Accelerated Gradient Methods Combining Tikhonov Regularization with Geometric Damping Driven by the Hessian

被引:15
作者
Attouch, Hedy [1 ]
Balhag, Aicha [2 ]
Chbani, Zaki [3 ]
Riahi, Hassan [3 ]
机构
[1] Univ Montpellier, IMAG, CNRS, Montpellier, France
[2] Univ Bourgogne Franche Comte, Inst Math Bourgogne, UMR 5584 CNRS, F-2100 Dijon, France
[3] Cadi Ayyad Univ, Semlalia Fac Sci, Marrakech 40000, Morocco
关键词
Accelerated gradient methods; Convex optimization; Damped inertial dynamics; Hessian-driven damping; Hierarchical minimization; Nesterov accelerated gradient method; Tikhonov approximation; STRONG ASYMPTOTIC CONVERGENCE; EVOLUTION-EQUATIONS; CONVEX-OPTIMIZATION; INERTIAL DYNAMICS; MONOTONE INCLUSIONS; SYSTEM; STABILIZATION; ALGORITHM; BEHAVIOR;
D O I
10.1007/s00245-023-09997-x
中图分类号
O29 [应用数学];
学科分类号
070104 ;
摘要
In a Hilbert framework, for general convex differentiable optimization, we consider accelerated gradient dynamics combining Tikhonov regularization with Hessian-driven damping. The temporal discretization of these dynamics leads to a new class of first-order optimization algorithms with favorable properties. The Tikhonov regularization parameter is assumed to tend to zero as time tends to infinity, which preserves equilibria. The presence of the Tikhonov regularization term induces a strong property of convexity which vanishes asymptotically. To take advantage of the fast convergence rates attached to the heavy ball method in the strongly convex case, we consider inertial dynamics where the viscous damping coefficient is proportional to the square root of the Tikhonov regularization parameter, and hence converges to zero. The geometric damping, controlled by the Hessian of the function to be minimized, induces attenuation of the oscillations. Under an appropriate setting of the parameters, based on Lyapunov's analysis, we show that the trajectories provide at the same time several remarkable properties: fast convergence of values, fast convergence of gradients towards zero, and strong convergence to the minimum norm minimizer. We show that the corresponding proximal algorithms share the same properties as continuous dynamics. The numerical illustrations confirm the results obtained. This study extends a previous paper by the authors regarding similar problems without the presence of Hessian driven damping.
引用
收藏
页数:52
相关论文
共 50 条
[1]   Newton-Like Dynamics and Forward-Backward Methods for Structured Monotone Inclusions in Hilbert Spaces [J].
Abbas, B. ;
Attouch, H. ;
Svaiter, Benar F. .
JOURNAL OF OPTIMIZATION THEORY AND APPLICATIONS, 2014, 161 (02) :331-360
[2]   Newton-Type Inertial Algorithms for Solving Monotone Equations Governed by Sums of Potential and Nonpotential Operators [J].
Adly, Samir ;
Attouch, Hedy ;
Van Nam Vo .
APPLIED MATHEMATICS AND OPTIMIZATION, 2022, 85 (03)
[3]  
Alvarez F, 2006, DISCRETE CONT DYN-A, V15, P921
[4]  
[Anonymous], 1972, Lecture Notes
[5]  
Attouch H, 2022, Arxiv, DOI arXiv:2201.11643
[6]   A CONTINUOUS DYNAMICAL NEWTON-LIKE APPROACH TO SOLVING MONOTONE INCLUSIONS [J].
Attouch, H. ;
Svaiter, B. F. .
SIAM JOURNAL ON CONTROL AND OPTIMIZATION, 2011, 49 (02) :574-598
[7]   Asymptotic control and stabilization of nonlinear oscillators with non-isolated equilibria [J].
Attouch, H ;
Czarnecki, MO .
JOURNAL OF DIFFERENTIAL EQUATIONS, 2002, 179 (01) :278-310
[8]   A dynamical approach to convex minimization coupling approximation with the steepest descent method [J].
Attouch, H ;
Cominetti, R .
JOURNAL OF DIFFERENTIAL EQUATIONS, 1996, 128 (02) :519-540
[9]   Viscosity solutions of minimization problems [J].
Attouch, H .
SIAM JOURNAL ON OPTIMIZATION, 1996, 6 (03) :769-806
[10]  
Attouch H., arXiv