Quantum Algorithms for Escaping from Saddle Points

We initiate the study of quantum algorithms for escaping from saddle points with provable guarantee. Given a function f:Rn→R, our quantum algorithm outputs an ϵ-approximate second-order stationary point using O~(log2⁡(n)/ϵ1.75) queries to the quantum evaluation oracle (i.e., the zeroth-order oracle). Compared to the classical state-of-the-art algorithm by Jin et al. with O~(log6⁡(n)/ϵ1.75) queries to the gradient oracle (i.e., the first-order oracle), our quantum algorithm is polynomially better in terms of log⁡n and matches its complexity in terms of 1/ϵ. Technically, our main contribution is the idea of replacing the classical perturbations in gradient descent methods by simulating quantum wave equations, which constitutes the improvement in the quantum query complexity with log⁡n factors for escaping from saddle points. We also show how to use a quantum gradient computation algorithm due to Jordan to replace the classical gradient queries by quantum evaluation queries with the same complexity. Finally, we also perform numerical experiments that support our theoretical findings.

[1]  Y. Nesterov.  A method for solving the convex programming problem with convergence rate O(1/k^2) , 1983 .

[2]  Ivo Babuška,et al.  The h-p version of the finite element method , 1986 .

[3]  Christof Zalka,et al.  Efficient Simulation of Quantum Systems by Quantum Computers , 1998 .

[4]  D. Manolopoulos,et al.  Symplectic integrators tailored to the time‐dependent Schrödinger equation , 1996 .

[5]  Lov K. Grover.  A fast quantum mechanical algorithm for database search , 1996, STOC '96.

[6]  S. Wiesner.  Simulations of Many-Body Quantum Systems by a Quantum Computer , 1996, quant-ph/9603028.

[7]  Seth Lloyd,et al.  Universal Quantum Simulators , 1996, Science.

[8]  Christof Zalka.  Simulating quantum systems on a quantum computer , 1996, Proceedings of the Royal Society of London. Series A: Mathematical, Physical and Engineering Sciences.

[9]  Christof Zalka.  Efficient Simulation of Quantum Systems by Quantum Computers , 1996, quant-ph/9603026.

[10]  J. Bourgain.  Growth of Sobolev Norms in Linear Schrödinger Equations with Quasi-Periodic Potential , 1999 .

[11]  J. Bourgain.  On growth of sobolev norms in linear schrödinger equations with smooth time dependent potential , 1999 .

[12]  S. Jordan.  Fast quantum algorithm for numerical gradient estimation. , 2004, Physical review letters.

[13]  David W. Bulger.  Quantum basin hopping with gradient-based local optimisation , 2005, quant-ph/0507193.

[14]  Jianping Li.  General explicit difference formulas for numerical differentiation , 2005 .

[15]  Yurii Nesterov,et al.  Cubic regularization of Newton method and its global performance , 2006, Math. Program..

[16]  Yan V Fyodorov,et al.  Replica Symmetry Breaking Condition Exposed by Random Matrix Calculation of Landscape Complexity , 2007, cond-mat/0702601.

[17]  A. Bray,et al.  Statistics of critical points of Gaussian fields on large-dimensional spaces. , 2006, Physical review letters.

[18]  R. Cleve,et al.  Efficient Quantum Algorithms for Simulating Sparse Hamiltonians , 2005, quant-ph/0508139.

[19]  William A. Webb,et al.  Wavefunction preparation and resampling using a quantum computer , 2008, 0801.0342.

[20]  Stephen P. Jordan,et al.  Quantum computation beyond the circuit model , 2008, 0809.2307.

[21]  Andrew M. Childs,et al.  Limitations on the simulation of non-sparse Hamiltonians , 2009, Quantum Inf. Comput..

[22]  Sham M. Kakade,et al.  A tail inequality for quadratic forms of subgaussian random vectors , 2011, ArXiv.

[23]  Seth Lloyd,et al.  Gaussian quantum information , 2011, 1110.3234.

[24]  Surya Ganguli,et al.  Identifying and attacking the saddle point problem in high-dimensional non-convex optimization , 2014, NIPS.

[25]  Prateek Jain,et al.  Computing Matrix Squareroot via Non Convex Local Search , 2015, ArXiv.

