Quantum approaches could be used to improve the performance of different aspects of machine learning development or use. Which of these areas is best suited to gain the most advantage from the use of quantum? I.e. where should we focus our attention to get the most advantage, and how much advantage can we expect to achieve?

There are several different facets involved in machine learning. Which of these facets is quantum best suited to enhancing? How much improvement can we expect to see when quantum approaches are used in comparison to when classical methods are used alone?

What tools or methods could be developed to support end users in benchmarking the performance of quantum hardware simulations, and their likelihood of being accurate to the real performance on hardware?

End users do expect to gain real business value from the use of quantum hardware in the future, but this will often require them to make significant investment to get access to the hardware. It is possible to test the possible performance of running algorithms on quantum hardware by simulating the devices on classical computers, but it can be difficult for end users to know whether the results they see in simulation will be emulated closely on real hardware – there is some variation in how well simulations perform relative to hardware. Better benchmarking techniques and tools would support end users in investing in quantum computing development, as they can be more confident that the results they see in simulation will be emulated on the hardware.

Where should we focus to obtain greatest business value in the energy sector before full fault tolerance is achieved?

The potential future benefit of fully fault tolerant devices are clear to see. As we await their development, where can we expect to achieve the greatest benefits by using intermediate quantum approaches?

What hybrid approaches to dealing with errors in quantum computing are there to bring value sooner?

A key difference from classical computing and quantum computing, is the prevalence of errors. In classical computing, the high redundancy in how information is stored means hardware is naturally error correcting and there has been little need to consider this. In contrast, for quantum computing errors are a fundamental challenge that we need to deal with. A range of techniques have been developed to tackle errors, from error correction (correcting the error irrespective of the circuit) to error mitigation (using statistical techniques to increase the signal to noise ratio). To get to quantum value sooner, it is likely we will need to take a hybrid approach to get the maximize value out of quantum hardware. This will involve developing novel techniques, understanding how different ideas work together and making sure we are squeezing everything out of the computation (for example using all the extra information available).

6721: How should we load balance quantum and classical resources in heterogenous compute environments?

6749: In your opinion, which steps and platforms are needed for the integration of HPC and quantum computers?

6721: Quantum computing is not going to exist in isolation instead it will be part of a larger computational framework often in a HPC. Unfortunately, the operational requirements for quantum processing units are significantly different from classical computation. For example this includes, the none interpretability of quantum circuits leading to the inability to suspend execution. This means that many of the traditional approaches to load balancing and fair usage do not work or lead to poor performance. This issue is likely to become more acute in regime where multiple quantum processing units co-exist within a single platform, each with different performance characteristics. Novel ideas need to understand how we should model such heterogeneous systems and to support the development and implementation of compatible load balancing and work orchestration protocols.

6749: Useful quantum computing will likely come from the integration of HPC and quantum computers. In this context, a platform for the unified programming of these resources is needed. Additionally, a series of steps that go beyond programming are likely to be needed as well. A description of these steps and timeline would be greatly appreciated.

What can we do to help computer scientist make use of quantum computers

Assuming hybrid execution, using quantum algorithm as subroutines is simple (like using classic ones). Learning how to create quantum algorithms is really hard. What are the choices in the middle and what should the industry prepare that will enable CS people to make as much use of quantum computers as possible?

What support would the quantum ecosystem most benefit from and how can we best strengthen the connection between research and industry?

The Digital Catapult has recently performed a ecosystem map of the supply side of the UK Quantum Ecosystem. Based on this analysis the Catapult network is looking to support the Ecosystem grow a sustainably as possible.

What kind of optimization problems might benefit most from quantum computing? Through which model?

Optimization is thought to be one application where quantum advantage might happen. While some optimization problems have a formulation that suits naturally quantum computing, it is not the case for all of them. In that light, is there a particular structure from which advantage might emerge and if so, is adiabatic quantum computing or gate based quantum computing best suited.

6622: What are the near term opportunities for QML in bioinformatics.

The rise of antimicrobial resistance is such that it is a top ten global health threat and estimates indicate by 2050 it will kill over 10 million people per annum. Hence, the need to develop new computational approaches, including quantum, to genomic sequencing and drug development. QBA is working with the Medicines Discovery Catapult and the EPCC/UK Quantum Software Lab on this challenge.

Our research has two principal areas: firstly, augmenting existing machine learning (ML) approaches by adding quantum layers, and secondly, proposing novel fully quantum models (assuming full-scale fault-tolerant models). Our initial tests and desk studies suggest that a hybrid model could outperform classical models, giving tangible computational advantages in speed, accuracy or produce results on smaller initial data samples .However, our exploration of variational circuit architectures has encountered barriers in efficiently simulating quantum circuits particularly with memory resources needed. Our questions are : which quantum models are resource-efficient and show promise enough to be prioritised? Can we improve on leverage techniques from classical ML, such as transfer learning, to produce quantum models that are versatile enough to apply across a range of viruses and bacteria? For a fully quantised version which approaches would yield nearer term results?

What are your views for programming languages, compilers and circuit optimisation in hybrid clasical-quantum computing?

Quantum programming languages, compilers and circuit optimisation can be accelerated by classical computing, and, specifically, GPU acceleration. Views about this and current work in this field would be very interesting.

6721: How should we load balance quantum and classical resources in heterogenous compute environments?

6749: In your opinion, which steps and platforms are needed for the integration of HPC and quantum computers?

6721: Quantum computing is not going to exist in isolation instead it will be part of a larger computational framework often in a HPC. Unfortunately, the operational requirements for quantum processing units are significantly different from classical computation. For example this includes, the none interpretability of quantum circuits leading to the inability to suspend execution. This means that many of the traditional approaches to load balancing and fair usage do not work or lead to poor performance. This issue is likely to become more acute in regime where multiple quantum processing units co-exist within a single platform, each with different performance characteristics. Novel ideas need to understand how we should model such heterogeneous systems and to support the development and implementation of compatible load balancing and work orchestration protocols.

6749: Useful quantum computing will likely come from the integration of HPC and quantum computers. In this context, a platform for the unified programming of these resources is needed. Additionally, a series of steps that go beyond programming are likely to be needed as well. A description of these steps and timeline would be greatly appreciated.

6755: For which algorithms do you see quantum simulation as most useful? In which sectors? Using which techniques?

6758: Do you think quantum simulation is useful for the development of quantum hardware? If so, which are the fields that could benefit the most?

6755: Quantum simulation can be used in almost every algorithm, but knowing which algorithms will benefit most can be non-trivial. Similarly to sectors that could benefit from applications of those algorithms. Techniques most promising in quantum simulation are another aspect to consider in quantum simulation adoption.

6758: The development of quantum hardware is a challenging effort, where all resources that can be gathered would be useful in principle. Quantum simulation could be one of these. Simulation of noise and quantum error correction are two areas that could benefit, as well as others.

With advances in QuBit error mitigation and suppression, do researchers feel they now have enough logical QuBits to be able to solve any intractable classical optimisation challenges? or is this gong to take several more years of hardware development.

Fire Opal’s error suppression methods are both circuit and hardware-agnostic, meaning that users can be confident that algorithmic performance will be improved regardless of the circuit or hardware type. In the background, Fire Opal uses various methods to address different sources of errors. Just to list a few examples of how Fire Opal reduces error (details can be found in our published paper:

https://journals.aps.org/prapplied/abstract/10.1103/PhysRevApplied.20.024034):
-> Compilation: Reduces the number of gates and the circuit depth/duration. This reduces T1 effects, depolarization, and the effects of noisy gates and bad qubits.
Hardware-aware mapping: Chooses the best qubits to map the circuit to the device, reducing the impact of readout, T1, and gate error.
-> Dynamical decoupling: Addresses dephasing and quantum crosstalk.
-> Gate optimization: Reduces coherent, control, leakage, and decoherence errors in gates.
Measurement error mitigation: reduces the effect of readout misclassification.

Best of all, Fire Opal is deterministic, meaning that you execute your algorithm once and get the best possible result with no need for heavy post-processing. If you want to include post-processing, error suppression reduces the amount required.

https://q-ctrl.com/topics/differentiating-quantum-error-correction-suppression-and-mitigation

6622: What are the near term opportunities for QML in bioinformatics.

The rise of antimicrobial resistance is such that it is a top ten global health threat and estimates indicate by 2050 it will kill over 10 million people per annum. Hence, the need to develop new computational approaches, including quantum, to genomic sequencing and drug development. QBA is working with the Medicines Discovery Catapult and the EPCC/UK Quantum Software Lab on this challenge.

Our research has two principal areas: firstly, augmenting existing machine learning (ML) approaches by adding quantum layers, and secondly, proposing novel fully quantum models (assuming full-scale fault-tolerant models). Our initial tests and desk studies suggest that a hybrid model could outperform classical models, giving tangible computational advantages in speed, accuracy or produce results on smaller initial data samples .However, our exploration of variational circuit architectures has encountered barriers in efficiently simulating quantum circuits particularly with memory resources needed. Our questions are : which quantum models are resource-efficient and show promise enough to be prioritised? Can we improve on leverage techniques from classical ML, such as transfer learning, to produce quantum models that are versatile enough to apply across a range of viruses and bacteria? For a fully quantised version which approaches would yield nearer term results?

What would be an interesting competition-style challenge for the Quantum community?

The task of image classification using AI was greatly boosted by good training data and public challenges to implement algorithms. As an early example of this, ImageNet Large Scale Visual Recognition Challenge (ILSVRC) is widely considered the catalyst that sparked the AI boom.

To formulate an equivalent challenge in quantum computing what concretely is needed in terms of datasets, good metrics to evaluate quantum algorithms and how could such a challenge engage the competitive spirit of the quantum community?

In QunaSys we are trying to do a version of this challenge at qagc.org and are interested in the community’s thoughts.

What do you find most challenging in the bring-up, day-to-day operation, and maintenance of a quantum experiment?

We are dedicated to useful quantum computing. As such we would like to support qubit labs in doing their research to bring this about sooner.

We have expertise and interest in building software automations which make this research easier. Our open-source product, Aqueduct (https://www.riverlane.com/products/aqueduct), aims to simplify quantum workflows. Our upcoming plug-in SDK will allow all users to build extensions that tie in deeply to Aqueduct, including integrations to control systems such as Artiq and to data processing workflows.

We would like to learn more about the processes and challenges in qubit research, and figure out together how we can bring our expertise to bear to accelerate and make easier the development of better qubits and better quantum computers.

We would like to learn more about the processes and challenges in qubit research, and figure out together how we can bring our expertise to bear to accelerate and make easier the development of better qubits and better quantum computers.