Machine learning for Quantum Control and Quantum Computing, ML(QC)2

29 Aug 2022, 09:00 → 2 Sept 2022, 18:00 Europe/Stockholm

Conference center, room ... (Albano Building 3)

Anton Frisk Kockum, Eliska Greplova, Evert van Nieuwenburg , Mats Granath, Roger Melko

Scope

Quantum systems are inherently prone to noise and decoherence causing information loss. Given the enormous effort devoted to quantum technology and quantum computation, quantum control — i.e. how to preserve and manipulate quantum states — is a topic of great current interest and potential impact. Researchers have recently started to tap into the vast potential of machine learning (ML) to address many areas in physics, including quantum control and quantum error correction. The aim of the workshop is to build on and consolidate recent advances in using state-of-the-art ML techniques for the purpose of quantum control, quantum error correction, and methodologically related areas in condensed matter physics and beyond. The workshop will also address topics in quantum machine learning, such as quantum embeddings, quantum kernel methods and parametrised quantum circuits. It will bring together experts and pioneers in this rapidly developing field to further collaborations and raise the bar on machine learning applications for quantum physics. A special focus will be to engage researchers working on non-ML-related aspects of quantum computing and quantum technology in the Nordic region and beyond, to encourage cross-fertilization between the communities.

 

---

Program

See Timetable

---

Invited speakers

Confirmed speakers:

Isaac Tamblyn	University of Ottawa and University of Waterloo
Zohar Ringel	Hebrew University of Jerusalem
Jacob Biamonte	Skolkovo Institute of Science and Technology
Anasua Chatterjee	Center for Quantum Devices at Copenhagen University
Juan Carrasquilla	Vector Institute for Artificial Intelligence
Bernd Rosenow	University of Leipzig
Shai Machnes	Forschungscentrum Jülich
Devdatt Dubhashi	Chalmers University of Technology
Marin Bukov	Sofia University
Yi-Zhuang You	UC San Diego
Giulia Ferrini	Chalmers University of Technology
Göran Johansson	Chalmers University of Technology
Mikko Möttönen	Aalto University
André Carvalho	Q-CTRL
Yihui Quek	Freie Universität Berlin

 

---

Abstract submission

The deadline for abstract submission is July 1st. Submissions received after this deadline will only be considered if space allows. Acceptance or rejection messages will follow the week after. We have a maximum of 60 participants in total. 

 

---

Registration

Registration has now opened and the deadline is July 1st, 2022. Due to space limitations, there will be a selection process and all accepted participants will receive a letter of acceptance by the organizing committee by July 8th, at the latest.  

You are welcome to register without submitting an abstract.  

The workshop will be on-site, no remote attendance. 

 

---

Accommodation

The location of the workshop is the "Albano Building 3" on Albanovägen 29. Three nearby hotel suggestions are Biz Apartment Gärdet, the Elite Hotel Arcadia, and Best Western Plus Time Hotel, though you are of course entirely free to choose others. If you choose Biz Apartment Gärdet, please reference Nordita and the title of this event in your request to the hotel in order to receive a reduced rate. 

There is a limited number of rooms reserved at Biz Apartment Gärdet, primarily for invited speakers. Except for invited speakers you are responsible for booking your own accommodation.

---

Travel support

Travel support is unfortunately not available.

---

Sponsored by:

Monday 29 August

1 Registration

2 Welcome

3 New superconducting qubit and millikelvin electronics for it

We recently discovered a new kind of a superconducting qubit, the unimon, that can be fabricated using standard materials and techniques out of a single Josephson junction and a superconducting resonator, yet having higher anharmonicity than the transmon and resilience against charge and flux noise. Our first experiments on the unimon demonstrate single-qubit-gate fidelity of 99.9% stable for several hours without recalibration. In addition, we have developed qubit readout, reset, and control electronics that operates at millikelvin temperatures and can be integrated with the unimon in the future. These results have been obtained by the Quantum Computing and Devices (QCD) group in collaboration with several other groups. See https://www.aalto.fi/en/department-of-applied-physics/qcd-media for highlighted results.

Speaker: Mikko Möttönen (Aalto)

4 Improving quantum computer performance with machine learning

Quantum control provides a powerful framework for augmenting the performance of quantum devices, improving processes from gate design and calibration through to readout. In particular, measurement-based feedback control using machine learning techniques provides a path to control large systems, by allowing autonomous agents to determine the best control solutions even in the absence of a detailed model of the system.

In this talk, we will show how this fully automated closed-loop approach can be used to design logical gates that are resilient to hardware noise and more stable over time, showing robustness to system drift up to a month. We will also show a protocol that allows for autonomous and parallel gate tune up across entire devices. Next, we present an automated hardware-informed compilation, crosstalk mitigation, and optimized gate replacement routine, as well as a highly scalable neural-network-based measurement-error-mitigation protocol in post-processing after the execution on the quantum device.

We conclude by presenting the results of the implementation of these strategies in a 16-qubit superconducting quantum computer. We demonstrate benefits that increase with system size, reaching a 9000 times enhancement in quantum algorithm performance.

Speaker: Andre Carvalho (Q-CTRL)

5 Lunch

6 Machine-learning tools for rapid development of quantum technology

Development of quantum technology devices, and quantum technology in particular, is an arduous process, as it requires both scale and extreme accuracy.

Detailed and precise models of the devices are extremely rare, as characterization procedures of the numerous parameters comprising such models are often ad-hoc - requiring design of parameter-specific experiments and hand-coded scripts to execute and analyze the data.

The most-often used alternative are much simplified models which fail to predict gate fidelities to high accuracy and are therefore of limited utility for optimal control.

Subsequent closed-loop calibration of control pulses leave us with an equally unsatisfying situation of pulses whose precise operation we do not understand. Worst - they don't provide insight as to the causes of remaining infidelities.

Novel algorithmic and machine-learning techniques can go a long to rectify the situation. In this talk I will describe the ongoing work to develop such tools, including highly detailed TensorFlow digital twins of quantum devices, generalized model learning, optimizations based on reinforcement-learning, and automated Bayesian experiment design.

Together with more mundane quantum optimal control tools, they form a toolset which can help gain insight into the behavior of our systems, and significantly accelerate their development.

Speaker: Shai Machnes (Qruise)

7 Break

8 Machine learning optimization of Majorana hybrid nanowires

As the complexity of quantum systems such as quantum bit arrays increases, efforts to automate expensive tuning become increasingly worthwhile. We study machine learning-based tuning of gate arrays using the CMA-ES algorithm for the case study of Majorana wires with strong disorder. We find that the algorithm is able to efficiently improve the topological signatures, learn intrinsic disorder profiles, and completely eliminate disorder effects. For example, with only 20 gates, it is possible to fully recover Majorana zero modes destroyed by disorder with optimized gate voltages.

Speaker: Bernd Rosenow (Leipzig University)

17:30 Reception

Tuesday 30 August

9 The vacuum provides quantum advantage to otherwise simulatable architectures

Identifying the physical resources underlying quantum advantage — i.e., yielding the ability of quantum computers to solve computational problems faster than classical computers — is of crucial importance for the design of meaningful architectures for quantum computation. Often, the resource depends on the model. In the framework of infinite-dimensional continuous-variable (CV) systems, associated to bosonic fields, Gaussian circuits (where the input state, operations and measurements are all described by Gaussian functions) are classically efficiently simulatable. In other words, for these circuits a classical algorithm exists that can reproduce the outcome of the computation. To promote them to universal quantum computation specific non-Gaussian resources have to be provided, such as the so-called Gottesman-Kitaev-Preskill (GKP) state. The cost of producing these enabling resources with sufficient quality generally requires a significant overhead and their distinct features are typically complex and in stark contrast with respect to the elements of the corresponding simulatable architectures. It is a natural question to ask: are resources always complex and costly to produce?
In this work we provide a specific example of a CV quantum computing architecture that is classically efficiently simulatable, and that becomes universal by adding the vacuum state. The latter state is widely regarded as the simplest quantum state of a bosonic field, and in particular it is a Gaussian state. The architecture considered is based on GKP states, Gaussian operations and measurement of the quadratures of the bosonic field. First we prove that this class of circuits is classically efficiently simulatable for most Gaussian operations. Then, we leverage on recent results where the same architecture combined with the vacuum (or a thermal) state was shown to be universal for quantum computation, to conclude that the vacuum provides quantum advantage.

Speaker: Giulia Ferrini

10 Coffee

11 Quantum Approximate Optimization Algorithm applied to the binary perceptron

We apply digitized Quantum Annealing (QA) and Quantum Approximate Optimization Algorithm (QAOA) to a paradigmatic task of supervised learning in artificial neural networks: the optimization of synaptic weights for the binary perceptron.
At variance with the usual QAOA applications to MaxCut, or to quantum spin-chains ground state preparation, the classical Hamiltonian is characterized by highly non-local multi-spin interactions.

Yet, we provide evidence for the existence of optimal smooth solutions for the QAOA parameters, which are transferable among typical instances of the same problem, and we prove numerically an enhanced performance of QAOA over traditional QA.
We also investigate on the role of the QAOA optimization landscape geometry in this problem, showing that the detrimental effect of a gap-closing transition encountered in QA is also negatively affecting the performance of our implementation of QAOA.

Speaker: Mr Pietro Torta (SISSA)

12 Optimising quantum circuits with machine learning

tbd

Speaker: David Fitzek (Chalmers University of Technology)

13 Lunch

14 Word from Nordita Director

15 RG using information bottlenecks - venturing into unsolved models.

The information bottleneck (IB), an abstract mathematical framework for compressing relevant information, has attracted some attention recently due to newly discovered relations with deep learning. Specifically, state-of-the-art deep learning approaches now enable us to access IB quantities numerically. Having this new tool at our disposal, it is interesting to explore its relationship with the Renormalization Group (RG) where an apriori different notion of relevant information exists--- that of relevant operators. In a related manner, IB shows promise as an automated method of identifying relevant/slow degrees of freedom in complex interacting models. In this talk, I'll introduce the concept of IB and then report some of the progress we made on these theoretical and applicative fronts. I'll describe a concrete dictionary between relevant operators and bifurcation points in IB compression. In addition, I'll report some recent applications of this approach to self-dual criticality in 3 dimensions.

Speaker: Zohar Ringel (Assistant Professor)

16 Break

17 WACQT overview and airline logistics

Speaker: Göran Johansson

18 Headed beyond Kohn-Sham DFT with deep learning

Streamed talk

Speaker: Isaac Tamblyn

Wednesday 31 August

19 Adaptive Bayesian learning for quantum sensing and characterisation

Sophisticated learning techniques, such as Bayesian inference, neural networks or reinforcement learning are becoming increasingly popular as powerful tools to characterise and optimise quantum systems.
Here, I will report on our theoretical and experimental work to speed up the characterisation of quantum systems by exploiting online adaptive measurements and Bayesian inference. We have developed a system, comprising a hard-realtime microcontroller, that implements real-time adaptation of measurement settings, based on previous measurement outcomes (through Sequential MonteCarlo, with <100 microseconds update time). Such a system always operate near maximum sensitivity, even in cases where the optimal settings depend on the unknown parameter to be estimated. We have used this system to characterise decoherence timescales for a single spin qubit associated to a nitrogen-vacancy centre in diamond, showing an improvement in speed of one order of magnitude. We have also benchmarked different adaptive heuristics based on different Bayesian and frequentist statistical estimation bounds. These results offer opportunities in the characterisation of large-scale quantum systems, and in quantum sensing, by considerably increasing the measurement bandwidth.
Time permitting, I will also discuss our on-going work on learning models for quantum emitters from photon arrival times, utilising a Markov-chain MonteCarlo approach.

Speaker: Cristian Bonato (Heriot-Watt University)

20 Coffee

21 Self-Correcting Quantum Many-Body Control using Reinforcement Learning with Tensor Networks

Quantum many-body control is a central milestone en route to harnessing quantum technologies. However, the exponential growth of the Hilbert space dimension with the number of qubits makes it challenging to classically simulate quantum many-body systems and consequently, to devise reliable and robust optimal control protocols. I will present a novel framework for efficiently controlling quantum many-body systems based on reinforcement learning (RL). We tackle the quantum control problem by leveraging matrix product states (i) for representing the many-body state and, (ii) as part of the trainable machine learning architecture for our RL agent. The framework is applied to prepare ground states of the quantum Ising chain, including critical states. It allows us to control systems far larger than neural-network-only architectures permit, while retaining the advantages of deep learning algorithms, such as generalizability and trainable robustness to noise. In particular, I will demonstrate that RL agents are capable of finding universal controls, of learning how to optimally steer previously unseen many-body states, and of adapting control protocols on the fly when the quantum dynamics are subject to stochastic perturbations.

Speaker: Marin Bukov (MPI-PKS)

22 Quantum Annealing for Neural Network optimization problems: a new approach via Tensor Network simulations

Quantum Annealing (QA) is one of the most promising frameworks for quantum opti-mization. In this work, we focus on the problem of minimizing complex classical cost functions associated with prototypical discrete neural networks, specifically the paradigmatic binary perceptron and the Hopfield model. We show that the adiabatic time evolution of QA can be efficiently represented as a suitable Tensor Network. This representation allows for simple classical simulations, well-beyond small sizes amenable to exact diagonalization techniques. We show that the optimized state, expressed as a Matrix Product State (MPS), can be recast into a Quantum Circuit, whose depth scales only linearly with the system size and quadratically with the MPS bond dimension. This may represent a valuable starting point allowing for further circuit optimization on near-term quantum devices.

Speaker: Guglielmo Lami (SISSA)

23 Lunch

24 Increasing the complexity of quantum devices in an algorithmic age

Advances in fabrication and quantum control methods are leading to gate-controlled nanoscale devices becoming progressively more complex, yet affording more and more exquisite local control over gate potentials. Automated solutions for tuning and characterisation are also advancing, as the parameter space of these complex devices are quickly becoming too large to manually investigate. In this talk, I will present our efforts to harness these two complementary advances, to impact the fields of quantum information as well as fundamental condensed matter physics. Hardware and software techniques working in concert assist us to find operation points in spin qubit arrays, as well as to read them out; subsequently, these techniques are also applied to mesoscopic regimes such as quantum Hall systems. Our results propose a path towards the application of algorithmic techniques to fundamental problems in the solid state.

Speaker: Anasua Chatterjee (Niels Bohr Institute)

25 Realizing a deep reinforcement learning agent for the discovery of real-time feedback control strategies for a quantum system

Finding strategies to control quantum information processing devices in real-time becomes increasingly demanding as they grow in size and complexity. Reinforcement learning promises to overcome this challenge by uncovering the underlying system dynamics without relying on a specific model. Here, we implement a deep neural network on a field-programmable gate array (FPGA) and demonstrate its use as a real-time reinforcement learning agent to efficiently initialize a superconducting qubit into its ground state. The agent repeatedly measures the state of the qubit and chooses on a sub-microsecond time scale whether to idle, to apply a bit-flip gate, or to terminate. After the agent chooses to terminate the initialization process, we perform a validation measurement to infer the probability of having successfully initialized the ground state. To train the agent, we use model-free reinforcement learning that is based solely on measurement data.

Speaker: Kevin Reuer (Deparment of Physics, ETH Zürich)

26 Break

27 A tutorial on designing quantum gates with optimal control theory and automatic differentiation

I will present a hands-on tutorial (with code) on optimal coherent control for quantum gate design, leveraging automatic differentiation and tensor networks.

Speaker: Giacomo Torlai

18:00 Dinner

Thursday 1 September

28 Coffee

29 Quantum variational learning for quantum error-correcting codes

Quantum error-correcting codes (QECCs) are believed to be a necessity for large-scale fault-tolerant quantum computation. In the past two decades, various methods of QECC constructions have been developed, leading to many good families of codes. However, the majority of these codes are not suitable for near-term quantum devices. Here we present VarQEC, a noise-resilient variational quantum algorithm to search for quantum codes with a hardware-efficient encoding circuit. The cost functions are inspired by the most general and fundamental requirements of a QECC, the Knill-Laflamme conditions. Given the target noise channel (or the target code parameters) and the hardware connectivity graph, we optimize a shallow variational quantum circuit to prepare the basis states of an eligible code. In principle, VarQEC can find quantum codes for any error model, whether additive or non-additive, degenerate or non-degenerate, pure or impure. We have verified its effectiveness by (re)discovering some symmetric and asymmetric codes, e.g., $((n,2^{n-6},3))_2$ for $n$ from 7 to 14. We also found new $((6,2,3))_2$ and $((7,2,3))_2$ codes that are not equivalent to any stabilizer code, and extensive numerical evidence with VarQEC suggests that a $((7,3,3))_2$ code does not exist. Furthermore, we found many new channel-adaptive codes for error models involving nearest-neighbor correlated errors. Our work sheds new light on the understanding of QECC in general, which may also help to enhance near-term device performance with channel-adaptive error-correcting codes.

Speaker: Mr Chenfeng Cao (The Hong Kong university of Science and Technology)

30 Quantum simulation and Rydberg atom arrays

Speaker: Roger Melko

31 Lunch

32 Variational Neural Annealing

(Streamed) Many important challenges in science and technology can be cast as optimization problems. When viewed in a statistical physics framework, these can be tackled by simulated annealing, where a gradual cooling procedure helps search for ground state solutions of a target Hamiltonian. While powerful, simulated annealing is known to have prohibitively slow sampling dynamics when the optimization landscape is rough or glassy. In this talk I will show that by generalizing the target distribution with a parameterized model, an analogous annealing framework based on the variational principle can be used to search for ground state solutions. Autoregressive models such as recurrent neural networks provide ideal parameterizations since they can be exactly sampled without slow dynamics even when the model encodes a rough landscape. We implement this procedure in the classical and quantum settings on several prototypical spin glass Hamiltonians, and find that it significantly outperforms traditional simulated annealing in the asymptotic limit, illustrating the potential power of this yet unexplored route to optimization.

Speaker: Juan Carrasquilla (Vector Institute)

33 Quantum process tomography with gradient descent

Quantum process tomography (QPT) obtains the representation of a quantum process using experimentally obtained measurement data. We can cast the QPT problem into a learning task where machine learning methods have been recently successful in using generative models for QPT. In this talk, we show how simple gradient-based learning with appropriate constraints on the representation of process, along with restrictions on the gradients can solve QPT. We will demonstrate gradient-based learning of processes for 2- 5 qubits as well as single-mode bosonic systems. We compare our simple approach to existing techniques such as compressed sensing and projected least squares QPT. We also show that using neural networks rather than standard process representations provides no significant advantage which may indicate that good representations of process combined with gradient-based learning might be sufficient for QPT tasks.

Speaker: Shahnawaz Ahmed (Chalmers University of Technology)

34 Break

35 Machine Learning Renormalization Group and Generative Modeling

In this talk, I will introduce the machine learning renormalization group method, a hierarchical flow-based generative model motivated by the idea of the renormalization group in physics. Given the action of a field theory, the algorithm learns the optimal renormalization group transformation and maps the field configuration from the holographic boundary to the bulk, which enables efficient sampling and error correction. Beyond physics applications, I will also demonstrate the application of this method in the image and language processing domain.

Speaker: Yizhuang You

36 How to learn a quantum state (and how not to)

streamed
Learning an unknown n-qubit quantum state is a fundamental challenge in quantum computing. Full tomography, however, requires exponential-in-n many copies of \rho for a good estimate. Is it possible to circumvent this exponential tax on resources? We consider two variants of this question:
1. “Pretty-good tomography” (based on https://arxiv.org/abs/2102.07171, NeurIPS 2021 (Spotlight)): Aaronson and others introduced several “reduced” models of learning quantum states which impose weaker requirements on the learner: PAC-learning, shadow tomography for learning ``shadows” of a quantum state, online learning, whose complexities scale only linearly in n. We show implications and reductions between the many models in this menagerie, and further introduce a combinatorial parameter that characterizes the complexity of learning. As an application, we improve shadow tomography (for classes of quantum states).
2. Probabilistic modelling (based on https://arxiv.org/abs/2110.05517 and https://arxiv.org/pdf/2207.03140.pdf): Deep generative models have recently empowered many impressive scientific feats, ranging from predicting protein structure to atomic accuracy (Alpha-Fold) to achieving human-level language comprehension (GPT-3). At the heart of these models is the question: by drawing very few samples from a probability distribution, can we learn an algorithm that generates more samples from the same distribution? Even more intriguingly: could there be a quantum advantage for such a task? We present both go and no-go results for this setting.

Speaker: Yihui Quek

Friday 2 September

37 Bandits and Bayesian Optimization for Quantum Circuits

We survey bandit and Bayesian optimization approaches for optimizing parameters for Quantum circuits.

Speaker: Devdatt Dubhashi

38 Closing remarks