En quelques années, de 1900 à 1930, une poignée de jeunes physiciens repensèrent entièrement notre compréhension de la structure de la matière. D’abord accepter que l’énergie puis bientôt toutes les grandeurs physiques étaient discrètes, autrement dit qu’elles ne pouvaient prendre que des valeurs bien précises, dont le spectre s’étend en les répétant par multiplication par des entiers naturels. La physique quantique (de quanta ou quanton, le quantum élémentaire d’énergie — certains la nomment de ce fait physique “intégrale”) était née. Simultanément, mais sans lien direct de cause à effet, si ce n’est que le même génie était à l’œuvre (Einstein), il fallut renoncer à l’existence d’un cadre rigide et statique d’espace et de temps immuable indépendant de son contenu au profit d’un espace-temps dynamique dont la géométrie est dessinée par son contenu de matière. C’était l’explosion de la relativité générale.

Pour décrire l’étrange comportement quantique du monde sub-atomique, un nouveau formalisme naquit (Heisenberg, Schrodinger, Dirac) dont les physiciens se servent depuis et qui leur permirent d’aller de découvertes en découvertes. Mais il fallut aller encore plus loin en sacrifiant notre compréhension spatio-temporelle de ce qui nous entoure. Le noeud gordien prit 50 ans à être tranché. Ce furent les expériences d’Aspect qui démontrèrent sans ambiguïté (par la violation d’inégalités de probabilités, imaginées par Bell) qu’un autre absolu ne tenait plus : la localité. Accepter que les briques de ce monde ne soient plus localisables, mais puissent être dans une juxtaposition d’une infinité de positions spatiales et temporelles possibles comme le pose le principe de superposition d’états. Accepter aussi une conséquence de ce principe, sans doute une des propriétés les plus déconcertantes et contre-intuitives qui puissent être : l’intrication. Qui stipule que deux régions très éloignées d’espace-temps peuvent abriter une entité commune ne formant qu’un.

Et l’on attend désormais qu’une nouvelle génération de physiciens réconcilient physique quantique et relativité générale, théorie incomplète puisque incompatible avec les propriétés de superposition et d’intrication.

Il faut imaginer le choc que l’abandon du vieux cadre physique et l’adoption des nouveaux principes fut pour la communauté scientifique de l’époque…et la nôtre ! On n’efface pas des millénaires de géométrie euclidienne, de mathématique et mécanique classique.

Et pourtant, les succès éclatants de cette nouvelle science, et de ces deux immenses piliers, nourrirent tout le XXème siècle en applications technologiques et industrielles majeures. Du transistor, dont le fonctionnement est fondé sur l’effet tunnel, au laser, lumière cohérente stimulée, aux matériaux supraconducteurs, résultant du comportement grégaire des bosons, de la maîtrise de l’atome et ses innombrables applications dont la fission et la fusion nucléaire à l’IRM, aux horloges atomiques qui mesure le temps et ses effets relativistes et sans lesquelles pas de GPS, c’est toute nos technologies contemporaines bref toute notre vie matérielle qui utilise les propriétés de la physique quantique et de la relativité générale.

Les progrès technologiques réalisés depuis les années 2000 dans les lasers modulables et les matériaux supraconducteurs, conjugués au développement de capacités de calcul de plus en plus performantes, ont permis aux physiciens de réaliser des manipulations de particules uniques et d’explorer cette nouvelle physique quantique dans ses tréfonds afin d’en manifester les effets les plus spectaculaires. Ainsi de l’une de ses propriétés, l’intrication et sa richesse énorme pour coder de l’information, dont les applications en cryptographie et en calcul sont désormais incontournables.

C’est l’intrication des états quantiques qui, en les isolant du reste du monde, les rend capables de ce support additionnel d’information. Dans un ensemble de n objets, pouvant prendre p caractéristiques, la combinatoire nous indique que le nombre d’arrangements avec répétition est en effet de p^n. Qui est égal aussi au nombre d’applications d’un ensemble à n éléments vers un ensemble à p éléments.

L’intrication quantique permet de créer un tel ensemble (isolé du monde extérieur) en rassemblant n qubits qui ne peuvent prendre qu’un certain spectre de valeurs discrètes lorsqu’on les observe. Dans le cas où une mesure de ces états donnent deux résultats possibles, on va avoir ainsi 2^n configurations ordonnées possibles où n est le nombre d’états. Donc 2^n possibilités. Avec l’intrication, on introduit en fait une forme d’ordre dans un ensemble de “chiffres”, on va pouvoir bâtir des mots, avec 2 lettres, 0 et 1, cela donnera autant d’arrangements à répétition soit 2^n.

C’est là que réside le formidable potentiel du codage et du calcul quantique.

Avec 78 qubits, 2⁷⁸ = 40 zettaoctets (qui est, en 2020, la somme des données créées par l’humanité depuis l’aube des temps…).

Ainsi, 250 qubit c’est 2 puissance 250 configurations possibles soit environ 10 puissance 80 qui est supérieur au nombre d’atomes contenus dans l’univers visible. Et l’algorithme quantique traite les 10 puissance 80 solutions potentielles en même temps …L’algorithme bien conçu consistant à faire en sorte que l’amplitude associée à une mauvaise réponse s’annule au cours du calcul par interférence destructive, les amplitudes s’annulant, tandis que les chemins de calcul conduisant à une réponse correcte, qui ont des amplitudes de même signe, voient sa probabilité d’apparaître se renforcer lorsqu’on mesure l’état final des particules.

La seconde révolution quantique dessine ainsi une nouvelle science de l’information. Une informatique fondée sur la puissance quantique, une informatique « intégrale »…

L’explosion des besoins en traitement de données numériques se heurte non seulement aux limites physiques de la microélectronique mais aussi à la part prépondérante et non soutenable des technologies de l’information dans la consommation énergétique mondiale. De nouvelles capacités de stockage, véritables mémoires quantiques, seraient une avancée majeure. Des calculs de physique et chimie des matériaux ou de découverte de molécules aujourd’hui impossibles avec des processeurs classiques commencent à devenir envisageables par simulation quantique. Et cette disruption éco-responsable car beaucoup plus économe en énergie se propage aussi à la cybersécurité avec une nouvelle cryptographie elle aussi fondée sur l’intrication et la métrologie où la sensibilité des qubits à l’environnement est utilisée pour réaliser des capteurs de grandeurs physiques extrêmement performants.

Il est grand temps de rallumer les étoiles », disait Apollinaire. La nouvelle physique nous pousse à ce rêve. Sa révolution ne fait que commencer.

Charles Beigbeder

Merci aux ouvrages du mathématicien Alain Connes et des physiciens Serge Haroche, Nicolas Gisin et Alain Aspect, qui m’ont inspiré ce texte…

La révolution quantique ne fait que commencer ! was originally published in Quantonation, Quantum Investors on Medium, where people are continuing the conversation by highlighting and responding to this story.

High-Performance Computing (HPC or supercomputer) is omnipresent in today’s society. For example, every time you watch Netflix, the recommendation algorithm leverages HPC resources remotely to offer you personalized suggestions. Thanks to the growth of Cloud Computing and sheer computational power, the number of applications is skyrocketing in the industry and academic world, but also in other fields such as cosmetics and finance.

What is HPC?

When people talk about HPC, it is not always clear what that term means to them. HPC has many different definitions which vary from one expert to another.

Let’s dive into the literal meaning of HPC. It stands for High-Performance Computing. The ability to carry out large scale computations to solve complex problems, that either need to process a lot of data, or to have a lot of computing power at their disposal. Basically, any computing system that doesn’t fit on a desk can be described as HPC.

Apart from this conceptual definition, HPC can also be described in terms of hardware. HPC systems are actually networks of processors. The key principle of HPC lies in the possibility to run massively parallel code to benefit from a large acceleration in runtime. HPC systems sometimes reach impressive sizes since applications are accelerated when you add parallelism to the process, which is to say, when you add computing cores. A common HPC capability is around 100,000 cores. Most HPC applications are complex tasks which require the processors to exchange their results. Therefore, HPC systems need very fast memories and a low-latency, high-bandwidth communication systems (>100Gb/s) between the processors as well as between the processors and the associated memories.

We can differentiate two types of HPC systems: the homogeneous machines and the hybrid ones. Homogeneous machines only have CPUs while the hybrids have both GPUs and CPUs. Tasks are mostly run on GPUs while CPUs oversee the computation. As of June 2020, about 2/3 of supercomputers are hybrid machines. They have more computing power since GPUs can handle millions of threads simultaneously and are also more energy efficient. GPUs have faster memories, require less data transfer and are capable to exchange with other GPUs, which is the most energy-intensive part of the machine.

HPC systems also include the software stack. That can be divided into three categories. First the user environment encompasses the applications known as workflows. Then the middleware linking applications and their implementation on the hardware. It includes the runtimes and frameworks. Last, the Operating system, at system level with the job scheduler, management software for load balancing and data availability. Its role is to assign tasks to the processors and organize the exchange of data between the processors and the memories to ensure the best performance.

HPC applications

HPC provides many benefits and value when used for commercial and industrial applications. Applications that can be classified in five categories:

- Fundamental research aims to improve scientific theories to better understand natural or other phenomena. HPC enables more advanced simulations leading to breakthrough discoveries.

- Design simulation allows industries to digitally improve the design of their products and test their properties. It enables companies to limit prototyping and testing, making the designing process quicker and less expensive.

- Behavior prediction enables companies to predict the behavior of a quantity which they can’t impact but depend on, such as the weather or the stock market trends. HPC simulations are more accurate and can look farther into the future thanks to their superior computing abilities. It is especially important for predictive maintenance and weather forecasts.

- Optimization is a major HPC use case. It can be found in most professional fields, from portfolio optimization to process optimization, to most manufacturing challenges faced by the industry.

HPC is more and more used for data analysis. Business models, industrial processes and companies are being built on the ability to connect, analyze and leverage data, making supercomputers a necessity in analyzing massive amounts of data.

HPC: a major player for society’s evolution

HPC needs are skyrocketing. A lot of sectors are beginning to understand the economic advantage that HPC represents and therefore are developing HPC applications. Beyond the historical fields of simulation for industry and academia, HPC is penetrating finance, customer services and medicine.

Industrial companies in the field of aerospace, automotive, energy or defence are working on developing digital twins of a machine or a prototype to test certain properties. This requires a lot of data and computing power in order to accurately represent the behavior of the real machine. This will, moving forward, render prototypes and physical testing less and less standard.

According to research firm Hyperion, the HPC market was worth $27.7 billion US dollars in 2018. They anticipate an increase to a $39.2 billion US dollars market by 2023, which represents a compound annual growth rate of more than 7% driven by these numerous new applications.

The limits of a model

Unfortunately, supercomputers are revealing some limits. First of all, some problems are not currently solvable by a supercomputer. The race to the exascale (a supercomputer able to realize 10^18 floating point operations per second) is not necessarily going to solve this issue. Some problems or simulations might remain unsolvable, or at least, unsolvable in an acceptable length of time. For example, in the case of digital twins or molecular simulation, calculations have to be greatly simplified in order for current computers to be able to make them in an acceptable length of time (for product or drug design).

Moreover, a second very important challenge is the power consumption. The consumption of computing and data centers represents 1% of power consumption in the world and this is bound to significantly increase. It shows that this model is unsustainable in the long term, especially since exascale supercomputers will most surely consume more than current ones. Not only is it technically unsustainable, it is also financially so. Indeed, a supercomputer can cost as much as 10mUSD per year in electricity consumption.

The new chips revolution

CPUs and GPUs are not the only solutions to tackle the two previously stated issues.

Although most efforts are focused on developing higher-performance CPU and GPU-powered supercomputers in order to reach the exascale, new technologies, in particular “beyond Silicon”, are emerging. Innovative chip technologies could act as accelerators like GPUs did in the 2010s and significantly increase the computing power. Moreover, some technologies, such as quantum processors for example, would be able to solve new categories of problems that are currently beyond our reach.

In addition, 70% of the energy consumption in a HPC is accounted for by the processors. Creating new chips, more powerful and more energy efficient would enable us to solve both problems at once. GPUs were the first step towards this goal. Indeed, for some applications, GPUs can replace up to 200 or 300 CPUs. Although one GPU individually consumes a bit more than a CPU (400W against 300W approximately), overall, a hybrid supercomputer will consume less than a homogeneous supercomputer of equal performance.

The model needs to be reinvented to include disruptive technologies. Homogeneous supercomputers should disappear, and it is already underway. In 2016, only 10 out of the supercomputers in the Top500 were hybrid. By 2020, within only four years, it rose to 333 out of 500, including 6 in the top 10.

At Quantonation, are convinced that innovative chips integrated in hetereogeneous supercomputing architectures, as well as optimized softwares and workflows, will be key enablers to face societal challenges by significantly increasing sustainability and computing power. We trust that these teams are ready to face the challenge and be part of the future of compute:

• Pasqal’s neutral atoms quantum computer, highly scalable and energy efficient;
• Lighton’s Optical Processing Unit, a special purpose ligh-based AI chip fitted for tasks such as Natural Language Processing;
• ORCA Computing’s fiber based photonic systems for simulation and fault tolerant quantum computing;
• Quandela’s photonic qubit sources that will fuel next generation of photonic devices;
• QuBit Pharmaceuticals software suites leveraging HPC and quantum computing resources to accelerate drug discovery ;
• Multiverse Computing’s solutions using disruptive mathematics to resolve finance’s most complex problems on a range of classical and quantum technologies.

Thank you very much to the experts that helped us conduct this research including Christelle Piechursky, Cyril Baudry, Jean-Charles Cabelguenne, Jean-Paul Marinier, Julien Nauroy, Eric Rodriguez, et Laurent Seror.

Written by Jean-Gabriel Boinot-Tramoni & Marie Gruet

Please reach jg@quantonation.com if you have any comment

A Beginner’s Guide to High Performance Computing was originally published in Quantonation, Quantum Investors on Medium, where people are continuing the conversation by highlighting and responding to this story.

Quantum Computing startup Pasqal collaborates with the R&D department of electric utility EDF to bring fast solutions to hard optimization problems.

Quantum computers have the potential to solve hard computational problems more efficiently than their classical counterparts. Applications notably encompass computational drug design, materials science, machine learning, and optimization problems. With the rapid developments of quantum hardware, practical quantum advantage is within reach.

With many cities turning to e-mobility to tackle environmental challenges, electric utilities have to account for a growing and more complex load to manage for their production facilities and the grid. One example is the need to schedule resource allocation for shared electric vehicles while taking into considerations their expected and real time availability as well as charging constraints. This class of problem is computationally hard to solve even with large supercomputers and it is expected that a quantum algorithm called Quantum Approximate Optimization Algorithm (QAOA) could improve its resolution.

EDF made smart charging and the development of its infrastructures one of the strong point of its Electric Mobility Plan, launched in October 2018. EDF views smart charging as a true asset for electric vehicle’s users and for the electrical system. Through its subsidiaries, IZIVIA and DREEV, the EDF Group already provides V2G solutions.

Through its Pulse Explorer Program, EDF R&D routinely reaches out to start-ups to explore new ideas in a collaborative way. EDF and Pasqal have formalized a partnership to explore how this algorithm could be implemented on the neutral atoms’ quantum processor developed at Pasqal and take benefit from its unique properties.

The core of the partnership is to finely tune the algorithms according to the hardware’s possibilities and to mitigate the impact of the errors. The level of performance will be gauged on a classical emulator, prior to a real hardware implementation.

Loïc Henriet, head of software development at Pasqal explained: “we have developed our full software stack with specific tools for generic optimization problems, but it is very important that we engage directly with partners working on applications. We need to focus on practical use cases to show that quantum processors can provide a real advantage.”

Marc Porcheron, head of EDF R&D’s Quantum Computing project, said: “utilities such as EDF have to be at the forefront of innovation in high performance computing. It is great to collaborate with Pasqal to explore the new possibilities opened by Quantum Computing for hard optimization problems like the ones we face in the decisive field of smart-charging. I am impressed with the results that have already been achieved with Pasqal, and look forward to implement on their upcoming hardware the quantum algorithms we investigate together.”

To know more

More information on Pasqal’s Quantum Computing stack is available in Pasqal’s technical whitepaper. An abstract is provided here.

Minimizing the total charging time of N vehicles on k distinct charging stations amounts, under some constraints, to solving a combinatorial optimization problem called Max-k-Cut. The objective of Max-k-Cut is to partition the N vertices of a graph into k ensembles, so that the cut is maximal (the cut is the sum of all the weights of edges between nodes that are not in the same group). An example is illustrated on the following figure:

An efficient resolution of Max-k-Cut with a variational procedure involving N ln k qubits has recently been devised by Pasqal’s research team, thus unlocking the potential of Quantum Processors for real-world operational problems. One illustrates in the figure (b) the optimization landscape for one instance of a Max-k-cut problem. As can be seen in the plot, the optimization landscape is very irregular, requiring the design of efficient software methods in the classical optimization loop of the procedure as described in Pasqal’s whitepaper, a genetic algorithm in this instance.

Pasqal and EDF partner to study smart-charging challenges with Quantum Computing was originally published in Quantonation, Quantum Investors on Medium, where people are continuing the conversation by highlighting and responding to this story.

From approximation to prediction of molecular properties for better drugs

Despite the investment of tens of billions of dollars annually, the development of new therapeutic molecules remains expensive, time-consuming and risky. In Silico drug discovery already accelerates the pace of R&D but the precise and accurate simulation of molecules and their interaction with targets has been so far hampered by the complexity of the physics involved at the microscopic level. Thanks to major advances in molecular modelling taking into account highly complex quantum effects, Qubit Pharmaceuticals can perform precise and accurate calculations such as finding a drug binding site, predicting the binding of a drug to its target etc. up to 1 000 000 times faster than traditional high-resolution solutions. Qubit Pharmaceuticals will solve 3 key problems in the field: quality of predictions, interpretability and speed of simulations generating improvement and acceleration of therapeutic pipelines.

Biopharma companies that take the right approach to quantum computing now may gain a lasting advantage according to an analysis of the Boston Consulting Group

Qubit Pharmaceuticals has the vision to become a world leader in quantum-assisted drug development. The company is already taking full advantage of the spectacular improvement in the computing power of supercomputers thanks to its algorithms inspired by quantum physics that run on the most powerful computers in the world. It is establishing partnerships with suppliers of next generation processors in order to develop hybrid classical — quantum approaches within the next five years and full quantum approaches as soon as hardware will be available, which will allow to produce better results faster.

4 Academic labs across the Atlantic, 2 European Research Council grants, the Atos Joseph Fourier Prize and over 15 years of common history

Qubit Pharmaceuticals is the result of the spin-off of the research work of 5 internationally renowned scientists[1] in the United States and in France, collaborating together on the science and the technology for more than 15 years, all 5 providing scientific support to the company. Major technological breakthroughs recently discovered by the founders of the company have been awarded numerous prizes: 2 ERCs (Synergy 2018 and Starting Grant in 2015) and the Atos Joseph Fourier prize in 2018. The use of their technology in the fight to find a cure against COVID-19 convinced the founders to create the company in order to apply their platform to other diseases.

Jean-Philip Piquemal, Chief Scientist of Qubit Pharmaceuticals, said: “Qubit Pharmaceuticals represents more than 15 years of join work from the founding academic groups in France and in the United States. It brings to the Industry a new generation of fast and predictive quantum inspired molecular modelling tools coupled to new High Performance Computing acceleration and visualization strategies enabling enhanced molecular simulations.”

Robert Marino, Qubit Pharmaceuticals CEO added: ”With Atlas and the other tools we develop, we aim to halve the time and cost of preclinical development. Thanks to the predictive power of our tools we want to become the trusted third party for companies in their R&D development. Atlas is a real revolution, the result of a risky but successful gamble, which requires mathematicians and physicists to work together to ensure the robustness of the calculations, HPC specialists to massively accelerate calculations and drug development experts to succeed in finding valuable candidates and understanding data. Without this interdisciplinarity and the scientific excellence of the founders, we would never have been able to achieve such results.

Venture Capital fund Quantonation has worked closely with the team to bring the founders’ project to maturity.Quantonation Partner Christophe Jurczak said: “Quantum computing is an emerging technology with the potential to transform many industries, with biopharma and materials among the most promising. But the pathway to practical application remains under construction and it is clear to us that successful players on the quantum computing software need to be at the forefront of classical computing too ! That’s one of the reasons why we were seduced by the team at Qubit Pharmaceuticals, a unique combination of top experts in quantum physics, chemistry, and high-performance computing. This team has been working on the modeling of the dynamics of objects as large as viruses, they are the best positioned to make the most of emerging quantum computers.”

Qubit Pharmaceuticals is already attracting interest from industry. It’s a laureate of the EIT Health Headstart Program and has joined the Paris Biotech Santé incubator.

[1] Laboratoire de Chimie Théorique, Sorbonne Université — CNRS ; Laboratoire Génomique, Bioinformatique, et Applications du CNAM ; Ponder Lab, Washington University in St Louis ; Computational Biomolecular Engineering Lab, University of Texas in Austin.

Qubit Pharmaceuticals closes a pre-seed round with Quantonation was originally published in Quantonation, Quantum Investors on Medium, where people are continuing the conversation by highlighting and responding to this story.

The application of quantum computers has the potential to reduce the time it takes to compute the solution of hard problems, such as computational drug design, classification challenges, or optimisation problems, compared to current classical supercomputers. Furthermore, quantum computers promise to not be only faster, but also consume less energy. Scientific discovery is poised to benefit most from these new hybrid platforms.

Pasqal was founded in 2019 with the vision to leverage the technology developed at Institut d’Optique in Palaiseau (France) for more than 10 years to build quantum processors based on neutral atoms ordered in large 2D arrays. Pasqal is funded by Quantonation, a Paris based VC fund focusing on Deep Physics and Quantum Technologies.

Pasqal is collaborating with application developers and end-users to explore the capabilities of its architecture and prepare for the upcoming quantum advantage. Each approach to quantum computing (atoms, ions, photons, superconducting circuits…) is based on different concepts with an impact on the software stack, from the physical qubit up to the application. Some parts of the software are specific to the underlying physics, which makes it challenging for software programmers to acquire familiarity and confidence to make the best implementation decisions, ideally regardless of the underlying platform.

With this in mind, Pasqal has been collaborating with Google to give access to its technology through Cirq, an open-source framework dedicated to the development and implementation of quantum algorithms on arbitrary platforms. Pasqal’s API is currently able to run instances on an emulator, while the first processor is being built.

Dr. Markus Hoffmann from Google explains:

“It’s great to see the adoption of Cirq following the spirit of the Apache 2.0 open source license and making further hardware platforms accessible to the Cirq developer community.

Pasqal’s CEO, Dr. Georges-Olivier Reymond, says: “We are delighted to provide such an intuitive and effective interface between our quantum computing hardware and software developers all over the world. This will greatly benefit the emergence of a global quantum ecosystem that is necessary for this new industry to thrive.”

It is to be noted that the development of Pasqal’s API has benefitted from discussions with and support by the team at AQT, an Austrian startup developing a quantum computer based on trapped ions. AQT has been one of the early adopters of Cirq. Dr. Thomas Monz, CEO at AQT, explains that “together with Dr. Markus Hoffmann, AQT happily supports Pasqal and the European quantum ecosystem to work towards a versatile quantum algorithm and quantum application platform.”

Next, we present a tutorial on how to use Pasqal custom classes in cirq (Jupyter notebook).

To know more about Pasqal — Cirq integration and start running programs on Pasqal’s platform emulator:

1. Start working with custom Pasqal objects in Cirq with getting_started.ipynb
2. See how to use Tensorflow Quantum with Pasqal devices with TFQ-cirq-pasqal.ipynb (advanced).

Documentation for working with Pasqal classes in Cirq is available on Github, where the two tutorials are hosted.

Originally published at https://pasqal.io on April 28, 2020.

Bringing arrays of atoms to quantum computing software developers was originally published in Quantonation, Quantum Investors on Medium, where people are continuing the conversation by highlighting and responding to this story.

Today, we are proud to announce a major upgrade to LightOn Cloud service, now available for the AI community worldwide. LightOn Cloud 2.0 features a substantially increased capacity with Aurora 1.5 latest-generation OPUs, the latest version of LightOnML library and simpler pay-per-use payment, booking, and support processes. LightOn technology is now deployed in two datacenters through partnerships with France’s largest cloud providers, OVHcloud and Scaleway.

LightOn Cloud 2.0 users will now have access to increased power and flexibility for the creation of their machine learning models, combining LightOn OPU with a high-end Intel CPU and a V100 NVIDIA GPU. In the past few months, LightOn Cloud users have built a portfolio of use cases with the OPU technology, highlighting its advantage on several neural network architectures. Such examples include Natural Language Processing, Computer Vision, Computational Chemistry or Reinforcement Learning. Most of these examples are explained in blog posts on our Medium with code directly accessible through LightOn AI Research public GitHub.

In such highly-demanding problems, typical training speedups are x8 to a whopping x40 compared to GPU only, at similar accuracies: results are obtained in minutes instead of hours, making LightOn Aurora OPU the perfect technology for data or architecture exploration.

At LightOn, we support research through the LightOn Cloud for Research Program. Machine learning researchers working in academia or non-profit organizations can apply for free LightOn Cloud access. In particular, any use of LightOn’s technology on COVID19 related problems will be prioritized — LightOn AI Research team has already demonstrated how the OPU can be used to detect conformational changes in coronavirus-related HPC simulations. You can find the article here.

LightOn Cloud 2.0 demonstrates that a hybrid approach, featuring beyond-pure-silicon technology, is already here to enhance large scale AI workloads while at the same time lowering data centers’ power consumption.

In order to provide the best quality of service for our users while allowing for the scalable growth of LightOn Cloud and attendant technology, applications to access the commercial service and the Research Program are required. Data scientists and Machine Learning engineers that are interested can apply for LightOn Cloud here. LightOn Cloud for Research Program here.

About Us

LightOn is a hardware company that develops new optical processors that considerably speed up Machine Learning computations. LightOn’s processors open new horizons in computing and engineering fields that are facing computational limits. 🌈

Please follow us on Twitter at @LightOnIO , subscribe to our newsletter and/or register to our workshop series. We live stream, so you can join from anywhere. 🌍

The author

Victoire Louis, Community Builder at LightOn.

Acknowledgments

We thank Igor Carron, Laurent Daudet and Charidimos Chaintoutis for reviewing this blog post.

Welcome to LightOn Cloud 2.0 featuring LightOn Aurora OPUs was originally published in Quantonation, Quantum Investors on Medium, where people are continuing the conversation by highlighting and responding to this story.

These are unprecedented times we’re living in — as we all work to protect our family and friends, our NHS, and our nation from the effects of Coronavirus. First from our family at KETS to yours, we hope you’re safe and well. To anyone dealing with the direct effects of the virus, our thoughts are with you. And to all those in the NHS, on the frontlines, and volunteering a huge thanks from all of us.

As I sit reflecting after my first week working from home, now that we’ve started to get accustomed to our new ways of working in the real world, I wonder how many have taken the time to think about how we’re working in the virtual world? I imagine, like us, many of you were in a huge rush to bring laptops, computers, and other equipment home; to copy all the files you think you might need; to figure out which teleconferencing app you were going to use; and a myriad of other things. You might have even had a home-working setup competition like we did at KETS.

Hopefully with our real selves as safe as can be, now is a great time to take stock of the safety of our virtual selves as well as our online presence which has likely gotten a whole lot bigger working from home.

The National Cyber Security Centre has a huge number of resources dedicated to exactly this for individuals and families, the self-employed and sole traders, start-ups and SMEs, big business, the public sector, and cyber professionals. Particularly good is their article Home working: preparing your organisation and staff as many are looking to exploit our fears over the current situation.

It’s also a great time to take stock of your online vulnerabilities, something you might have been putting off for a while, but now is the perfect time. Again, the NCSC has a great resource called Cyber Essentials to help you protect your organisation from cyber-attack. At the most basic level it contains the following 5 self-help steps to protect yourself and your organisation online:

· secure your internet connection,

· secure the devices and software you’re using,

· control access to your data and the services you’re using,

· protect yourself from viruses and malware,

· and keep your devices and software up to date.

Beyond this, you can apply for a Cyber Essentials Certificate and even get certified by an independent expert.

By implementing these 5 steps we can cut down on something like 80% of the most common cyber-attacks. So now that we’re likely to have gotten to the end of our first week working from home and hopefully done all we can do to ensure our family and friends are safe, it’s the perfect time to improve the safety of our cyber-selves as well.

Keeping yourself and your cyber-self, safe. was originally published in Quantonation, Quantum Investors on Medium, where people are continuing the conversation by highlighting and responding to this story.

Today, about 60% of high-performance computing (HPC) workload performs computational chemistry and material sciences tasks. These undertakings open the door to the discovery of new materials, enhance our understanding of biological mechanisms and potentially allow for the discovery of new drugs.

One popular and straightforward technique for performing such computations is Molecular Dynamics (MD). MD computations follow trajectories of atoms over an extended period of time thereby providing detailed physical information as well as quantitative data on the chemical system at hand. Developed in the early 1950s, the technique has since increased in systems complexity with the increase in computational power afforded by Moore’s law. Larger and larger molecular trajectories can be achieved over longer periods of time.

HPC and computational chemistry developments have been intertwined over the years. On a generic hardware level, HPC architectures have evolved tremendously over the past 60 years as an answer to larger system computations. In parallel, the discovery of new algorithms has permitted ever faster implementations of these simulations (eg. Fast Multipole Method). Eventually, specific elements of the computation pipeline have been required to speed up some aspects of these simulations (e.g. Gravity Pipe (GRAPE) or more recently Anton). While the bulk of the computation is focused on obtaining trajectories, a real insight can only come through the analysis of said trajectories which in and of itself may require large resources.

In this post, we use LightOn’s Optical Processing Unit (OPU) to analyze MD trajectories computed by Cespugli, Durmaz, Steinkellner, and Gruber [1] in the particular case of the coronavirus responsible for the COVID-19 disease.

Molecular Dynamics 101

“…if we were to name the most powerful assumption of all, which leads one on and on in an attempt to understand life, it is that all things are made of atoms, and that everything that living things do can be understood in terms of the jigglings and wigglings of atoms.”

Richard P. Feynman, The Feynman Lectures on Physics

Biomolecular systems such as viruses are indeed made of atoms constantly jiggling. They can take different shapes called conformations. Identifying transitions from one conformation to another one is of biological interest. Indeed conformational changes link the structure of a biomolecule to its function within an organism. Thus, knowing these changes are key in the design of new drugs.

As the number of atoms studied has grown bigger, larger data streams are being produced at every time step of these simulations. As a result, conformational changes can be tricky to identify, especially for large molecular systems, as changes can sometimes be very local. Processing and memory limitations are also in the way of understanding these changes as they require full trajectories to be post-processed. As a consequence, detecting these changes is expensive and is performed by batch thereby requiring many computational resources. This situation calls for a more agile online process.

In the next sections, we examine the detection of SARS-CoV-2 (responsible for COVID19) conformational changes in an online fashion using LightOn’s OPU. Our approach not only speeds up the detection of conformational changes compared to methods using CPUs, but it does so with a lower memory footprint.

Sampling the Free Energy Landscape: What is It Good For?

MD is used to explore the conformational space of large molecules i.e all its possible structures. The challenge is sampling the free energy landscape of our molecule that varies with changes in conformations. MD simulations solve Newton’s laws of motion for a group of many atoms. Since they all move around constantly, their bonds constantly fluctuate and some can do so at high frequency. Hence, the time step Δt used in integration solvers is limited by the highest frequency vibration of the system. In most cases, the Carbon-Hydrogen bond stretching puts an upper bound in that frequency thereby requiring integration steps Δt’s to be of the order of the femtosecond: that’s 10⁻¹⁵ s!

In the lower frequency range, the different conformations of a protein — or metastable states — are separated by high-energy barriers. As a result, transitions between such structures are rare events that occur on a timescale of the microsecond and sometimes up to the millisecond. In order to see a single change in the free energy landscape, we potentially need to perform a sequential integration of Newton’s equations more than a billion times!

To counter this issue, enhanced sampling methods, such as metadynamics [2], have been developed by theoretical chemists. The idea behind this strategy is to modify the potential energy of a molecule by adding an extra term called bias. This bias will decrease the energy barrier between conformations, forcing the system out of the metastable state where it is located. Such algorithms make use of collective variables (CVs) that are one-dimensional coordinates assumed to describe the system in its current state. These CVs enable the “guiding” of the simulation as shown in the procedure featured in Figure 2. By adding a bias, we can push a system out of its “prison” through the addition of an extra force.

Diffusion Maps to Identify Collective Variables

You may be wondering at this stage: what are exactly those CVs, how do we find them? Typically, they can be physical coordinates such as the angle between certain bonds in molecules. In that case, they can be chosen using intuition. Unfortunately, it is not always that simple, especially in very large systems! Several methods have been developed to automatically detect such coordinates.

In particular, Trstanova, Leimkuhler, and Lelièvre [3] proposed a method based on a dimensionality reduction algorithm called diffusion maps (DMaps). This algorithm, first introduced by Coifman and Lafon [4], relies on the computation of the eigenvalues and eigenvectors of a diffusion operator on the data. As an output, we obtain a family of embeddings of our data set into Euclidean space, called diffusion coordinates. It has been argued that those diffusion coordinates directly correlate to CVs [3]. Pretty neat!

The strategy proposed in Ref. [3] goes as follows.

1. Treat batches of m = 2000 time frames as data points. As we are dealing with molecules in three dimensions, their number of features will be N = 3 x n_{atoms}, where n_{atoms} is the number of atoms of the structure.
2. Apply the DMaps algorithm to these points.
3. If the eigenvalues of the diffusion operators have converged, it means the system has reached a metastable state.
4. Using the diffusion coordinates, identify CVs and use them to enhance the sampling (for example, following the procedure illustrated in Figure 1).
5. Detect conformational changes as changes in the spectrum.

However, using the DMaps spectrum to detect conformational changes does not really seem optimal. How do we choose the hyperparameters of the DMaps algorithm? How do we characterize the changes in the spectrum? Do we really have to expand expensive computational power to extract eigenvalues of the DMaps matrix at every m time step — each representing an O(N³) cost?

The answer is: probably not. Indeed, the authors of Ref. [5] propose a method called no-prior-knowledge exponentially weighted moving average (NEWMA) that can solve most of these intricacies and expensive computational issues.

In the next section, we propose a new strategy to enhance sampling in MD simulations by relying on the NEWMA algorithm for the detection of conformational changes. To showcase our method, we study the case of the virus responsible for the Coronavirus outbreak.

NEWMA to Detect Conformational Changes in Molecular Dynamics

On March 11, 2020, the World Health Organization (WHO) has declared the ongoing coronavirus disease COVID-19 a pandemic. This infection is caused by the virus “severe acute respiratory syndrome coronavirus 2” (SARS-CoV-2). To face this challenge, “the greatest since World War Two” according to Angela Merkel, a significant portion of the scientific community has been devoted to the research on the virus. In particular, simulations of the SARS-CoV-2 virus have been conducted to determine the efficiency of certain drugs [1].

We propose to explore the use of the NEWMA method for conformational change detection. This corresponds to replacing the step (5) of the strategy of Ref. [2] by NEWMA, while keeping the rest unchanged. Figure 4 outlines this technique.

The idea behind NEWMA is to compute a detection statistics as the difference between two moving averages. This quantity is then compared to an adaptive threshold. If the statistics are above-said threshold, the algorithm flags that point as a changepoint. This method has the advantage of requiring no prior knowledge about the changepoints and can be computed online. Furthermore, the hyperparameters of the algorithms are selected by a heuristic computed only on the window size — that is, the number of recent samples that are to be compared with older ones. More details about the algorithm as well as the heuristics for its hyperparameters search can be found in Ref. [5].

The detection statistics are calculated using random projections (RP). These RP can be computed on CPU using methods such as Random Fourier Features (NEWMA RFF CPU) [6] or FastFood (NEWMA FF CPU) [7]. They can be computed using LightOn’s OPU (NEWMA RP OPU) as it has the advantage of being mostly insensitive to the number of features — hence, to the number of atoms! — of our dataset. LightOn’s OPU has also a much lower memory footprint as the random matrix does not need to be stored.

MD simulations introduced by Ref. [1] of the virus SARS-CoV-2 do not include any ground truth. That is, we do not know when exactly conformational changes are happening. To circumvent that issue, we compare in Fig. 4 the transitions detected using the NEWMA algorithm to the transitions detected using changes in the spectrum of the diffusion operators (DMaps). As shown in Fig. 5, the results obtained by the two algorithms match.

By using the change points detected by NEWMA, we can reduce the computation of the spectrum of the diffusion operator by a factor 4 (see Fig. 4). This is substantial as the spectrum of the diffusion operator requires the computation of pairwise distances between all atoms in the simulation and also requires the diagonalization of that matrix. As trajectories typically consider a much larger number of time frames, this effect is amplified. By computing the diffusion coordinates only at change points detected by NEWMA, much computational time can be saved!

In summary, in the new MD calculation pipeline shown in Fig. 4 and proposed in this blog post, NEWMA based detection can be efficiently computed at every time step while the coarser-grained, low-resolution DMaps algorithm can be used to obtain the coordinate representation in the manifold dataset when it is needed. This approach leads the way to faster global sampling studies, such as metadynamics.

Where Does LightOn’s OPU Shine in this New Computational Pipeline?

Finally, we also compare the performances of NEWMA using the different methods presented above to compute random projections. In addition to the SARS-CoV-2 trajectories [1], we benchmarked NEWMA using further MD simulations of molecular systems of various sizes provided by authors of Ref. [8]. Results of this investigation are shown in Fig. 6. While NEWMA FF CPU is very effective for smaller molecular systems, at 4000 atoms, computations performed using LightOn’s OPUs are already 40% faster than CPU based methods! Larger systems of atoms should see larger gains.

In summary, using the NEWMA algorithm in tandem with LightOn’s OPU is a very effective way of detecting change of conformations for large to very large molecular systems. While the largest system considered in this blog post had 5335 atoms, some systems such as lipid bilayers may be composed of tens of thousands of atoms (or more!). For such a large system, memory will become an issue for CPU based computations such as NEWMA RFF and NEWMA FF, while NEWMA performed on LightOn’s OPU will remain manageable.

In the future, studying conformational changes for SARS-CoV-2 in complex with drugs may become crucial to the discovery of a cure.

Want to identify conformational changes fast in your own molecular dataset?

The code used to generate the results of this blog post is publicly available here.

You can reproduce our results or make your own calculations with an OPU through our LightOn Cloud, available shortly. You can subscribe here.

LightOn and LightOn Cloud support Research. Please apply to the LightOn Cloud for Research Program on LightOn Cloud website.

About Us

LightOn is a hardware company that develops new optical processors that considerably speed up Machine Learning computation. LightOn’s processors open new horizons in computing and engineering fields that are facing computational limits. Interested in speeding your computations up? Try out our solution on LightOn Cloud! 🌈
Please follow us on Twitter at @LightOnIO , subscribe to our newsletter and/or register to our workshop series. We live stream, so you can join from anywhere. 🌍

The author

Amélie Chatelain, Machine Learning Engineer at LightOn AI Research.

Acknowlegments

We thank Igor Carron, Victoire Louis and Iacopo Poli for reviewing this blog post.

References

[1] Marco Cespugli, Vedat Durmaz, Georg Steinkellner, and Christian C. Gruber. “6 molecular dynamics simulations of coronavirus 2019-nCoV protease model in complex with different conformations of lopinavir” (2020). DOI: 10.6084/m9.figshare.11764158.v2

[2] Alessandro Laio and Francesco L. Gervasio. “Metadynamics: A method to simulate rare events and reconstruct the free energy in biophysics, chemistry and material science”. In: Reports on Progress in Physics 71.12, 2008. ISSN:00344885. DOI: 10.1088/0034–4885/71/12/126601.

[3] Zofia Trstanova, Ben Leimkuhler, and Tony Lelièvre. “Local and Global Perspectives on Diffusion Maps in the Analysis of Molecular Systems”, 2019. arXiv: 1901.06936.

[4] R.R. Coifman, S. Lafon, A.B. Lee, M. Maggioni, B. Nadler, F. Warner, and S.W Zucker. “Geometrics diffusions as a tool for harmonic analysis and structure definition of data: Diffusion maps”. In: PNAS.102(21):7426–7431, 2005. DOI: 10.1073/pnas.0500334102

[5] Nicolas Keriven, Damien Garreau, and Iacopo Poli. “NEWMA: a new method for scalable model-free online change-point detection”, 2018. arXiv: 1805.08061.

[6] A. Rahimi, and B. Recht. “Random Features for Large Scale Kernel Machines”. In Advances in Neural Information Processing Systems (NIPS), 2007.

[7] Q. V. Le, T. Sarlós, and A. J. Smola. “Fastfood — Approximating Kernel Expansions in Loglinear Time.” In: International Conference on Machine Learning (ICML), volume 28, 2013.

[8] Kresten Lindorff-Larsen, Stefano Piana, Ron O. Dror, and David E. Shaw. “How Fast-Folding Proteins Fold”. In: Science 28 Oct 2011, Vol. 334, Issue 6055, pp. 517–520. DOI: 10.1126/science.1208351

Accelerating SARS-COv2 Molecular Dynamics Studies with Optical Random Features was originally published in Quantonation, Quantum Investors on Medium, where people are continuing the conversation by highlighting and responding to this story.

Machine Learning aims at automatically identifying structures and patterns in large data sets. In order to identify these patterns, algorithms often resort to standard linear algebra routines such as matrix inversion or eigenvalue decomposition. For example, support vector machines, one of the most successful traditional machine learning approaches for classification, can be cast to a linear system of equations, and then be solved using matrix inversion. Similarly, identifying the important signals in a data set can be done by identifying the leading eigenvalues and vectors of the data matrix, a method called principal component analysis. The large dimensionality of the vector spaces involved in these operations makes their implementation at large scale very resource intensive, thus motivating the development of innovative methods to lower their computational cost.

A first wave

Researchers at MIT proposed already at the end of the 90s to use a quantum computer to perform these linear algebra routines more efficiently (see reviews [1],[2]). The key idea behind this proposal is that an ensemble of n quantum bits is described by a large 2ⁿ-dimensional complex vector, or quantum state, and operations/gates on these quantum bits by 2ⁿ times 2ⁿ dimensional matrices. By taking advantage of the exponential amount of information potentially encoded in an ensemble of n quantum bits, several quantum algorithms were devised that could potentially outperform state-of-the-art classical algorithms for these linear algebra tasks.

The discovery of these algorithms triggered a lot of research efforts at the intersection of machine learning and quantum computing, but the resource requirements for their implementation in term of numbers of qubits and gate fidelities is unfortunately prohibitive for devices that are and will be available in the short term. Beyond these technological requirements, the main obstacle is the lack of an efficient protocol/device for encoding and accessing classical data into the quantum state, i.e. the need for a so-called quantum Random Access Memory. It is unclear today when such a device could be built and whether it would be cost-efficient. As a nice side effect, however, these results inspired novel classical methods [3] based on random subsampling of the data. Indeed, it was shown that using similar data access as in the quantum algorithms, one is also able to perform classical linear algebra routines exponentially faster on classical computers.

Noisy qubits and new applications

Motivated by these developments, people tried to find other ways to leverage Machine Learning methods on available quantum processors, albeit for other purposes than data intensive applications such as the ones mentioned above. The general aim of these methods is to optimize for a given objective function, and the key idea is to use both a quantum and a classical processor in conjunction, with only a minimal number of operations realized on the quantum processor. In this hybrid framework, the sole use of the quantum processor is to prepare a given quantum state, parameterized by a set of variational parameters. One then extracts from this state an estimate of the objective function through repeated measurements. This estimate is in turn used as the input to a classical optimization procedure, which returns new variational parameters to prepare the quantum state for the next iteration. This loop is repeated multiple times until some stopping criterium is fulfilled, or the optimization procedure converges. Then, a final estimate of the objective function is output.

In this framework, the quantum processor can be seen as a trainable quantum algorithm which output is adapted variationally to best match the task under consideration. Importantly, the variational nature of the procedure renders these algorithms relatively resilient to errors, making them prime candidates for an implementation on Noisy Intermediate Scale Quantum (NISQ) devices [4] that are currently being developped on a variety of platforms and made progressively available to researchers, startups and the general public. The general principle of these hybrid quantum-classical algorithms is illustrated in Fig. 1.

This approach, which bears similarities with classical neural networks, might lead to speedups for certain tasks thanks to the “exotic” correlations that can be encoded in the quantum state. The suitability of such quantum neural networks becomes evident when it comes to learning a complex quantum system, such as a molecule’s energy levels in chemistry. As already Feynman pointed out mid last century, it seems natural to use a quantum system as a computational resource for quantum chemistry calculations. The potential applications in this field are tremendous. It could for example lead to a better understanding of enzymes-ligand interactions and discoveries of new drugs. In a similar vein, we could use this procedure to enhance our knowledge of complex many-body physics effects [5], such as Many-Body Localization or High-Temperature Superconductivity. Beyond these fields, researchers are also actively exploring the use of quantum neural networks as classifiers [6] or for solving combinatorial optimization problems.

Hardware matters

In that respect, not all quantum hardware is equivalent, as the set of operations that can be natively implemented depends on the chosen qubit technology (superconducting circuits, neutral atoms, ions, photons …). One example notably includes the native resolution on a 2D platform of neutral atoms of a well-known graph problem, the Maximum Independent Set (MIS) problem.

Considering an undirected graph composed of a set of vertices connected by unweighted edges, an independent set of this graph is a subset of vertices where no pair is connected by an edge. The objective of the MIS problem is to find the largest of such subsets. This problem, which has various applications in network design or finance, becomes hard to solve on a classical computer when the size of the graph grows.

The MIS problem can be tackled by using an ensemble of interacting cold neutral atoms as a quantum resource, where each atom represents a vertex of the graph under study. As with any quantum system, the dynamics of the atoms are governed by the Schrödinger equation, involving a Hamiltonian (energy functional) depending on the atomic positions, the electronic energy levels and their interactions. Interestingly, the physical interactions encoded in the Hamiltonian constrain the dynamics to only explore independent sets of the graph under study, then leading to an efficient search in the set of possible solutions [7], as illustrated in Fig. 2. This example underlines the prime importance of targeting the trainable quantum network to the specific task at hand, so that the class of trial quantum states that are generated represents good candidates for solving the problem under consideration.

As such, well known notions in the field of Quantum Computing such as gate fidelity or quantum volume are not always the best criteria for assessing the performances of a given quantum hardware on a specific task. Exploring the same phase space by using only standard quantum computing gates in a nearest-neighbor architecture would be far more demanding than with the atoms. Optimisation of resources according to well suited performance criteria for emerging quantum computing platforms is an important topic for future research.

The way forward

The theoretical description of these systems is challenging, as we lack a complete theoretical understanding for such parameterized quantum circuits, similarly to classical neural networks. Additionally, most algorithms are of heuristic nature, and there is no efficient way to simulate large quantum devices, which makes the analysis and understanding even harder. As of today, no proof of a speed-up with respect to the best classical algorithms exists, but research is progressing rapidly, and several key points have been identified to understand and enhance performances:

• the choice of the parameterization determines the range of attainable solutions, which is generally a hard problem on its own.
• the second challenge is associated with the experimental difficulty of sampling enough repetitions of the quantum circuit to estimate accurately the objective function, which can seem prohibitive for some applications.
• another important point concerns the choice of the classical optimizer working in conjunction with the quantum processor, which is instrumental in exploring the parameter space and finding optimal solutions.
• finally, due to the imperfection of currently available quantum processors, one needs to set up techniques to mitigate the effect of errors.

All these aspects are the subject of active academic and industrial research and we understand better and better how each part of these quantum networks behave. For example, it has recently been shown how to compute the analytical gradient of the objective function with respect to the parameters [8]. Research has also led to the creation of classical optimizers specifically designed for quantum neural networks [9] as well as machine learning methods for lowering the sampling requirements [10] or explore the parameter space more efficiently [11].

As a conclusion, this whole new field, let’s call it Quantum Machine Learning, represents a powerful framework to describe and inspire activities in the field of algorithms development for short term NISQ machines. In that regard, the dynamism of the classical Machine Learning community could drive new advances in Quantum Machine Learning, as the core concepts of the field are progressively adapted to the quantum domain.

The Authors

Loïc Henriet is Quantum Engineer and Head of Software at Pasqal, a full stack Quantum Computing startup powered by arrays of neutral atoms

Christophe Jurczak is General Partner and founder at Quantonation, leading VC fund dedicated to Deep Physics and Quantum Technologies

Leonard Wossnig is the CEO and founder of Rahko, a quantum discovery startup developing a Quantum Machine Learning platform for chemical simulation

To learn more

[1] J. Biamonte, P. Wittek, N. Pancotti, P. Rebentrost, N. Wiebe, S. Lloyd, Nature 549, 195–202 (2017).

[2] C. Ciliberto, M. Herbster, A. Davide Ialongo, M. Pontil, A. Rocchetto, S. Severini, L. Wossnig, Proceedings of the Royal Society A, 474, 2209, 20170551 (2018)

[3] E. Tang, Proceedings of the 51st Annual ACM SIGACT Symposium on Theory of Computing — STOC 2019, pp. 217–228 (2018) and arXiv: 1807.04271 (2018)

[4] J. Preskill, Quantum 2, 79 (2018).

[5] C. Kokail, C. Maier, R. van Bijnen, T. Brydges, M. K. Joshi, P. Jurcevic, C. A. Muschik, P. Silvi, R. Blatt, C. F. Roos, P. Zoller, Nature 569, 355 (2019). J. Eisert, M. Friesdorf, C. Gogolin, Nature Physics 11, 124 (2015).

[6] E. Grant, M. Benedetti, S. Cao, A. Hallam, J. Lockhart, V. Stojevic, A. G. Green, S. Severini, npj Quantum Information 4, 65 (2018).

[7] H. Pichler, S.-T. Wang, L. Zhou, S. Choi, and M. D. Lukin, arXiv:1808.10816 (2018). L. Henriet, arXiv:1910.10442 (2019).

[8] K. Mitarai, M. Negoro, M. Kitagawa, K. Fujii, Phys. Rev. A 98, 032309 (2018).

[9]M. Ostaszewski, E. Grant, M. Benedetti, arXiv:1905.09692 (2019).

[10] G. Torlai, G. Mazzola, G. Carleo, and A. Mezzacapo, arXiv:1910.07596 (2019).

[11] G. Verdon, M. Broughton, J. R. McClean, K. J. Sung, R. Babbush, Z. Jiang, H. Neven, M. Mohseni, arXiv:1907.05415 (2019)

Machine Learning in the Quantum Era was originally published in Quantonation, Quantum Investors on Medium, where people are continuing the conversation by highlighting and responding to this story.

How would you pitch your company to our readers?

KETS is pioneering quantum encryption on a chip — both random number generation and key distribution technologies. This is incredibly important because there is a new technology on the horizon called quantum computing. Quantum computing is computing done with quantum systems — i.e. qubits — that has the potential to do many types of computation — like factoring, which underlies some of our current encryption algorithms; or simulation, which could allow us to find the next cancer drug — faster. While a great advancement in many areas, it will create a massive problem for our current encryption algorithms since quantum computers are tailormade to solve the math that underlies them. Many are hard at work on new and better software. At KETS, we also think part of the solution will be new quantum-safe hardware like ours.

Why did you become an entrepreneur?

Good question. A couple of reasons.

1) It sort of got baked into me during my undergraduate degree in Systems Design Engineering at the University of Waterloo. Waterloo (along with a few others like Stanford), pioneered the entrepreneurial-professor model. It’s very common for students, postdocs, and profs to be involved with start-ups in Waterloo. Research in Motion, which created the Blackberry out of Mike Lazaridis’ 4th year engineering project, was right next door. And I worked for a number of start-ups during my engineering co-op terms. So it was just kind of… normal.