[26]  Furong Huang,et al.  Escaping From Saddle Points - Online Stochastic Gradient for Tensor Decomposition , 2015, COLT.

[27]  Andrew M. Childs,et al.  Hamiltonian Simulation with Nearly Optimal Dependence on all Parameters , 2015, 2015 IEEE 56th Annual Symposium on Foundations of Computer Science.

[28]  John Wright,et al.  A Geometric Analysis of Phase Retrieval , 2016, 2016 IEEE International Symposium on Information Theory (ISIT).

[29]  Nathan Srebro,et al.  Global Optimality of Local Search for Low Rank Matrix Recovery , 2016, NIPS.

[30]  Krysta Marie Svore,et al.  Quantum Speed-ups for Semidefinite Programming , 2016, ArXiv.

[31]  Yair Carmon,et al.  Accelerated Methods for Non-Convex Optimization , 2016, SIAM J. Optim..

[32]  Kfir Y. Levy,et al.  The Power of Normalization: Faster Evasion of Saddle Points , 2016, ArXiv.

[33]  Tengyu Ma,et al.  Matrix Completion has No Spurious Local Minimum , 2016, NIPS.

[34]  Michael I. Jordan.  On Gradient-Based Optimization: Accelerated, Distributed, Asynchronous and Stochastic , 2017, SIGMETRICS.

[35]  Michael I. Jordan,et al.  On Gradient-Based Optimization: Accelerated, Distributed, Asynchronous and Stochastic , 2017, SIGMETRICS.

[36]  Xiaodi Wu,et al.  Exponential Quantum Speed-ups for Semidefinite Programming with Applications to Quantum Learning , 2017, ArXiv.

[37]  Iordanis Kerenidis,et al.  Quantum Recommendation Systems , 2016, ITCS.

[38]  Daniel P. Robinson,et al.  A trust region algorithm with a worst-case iteration complexity of O(ϵ-3/2)\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{docume , 2016, Mathematical Programming.

[39]  Prateek Jain,et al.  Global Convergence of Non-Convex Gradient Descent for Computing Matrix Squareroot , 2015, AISTATS.

[40]  Tianbao Yang,et al.  NEON+: Accelerated Gradient Methods for Extracting Negative Curvature for Non-Convex Optimization , 2017, 1712.01033.

[41]  Yuchen Zhang,et al.  A Hitting Time Analysis of Stochastic Gradient Langevin Dynamics , 2017, COLT.

[42]  I. Chuang,et al.  Optimal Hamiltonian Simulation by Quantum Signal Processing. , 2016, Physical review letters.

[43]  Michael I. Jordan,et al.  How to Escape Saddle Points Efficiently , 2017, ICML.

[44]  Tengyu Ma,et al.  On the optimization landscape of tensor decompositions , 2017, Mathematical Programming.

[45]  Tengyu Ma,et al.  Finding approximate local minima faster than gradient descent , 2016, STOC.

[46]  Michael I. Jordan,et al.  Accelerated Gradient Descent Escapes Saddle Points Faster than Gradient Descent , 2017, COLT.

[47]  Tong Zhang,et al.  SPIDER: Near-Optimal Non-Convex Optimization via Stochastic Path Integrated Differential Estimator , 2018, NeurIPS.

[48]  Chunhao Wang,et al.  Quantum-inspired sublinear classical algorithms for solving low-rank linear systems , 2018, ArXiv.

[49]  Tengyu Ma,et al.  Learning One-hidden-layer Neural Networks with Landscape Design , 2017, ICLR.

[50]  Seth Lloyd,et al.  Quantum-inspired low-rank stochastic regression with logarithmic dependence on the dimension , 2018, ArXiv.

[51]  Ewin Tang,et al.  Quantum-inspired classical algorithms for principal component analysis and supervised clustering , 2018, ArXiv.

[52]  Zeyuan Allen-Zhu,et al.  Natasha 2: Faster Non-Convex Optimization Than SGD , 2017, NeurIPS.

[53]  Ewin Tang,et al.  A quantum-inspired classical algorithm for recommendation systems , 2018, Electron. Colloquium Comput. Complex..

[54]  Tianbao Yang,et al.  First-order Stochastic Algorithms for Escaping From Saddle Points in Almost Linear Time , 2017, NeurIPS.

[55]  Tengyu Ma,et al.  Gradient Descent Learns Linear Dynamical Systems , 2016, J. Mach. Learn. Res..

[56]  Yuanzhi Li,et al.  Neon2: Finding Local Minima via First-Order Oracles , 2017, NeurIPS.

[57]  Nathan Wiebe,et al.  Hamiltonian Simulation in the Interaction Picture , 2018, 1805.00675.

[58]  Yair Carmon,et al.  Accelerated Methods for NonConvex Optimization , 2018, SIAM J. Optim..

[59]  Iordanis Kerenidis,et al.  A Quantum Interior Point Method for LPs and SDPs , 2018, ACM Transactions on Quantum Computing.

[60]  Michael I. Jordan,et al.  Stochastic Cubic Regularization for Fast Nonconvex Optimization , 2017, NeurIPS.

[61]  Mert Gürbüzbalaban,et al.  Global Convergence of Stochastic Gradient Hamiltonian Monte Carlo for Non-Convex Stochastic Optimization: Non-Asymptotic Performance Bounds and Momentum-Based Acceleration , 2018, Oper. Res..

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

[63]  John C. Platt,et al.  Quantum supremacy using a programmable superconducting processor , 2019, Nature.

[64]  Srinivasan Arunachalam,et al.  Optimizing quantum optimization algorithms via faster quantum gradient computation , 2017, SODA.

[65]  Liu Liu,et al.  Quantum algorithm for finding the negative curvature direction in non-convex optimization , 2019, ArXiv.

[66]  András Gilyén,et al.  Improvements in Quantum SDP-Solving with Applications , 2018, ICALP.

[67]  Xiaodi Wu,et al.  Quantum SDP Solvers: Large Speed-Ups, Optimality, and Applications to Quantum Learning , 2017, ICALP.

[68]  Michael I. Jordan,et al.  Stochastic Gradient Descent Escapes Saddle Points Efficiently , 2019, ArXiv.

[69]  Tongyang Li,et al.  Quantum-inspired classical sublinear-time algorithm for solving low-rank semidefinite programming via sampling approaches , 2019, ArXiv.

[70]  Zhouchen Lin,et al.  Sharp Analysis for Nonconvex SGD Escaping from Saddle Points , 2019, COLT.

[71]  I. Chuang,et al.  Hamiltonian Simulation by Qubitization , 2016, Quantum.

[72]  Michael I. Jordan,et al.  On Nonconvex Optimization for Machine Learning , 2019, J. ACM.

[73]  Pedro C. S. Costa,et al.  Quantum algorithm for simulating the wave equation , 2017, Physical Review A.

[74]  Ronald de Wolf,et al.  Quantum SDP-Solvers: Better Upper and Lower Bounds , 2017, 2017 IEEE 58th Annual Symposium on Foundations of Computer Science (FOCS).

[75]  Ronald de Wolf,et al.  Convex optimization using quantum oracles , 2018, Quantum.

[76]  Seth Lloyd,et al.  Quantum-Inspired Algorithms for Solving Low-Rank Linear Equation Systems with Logarithmic Dependence on the Dimension , 2020, ISAAC.

[77]  Tongyang Li,et al.  Sampling-based sublinear low-rank matrix arithmetic framework for dequantizing Quantum machine learning , 2019, STOC.

[78]  Zhao Song,et al.  An improved quantum-inspired algorithm for linear regression , 2020, Quantum.

[79]  Tongyang Li,et al.  Quantum-Inspired Sublinear Algorithm for Solving Low-Rank Semidefinite Programming , 2019, MFCS.

[80]  Shouvanik Chakrabarti,et al.  Quantum algorithms and lower bounds for convex optimization , 2018, Quantum.

[81]  Quanquan Gu,et al.  Stochastic Recursive Variance-Reduced Cubic Regularization Methods , 2019, AISTATS.

[82]  Andrew M. Childs,et al.  High-precision quantum algorithms for partial differential equations , 2020, Quantum.

[83]  Minh C. Tran,et al.  Theory of Trotter Error with Commutator Scaling , 2021 .

[84]  Michael I. Jordan,et al.  On Nonconvex Optimization for Machine Learning , 2021, J. ACM.

[85]  A. Montanaro,et al.  Faster Quantum-inspired Algorithms for Solving Linear Systems , 2021, ACM Transactions on Quantum Computing.