Barren plateaus in quantum neural network training landscapes

Many experimental proposals for noisy intermediate scale quantum devices involve training a parameterized quantum circuit with a classical optimization loop. Such hybrid quantum-classical algorithms are popular for applications in quantum simulation, optimization, and machine learning. Due to its simplicity and hardware efficiency, random circuits are often proposed as initial guesses for exploring the space of quantum states. We show that the exponential dimension of Hilbert space and the gradient estimation complexity make this choice unsuitable for hybrid quantum-classical algorithms run on more than a few qubits. Specifically, we show that for a wide class of reasonable parameterized quantum circuits, the probability that the gradient along any reasonable direction is non-zero to some fixed precision is exponentially small as a function of the number of qubits. We argue that this is related to the 2-design characteristic of random circuits, and that solutions to this problem must be studied.Gradient-based hybrid quantum-classical algorithms are often initialised with random, unstructured guesses. Here, the authors show that this approach will fail in the long run, due to the exponentially-small probability of finding a large enough gradient along any direction.

[1]  Ryan Babbush,et al.  The theory of variational hybrid quantum-classical algorithms , 2015, 1509.04279.

[2]  J. O'Brien,et al.  Witnessing eigenstates for quantum simulation of Hamiltonian spectra , 2016, Science Advances.

[3]  J. R. Ipsen.  Products of independent Gaussian random matrices , 2015, 1510.06128.

[4]  Jonathan Carter,et al.  Computation of Molecular Spectra on a Quantum Processor with an Error-Resilient Algorithm , 2018 .

[5]  Alán Aspuru-Guzik,et al.  Quantum Simulation of Electronic Structure with Linear Depth and Connectivity. , 2017, Physical review letters.

[6]  J. Gambetta,et al.  Hardware-efficient Quantum Optimizer for Small Molecules and Quantum Magnets , 2017 .

[7]  J. McClean,et al.  Application of fermionic marginal constraints to hybrid quantum algorithms , 2018, 1801.03524.

[8]  A. Harrow,et al.  Random Quantum Circuits are Approximate 2-designs , 2008, 0802.1919.

[9]  John Preskill,et al.  Quantum Computing in the NISQ era and beyond , 2018, Quantum.

[10]  H. Neven,et al.  Low-Depth Quantum Simulation of Materials , 2018 .

[11]  Alán Aspuru-Guzik,et al.  The theory of variational hybrid quantum-classical algorithms , 2015, 1509.04279.

[12]  Guigang Zhang,et al.  Deep Learning , 2016, Int. J. Semantic Comput..

[13]  M. Ledoux.  The concentration of measure phenomenon , 2001 .

[14]  Andreas Winter,et al.  Are random pure States useful for quantum computation? , 2008, Physical review letters.

[15]  P. Coveney,et al.  Scalable Quantum Simulation of Molecular Energies , 2015, 1512.06860.

[16]  Yoshua Bengio,et al.  Understanding the difficulty of training deep feedforward neural networks , 2010, AISTATS.

[17]  J. McClean,et al.  Strategies for quantum computing molecular energies using the unitary coupled cluster ansatz , 2017, Quantum Science and Technology.

[18]  J. Carter,et al.  Hybrid Quantum-Classical Hierarchy for Mitigation of Decoherence and Determination of Excited States , 2016, 1603.05681.

[19]  T. Monz,et al.  Quantum Chemistry Calculations on a Trapped-Ion Quantum Simulator , 2018, Physical Review X.

[20]  J. Gambetta,et al.  Hardware-efficient variational quantum eigensolver for small molecules and quantum magnets , 2017, Nature.

[21]  Ohad Shamir,et al.  Failures of Gradient-Based Deep Learning , 2017, ICML.

[22]  Alán Aspuru-Guzik,et al.  Exploiting Locality in Quantum Computation for Quantum Chemistry. , 2014, The journal of physical chemistry letters.

[23]  Ievgeniia Oshurko.  Quantum Machine Learning , 2020, Quantum Computing.

[24]  James Andrew Bagnell,et al.  Learning in modular systems , 2010 .

[25]  E. Farhi,et al.  A Quantum Approximate Optimization Algorithm , 2014, 1411.4028.

[26]  D. Gross,et al.  Most quantum States are too entangled to be useful as computational resources. , 2008, Physical review letters.

[27]  Joseph M. Renes,et al.  Symmetric informationally complete quantum measurements , 2003, quant-ph/0310075.

[28]  Christoph Dankert,et al.  Exact and approximate unitary 2-designs and their application to fidelity estimation , 2009 .

[29]  Peter D. Johnson,et al.  QVECTOR: an algorithm for device-tailored quantum error correction , 2017, 1711.02249.

[30]  Daniel A. Roberts,et al.  Chaos and complexity by design , 2016, 1610.04903.

[31]  H. Neven,et al.  Quantum Algorithms for Fixed Qubit Architectures , 2017, 1703.06199.

[32]  Alán Aspuru-Guzik,et al.  A variational eigenvalue solver on a photonic quantum processor , 2013, Nature Communications.

[33]  Jian Sun,et al.  Deep Residual Learning for Image Recognition , 2015, 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

[34]  Yoshua Bengio,et al.  Greedy Layer-Wise Training of Deep Networks , 2006, NIPS.

[35]  Alán Aspuru-Guzik,et al.  Quantum autoencoders for efficient compression of quantum data , 2016, 1612.02806.

[36]  Sergey Ioffe,et al.  Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift , 2015, ICML.

[37]  Alán Aspuru-Guzik,et al.  Quantum Neuron: an elementary building block for machine learning on quantum computers , 2017, ArXiv.

[38]  H. Neven,et al.  Characterizing quantum supremacy in near-term devices , 2016, Nature Physics.

[39]  Blake R. Johnson,et al.  Unsupervised Machine Learning on a Hybrid Quantum Computer , 2017, 1712.05771.

[40]  Yee Whye Teh,et al.  A Fast Learning Algorithm for Deep Belief Nets , 2006, Neural Computation.

[41]  A. J. Short,et al.  Entanglement and the foundations of statistical mechanics , 2005 .

[42]  R. Pooser,et al.  Cloud Quantum Computing of an Atomic Nucleus. , 2018, Physical Review Letters.

[43]  Yoshua Bengio,et al.  Gradient Flow in Recurrent Nets: the Difficulty of Learning Long-Term Dependencies , 2001 .

[44]  M. Hastings,et al.  Training A Quantum Optimizer , 2016, 1605.05370.

[45]  Stephen J. Benkovic,et al.  Corrigendum: RecG and UvsW catalyse robust DNA rewinding critical for stalled DNA replication fork rescue , 2014, Nature Communications.

[46]  Hartmut Neven,et al.  Classification with Quantum Neural Networks on Near Term Processors , 2018, 1802.06002.

[47]  E. Knill,et al.  Optimal quantum measurements of expectation values of observables , 2006, quant-ph/0607019.

[48]  Mikhail Smelyanskiy,et al.  Practical optimization for hybrid quantum-classical algorithms , 2017, 1701.01450.

[49]  M. Yung,et al.  Quantum implementation of the unitary coupled cluster for simulating molecular electronic structure , 2015, 1506.00443.

[50]  M. Hastings,et al.  Progress towards practical quantum variational algorithms , 2015, 1507.08969.

[51]  Andrew W. Cross,et al.  Quantum optimization using variational algorithms on near-term quantum devices , 2017, Quantum Science and Technology.

[52]  L. Lamata,et al.  From transistor to trapped-ion computers for quantum chemistry , 2013, Scientific Reports.