2) I think I consider myself an Engineering Physicist, I like building things. For my MSc and PhD, I built an entangled free-space quantum key distribution system that actually worked, in real-time, across the rooftops of Waterloo and the Perimeter Institute (I think there’s still some videos in YouTube land somewhere). It had real world applications that could actually move technology forwards, as did many other quantum technologies that I was immersed in and saw being developed beside me at the IQC. All of this potential, huge potential! But the only way its capitalised on rather than relegated to university display cases and scientific journals is if it is commercialised. Moreover, the only way the grants keep flowing and research can continue is if, in addition to continuing to move human knowledge forward, it also starts to show tangible outputs. I liked the idea of being part of that.

3) And maybe most importantly, with this age of social media and everyone wasting a lot of air being an arm-chair critic, perhaps the best thing about being an entrepreneur is you’re down in the trenches actually doing something! You’re not talking about trying to change the world… you’re actually trying to build something and change a small corner of the world. I think that’s a much more positive way to participate… pick up a screw driver, a laptop, or a whiteboard pen and build something. We seem to have forgotten that a little bit.

What direction do you see Deep Tech moving in?

It’s been really, really interesting to watch deep technology start-ups now starting to win significant private investment — and this includes quantum technologies.

But when you stop and think about it, it’s also quite natural. A lot of the challenges and problems we face are getting too large for academia to tackle. Don’t get me wrong, there’s still plenty of blue sky research to do. And we should continue to fund blue sky research (this funding trend that everything has to have a short-term application is worrying, I’d rather a portfolio approach).

But we’ve sort of exhausted much of the low hanging fruit in some areas. Things like building a quantum computer or maybe AI, are just so massive a task that they really do require a commercial entity to accomplish. One with a much bigger bank account; able to plan an R&D programme out to 5, 10, and in some cases 20 years; and professional project management practices, scientists, and engineers who know how to build products. Sort of the equivalent of the space programme 50 years ago.

The other direction that’s neat is deep-tech is becoming cool 😊 . Which I hope means a lot more school kids looking at perhaps what they thought of as their boring math class or science class and saying “yeah, it’s worth the slog because I want to design the next SpaceX rocket, or quantum sensor, or green energy source.”

Finally, I think the other interesting trend in deep tech and the people that fund it, is knowing that they have a secondary goal of just building the community and a bit of a deep-tech mixing pot. It’s that longer term vision.

Look, I’m sure that quantum technologies will go through a bit of a winter (like other technologies) because people are trying to do hard things and build very technically challenging devices. But there’s certainly a few VCs I’ve now met and many entrepreneurs (I had a great chat with Andrew Fursman from 1QBit about this) that understand this now and know that it might not be the first idea, or the second, or even the third, but eventually someone will hit on something that changes the world. And taking a long-term view, the way they support that is by supporting those riskier ventures now and backing good people. They’ll try and fail or succeed, reshuffle the people, bring more in, try again, and eventually we’ll watch someone converse with an android or walk on Mars.

And very finally, it’s good to see people investing in hardware again… challenging hardware… that could change the world. Don’t forget, all of these massive apps needed the smartphone first.

What’s your best memory at KETS?

There’s too many to choose from, but I’ll pick a recent one. I hadn’t been in the lab in a while and even though I get updates in meetings on how we’re doing, it’s not like seeing it in the flesh. So about a week or two ago I had a visitor, it was a quite day and I took them up to the lab at the end of the conversation to show them our stuff….and I didn’t recognise it!

Far from the physics project that it started out as (that literally was held together with some cellotape at various points), I was pointing at a slick RF break-out board with one of our little Quantum Key Distribution chips packaged on its own little daughter board, similar to a little SD card, and plugged into the main board. Our technology looked like a product! That was cool. And I felt so proud of how far our team has come in turning a university technology into an actual commercial product.

One common view you are sick of hearing?

“You’re a little early for us.” Typical response from an investor when we were looking for funding. Certainly, that got old, fast — but it’s par for the course when you’re hunting for money until you find the right investors (like Quantonation!) that get your vision.

Maybe the other one is the view that tech can save us from all of our problems. It’s maybe a weird thing to say coming from someone developing a deep-tech quantum encryption company, I’ll give you that. But it’s born from my childhood spent up at the cottage in Northern Ontario, Canada. It was during those formative years that I grew a deep affection for nature — it’s probably the closet I come to religion. And seeing rampant, lazy consumerist waste now is deeply upsetting.

Tech certainly has a role to play in solving many of the world’s problems. But we’ve become lazy as a species. And if we can just curb the worst of our bad habits — make a bit more food at home and create less takeway containers, turn off lights in an empty room, carry a reuseable coffee cup, and maybe wait one more year before buying the latest phone — then at the very least, tech will have a smaller problem that it can solve sooner.

Is Europe still in the race for Deep Tech ?

Absolutely! Absolutely… there’s so many cool things happening in deep tech in Europe. I mostly know quantum tech and some of it started in large part because of the big bet the UK placed with their National Quantum Technologies Programme 5 years ago and the 4 Hubs. But now you have the European Flagship project kicking off with all sorts of interesting projects. You have further EU calls to setup things like QKD testbeds and start thinking about the quantum internet. You’ve got QuTech in Delft (note, QTEC at the University of Bristol had the name first! Sorry, couldn’t help myself, our own fault for not trademarking the name 😊), Fraunhofers in Germany (which have been there for years doing good work).

What’s more, it’s a bit of an untapped market since it’s behind on the whole entrepreneurship thing and only becoming mainstream and cool now. Moreover, investment… serious investment in early stage, riskier companies, is now starting to get involved. And finally, it’s the different perspectives that Europe has compared to what they might have in places like the US, that made start-ups cool years ago. Diversity is usually where the next biggest ideas come from.

How will Proprietary Tech change the world?

I almost want to be contrarian and say that Proprietary Tech will change the world… when we learn to share it. What do I mean by that? I think there is this increasing movement in the world to be more collaborative, universities trying to collaborate with industry, start-ups with big companies, neighbourhood communities with artists and the local councils. And some of the biggest breakthroughs have come when people have put a multi-disciplinary team together. Right now, this happens when you have something the size of Google or Amazon put resource behind something, but I do not think that is the only way, nor should it be because then a very few people are deciding the agenda. And I appreciate the catch-22, new start-ups need to protect their ideas so they can benefit from what they create and continue to get the investment they need to grow. But I do think there will be a movement to share and collaborate more, and new systems worked out to do that.

What’s your favorite sci-fi?

Sadly, I’m the usual stereotype — loved Star Trek and Star Wars as a kid (and yes, I loved The Last Jedi), played all of the Mass Effect series… twice, on Xbox. I am re-reading IT at the moment — I think anyone who read it as a kid is drawn back to the story about the power of belief in kids. In terms of the last sci-fi movie I saw, I loved A Quiet Place — the use of sound was just so effective and visceral.

And I am itching to see Apollo 11 — I think space attracts me because we have quickly forgotten how hard it is and how massive an undertaking it was. Essentially a whole nation coming together for a decade, to transport a few people to the moon and back. And we all look up at the sky when we’re out at night walking home, wondering what is up there? (And not sci-fi, but a must-see movie I just saw with my wife is Booksmart. Laughed… our… a$$es off 😊).

One prediction for 2019?

That someone, somewhere experiences a crippling cyber attack. And worst of all, we might never hear about it…

Thanks Chris !

About Quantonation’s investment in KETS

Early stage Deep Physics focused venture fund Quantonation, together with Kx and others, invested over £2 million in KETS in December of 2018, including £1million provided from Innovate UK and the Industrial Strategy Challenge Fund.

More Information: website, Twitter

KETS Quantum Security is a University of Bristol start-up company developing a range of future-proof, cost-effective technologies for quantum-secured communications that have the power to improve the secure transmission of information such as banking details and medical records.

Towards a quantum proof future was originally published in Quantonation, Quantum Investors on Medium, where people are continuing the conversation by highlighting and responding to this story.