1: \begin{abstract}

2: 

3: First-order methods with momentum

4: such as Nesterov's fast gradient method

5: are very useful

6: for convex optimization problems,

7: but can exhibit undesirable oscillations

8: yielding slow convergence rates

9: for some applications.

10: An adaptive restarting scheme

11: can improve the convergence rate of the fast gradient method,

12: when the parameter

13: of a strongly convex cost function is unknown

14: or when the iterates of the algorithm

15: enter a locally strongly convex region.

16: Recently, we introduced the optimized gradient method,

17: a first-order algorithm

18: that has an inexpensive per-iteration computational cost

19: similar to that of the fast gradient method,

20: yet has a worst-case cost function rate

21: that is twice faster than that of the fast gradient method

22: and that is optimal for large-dimensional smooth convex problems.

23: Building upon the

24: success of accelerating the fast gradient method using adaptive restart,

25: this paper investigates similar heuristic acceleration of 

26: the optimized gradient method.

27: We first derive 

28: \cblue{a new first-order method 

29: that resembles the optimized gradient method}

30: %new step coefficients of the optimized gradient method

31: for strongly convex quadratic problems

32: with known function parameters,

33: yielding a linear convergence rate

34: that is faster than that of the analogous version of 

35: the fast gradient method.

36: We then provide a heuristic analysis

37: and numerical experiments

38: that illustrate that adaptive restart

39: can accelerate the convergence of the optimized gradient method.

40: Numerical results also illustrate

41: that adaptive restart

42: is helpful

43: for a proximal version of the optimized gradient method

44: for nonsmooth composite convex functions.

45: %

46: \keywords{

47: Convex optimization

48: \and First-order methods

49: \and Accelerated gradient methods

50: \and Optimized gradient method

51: \and Restarting}

52: \subclass{80M50 \and 90C06 \and 90C25}

53: \end{abstract}

54: