Autoregressive neural-network wavefunctions for ab initio quantum chemistry

NATURE MACHINE INTELLIGENCE (2022)

Authors:

Thomas D Barrett, Aleksei Malyshev, AI Lvovsky

Super-resolution linear optical imaging in the far field

Physical Review Letters American Physical Society 127 (2021) 253602

Abstract:

The resolution of optical imaging devices is ultimately limited by the diffraction of light. To circumvent this limit, modern superresolution microscopy techniques employ active interaction with the object by exploiting its optical nonlinearities, nonclassical properties of the illumination beam, or near field probing. Thus, they are not applicable whenever such interaction is not possible, for example, in astronomy or noninvasive biological imaging. Far field, linear optical superresolution techniques based on passive analysis of light coming from the object would cover these gaps. In this Letter, we present the first proof-of-principle demonstration of such a technique for 2D imaging. It works by accessing information about spatial correlations of the image optical field and, hence, about the object itself via measuring projections onto Hermite-Gaussian transverse spatial modes. With a basis of 21 spatial modes in both transverse dimensions, we perform two-dimensional imaging with twofold resolution enhancement beyond the diffraction limit.

Entangled resource for interfacing single- and dual-rail optical qubits

Quantum Verein zur Forderung des Open Access Publizierens in den Quantenwissenschaften 5 (2021) 416

Authors:

David Drahi, Demid V Sychev, Khurram K Pirov, Ekaterina A Sazhina, Valeriy A Novikov, Ian A Walmsley, Alexander Lvovsky

Abstract:

Today's most widely used method of encoding quantum information in optical qubits is the dual-rail basis, often carried out through the polarisation of a single photon. On the other hand, many stationary carriers of quantum information – such as atoms – couple to light via the single-rail encoding in which the qubit is encoded in the number of photons. As such, interconversion between the two encodings is paramount in order to achieve cohesive quantum networks. In this paper, we demonstrate this by generating an entangled resource between the two encodings and using it to teleport a dual-rail qubit onto its single-rail counterpart. This work completes the set of tools necessary for the interconversion between the three primary encodings of the qubit in the optical field: single-rail, dual-rail and continuous-variable.

Backpropagation through nonlinear units for the all-optical training of neural networks

Photonics Research Optical Society of America 9:3 (2021) B71-B80

Authors:

Xianxin Guo, Thomas D Barrett, Zhiming M Wang, Ai Lvovsky

Abstract:

We propose a practical scheme for end-to-end optical backpropagation in neural networks. Using saturable absorption for the nonlinear units, we find that the backward-propagating gradients required to train the network can be approximated in a surprisingly simple pump-probe scheme that requires only simple passive optical elements. Simulations show that, with readily obtainable optical depths, our approach can achieve equivalent performance to state-of-the-art computational networks on image classification benchmarks, even in deep networks with multiple sequential gradient approximation. With backpropagation through nonlinear units being an outstanding challenge to the field, this work provides a feasible path toward truly all-optical neural networks.

Reinforcement learning enhanced quantum-inspired algorithm for combinatorial optimization

Machine Learning: Science and Technology IOP Publishing 2:2 (2020) 025009

Authors:

Dmitrii Beloborodov, Ae Ulanov, Jakob N Foerster, Shimon Whiteson, Ai Lvovsky

Abstract:

Quantum hardware and quantum-inspired algorithms are becoming increasingly popular for combinatorial optimization. However, these algorithms may require careful hyperparameter tuning for each problem instance. We use a reinforcement learning agent in conjunction with a quantum-inspired algorithm to solve the Ising energy minimization problem, which is equivalent to the Maximum Cut problem. The agent controls the algorithm by tuning one of its parameters with the goal of improving recently seen solutions. We propose a new Rescaled Ranked Reward (R3) method that enables a stable single-player version of self-play training and helps the agent escape local optima. The training on any problem instance can be accelerated by applying transfer learning from an agent trained on randomly generated problems. Our approach allows sampling high quality solutions to the Ising problem with high probability and outperforms both baseline heuristics and a black-box hyperparameter optimization approach.