseminated through two publications, two published conference abstracts, three invited talks, one contributed talk, six poster presentations, one seminar and one code release, besides numerous press releases.

The fourth objective of Q-ANNTENNA concerns experimental implementations; i.e., bridging the theory developed during the action with actual experimental platforms, assessing their capabilities and limitations. In an international collaboration, we probed an integrated photonics device technology. On the other hand, the shallow-depth quantum computing algorithm has allowed the fellow to initiate new collaborations with Harvard University, assessing the Rydberg atoms technology. The results obtained have been disseminated through one first-release publication in Science, two conference abstracts, and three seminars, including numerous press releases.

Final results

The impact of Q-ANNTENNA can be viewed from several angles:

First, the action has been very successful on a scientific level, producing more than three times more publications than originally envisioned. The progress beyond the state of the art is thus clear, as the results have been reported in top international peer-reviewed journals. Q-ANNTENNA has pushed the boundaries of knowledge in several directions. To cite the most relevant: The design of a new quantum algorithm to prepare ground states of quantum Hamiltonians that achieves state of the art performance yet using a smaller number of ancillary qubits. Novel methods to obtain certificates of many-body properties assisted by reinforcement learning routines, and exponentially cheaper tests of nonlocal correlations based on convex hulls of semialgebraic sets. Realistically implementable witnesses of nonlocality/device-independent entanglement depth. Reinforcement-learning algorithm as part of a massive physics&outreach experiment (the Big Bell Test). Significant advances in the so-called self-testing problem of maximally entangled states. Implementation of that protocol in an integrated photonics prototype, yielding the highest fidelities reported with this technology so far.
Second, Q-ANNTENNA very positively impacted the fellow\'s carrer development. Its mobility allowance enabled him to strengthen ongoing scientific collaborations (ICFO, Warsaw); effectively co-supervise a PhD student, and grow new collaborations in Basel, Bristol, Leiden and Harvard; carry teaching activities beyond the co-supervision of a PhD student; carry activities as an independent reviewer as a service to the scientific community: in competitive projects, PhD committes and many articles in international peer-reviewed journals. Q-ANNTENNA has also contributed to establish contacts with the industry, including several DAX-indexed german companies, positioning the beneficiary uniquely to be a fruitful liaison between industry and academy.

Finally, all the know-how behind the results has been passed, not only onto researchers and graduate students at the host institution and other institutions during conferences, but it has also reached the general public, thanks to the institutional communication measures taken by MPQ (numerous press releases, open days, interviews, newsletters), but also thanks to personal communication measures proactively seeked by the beneficiary (personal website development, outreach talks, presence in social media, research portals and identifiers, broadcasted conferences in local radio stations, invited inaugural course lectures, etc.).

Website & more info

More info: