Question-3

Question-3: How can machine learning improve the performance of quantum algorithms for quantum chemistry?


Tutorials

Alan Aspuru-Guzik (University of Toronto)

There is no time for science as usual: Materials Acceleration Platforms

Wednesday 3rd March 2021, 15:00-16.30 GMT

The world is facing several time-sensitive issues ranging from climate change to the rapid degradation of our climate, as well as the emergence of new diseases like COVID-19. We need to rethink the way we do science and think of it as a workflow that could be optimized. Where are the pain points that can be solved with automation, artificial intelligence, or better human practices? My group has been thinking about this question with an application to the design of organic optoelectronic materials. In this talk, I will discuss the progress in developing materials acceleration platforms, or self-driving labs for this purpose.



Sophia Economou (VirginiaTech – USA)

Variational Quantum Eigensolvers
Wednesday 3rd March 2021, 17:00-18.30 GMT

Variational quantum eigensolvers (VQEs) constitute a class of hybrid quantum-classical algorithms that are envisioned to be appropriate for noisy intermediate scale quantum processors. The majority of VQEs focus on quantum simulation, and particularly finding properties of many-body quantum systems, such as the ground state energies of complicated molecules. In VQEs, the quantum processor is where the quantum state is variationally prepared and measurements are made, while the classical computer performs optimization. As such, VQEs can be thought of as training of quantum circuits. In this tutorial, I will present the concept of VQEs and focus on challenges in the state preparation, measurement, and optimization aspects, and discuss connections to machine learning.


Invited Speakers

Jarrod McClean (Google Quantum Artificial Intelligence Lab)

Probing the role of learning in computational chemistry with a quantum computation perspective
Thursday 4th March 2021, 17:00-17:45 GMT

Quantum algorithms for the simulation of chemical systems promise to open up the ability to simulate reactions and novel materials to previously inaccessible accuracy and size.  However, the perspective of quantum computation also teaches us about the limits of computation and its role in design or formulation of physical theories.  Here we address some of these limitations by taking a learning perspective and explore the role of a quantum computer in such a setting. We highlight this with recent results on how the availability of data can cause some quantum problems thought to be challenging to simulate classically to become easy in a classical learning setting, but use this to hone our perspective on the ideal role for a fault tolerant quantum computer in chemistry.



Min-Hsiu Hsieh (University of Technology Sydney – Australia)

The power of quantum neural networks
Friday 5th March 2021, 12:00-12:45 GMT

Quantum neural networks (QNNs) have been broadly used in various works with different levels of claimed benefits. One of my research interests in quantum machine learning is to understand the power of QNNs. In this talk, I will first compare the expressive power of QNNs with Boltzmann machines. Next, I will provide our results on the learnability of QNNs in terms of its trainability and generalization. Finally, I will provide a few applications of QNNs on machine learning tasks and ground state approximations. 



Roger Melko (Department of Physics & Astronomy, University of Waterloo – Canada)

Reconstructing quantum states with generative models
Friday 5th March 2021, 14:00-14:45 GMT

Generative models are a powerful tool in unsupervised learning, where the goal is to learn the unknown probability distribution that underlies a data set.  Recently, it has been demonstrated that modern generative models adopted from industry are capable of reconstructing quantum states, given projective measurement data on individual qubits. These virtual reconstructions can then be studied with probes unavailable to the original experiment. In this talk I will outline the strategy for quantum state reconstruction using generative models, and show examples on experimental data from a Rydberg atom quantum simulator.  I will discuss the continuing theoretical development of the field, including the exploration of powerful autoregressive models for the reconstruction of sign-problematic and mixed quantum states.


Contributed Talks

Abhinav Anand (University of Toronto – Canada)
Natural Evolutionary Strategies for Variational Quantum Computation
Thursday 4th March 2021, 17:45-18:10 GMT
We use natural evolutionary strategies (NES) for the optimization of randomly-initialized parameterized quantum circuits (PQCs) in the region of vanishing gradients. We show how using the NES gradient estimator the exponential decrease in variance can be alleviated. We also extend NES to deep circuits, by introducing batch optimization and hybrid approaches. Our empirical results indicate that one can use NES as a hybrid tool in tandem with other gradient-based methods for optimization of deep quantum circuits in regions with vanishing gradients. Our study contributes to the efforts towards developing better quantum algorithms for optimization and overcoming the associated pitfalls.
Tools: Natural Evolutionary Strategies

Matias Bilkis (Autonomous University of Barcelona – Spain)
Variable Ansatz VQE: a semi-agnostic approach

Friday 5th March 2021, 12:45-13:10 GMT
Ansatz design currently appears as a cornerstone of VQE, which is one of the most prominent algorithms for quantum chemistry. While trainability of large circuits becomes infeasible due to the presence of either Barren Plateaus or hardware noise, shallow circuits are simply insufficient to prepare meaningful quantum states. To overcome this issue, we propose a learning scheme to iteratively modify the ansatz structure, while keeping its depth as shallow as possible according to specific circuit-compression rules.
Tools: noise-aware circuit learning (NACL), TensorFlow-Quantum, evolutionary algorithms, hybrid-classical-quantum-variational-algorithm


Bryce Fuller (IBM Quantum – USA)
Accurate Simulation of Quantum Chemistry on Quantum Computers Integrated with Neural Networks
Friday 5th March 2021, 13:10-13:35 GMT
We show how integrating neural networks with near-term quantum chemistry algorithms can reduce the variance of chemical observable estimates and combat the impact of decoherence on the underlying state being measured. This requires fewer queries to be made on quantum hardware and could improve the scalability of techniques such as the variational quantum eigensolver. 
Tools: Neural Network Quantum States, Restricted Boltzmann Machines, Low-Rank Matrix Approximation, Monte Carlo Sampling, FKV Singular Value Decomposition, Quantum Neural Networks / Variational Quantum Eigensolver