DOI: 10.1201/9780429434358-7
C H A P T E R 7
Online optimization
algorithms
CONTENTS
7.1
General considerations of online optimization . . . . . . . . . . . . . 170
7.1.1 Need for online optimization of accelerators . . . . . . . 170
7.1.2 Formulating online optimization problem . . . . . . . . . 171
7.1.3 Practical considerations for online optimization
implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
7.2
Review of optimization algorithms . . . . . . . . . . . . . . . . . . . . . . . . 178
7.2.1 Deterministic optimization algorithms . . . . . . . . . . . . . 178
7.2.1.1
Gradient-based methods . . . . . . . . . . . . . . . 178
7.2.1.2
Gradient-free methods . . . . . . . . . . . . . . . . . 181
7.2.2 Stochastic optimization algorithms . . . . . . . . . . . . . . . . 185
7.2.3 Machine learning (ML) methods . . . . . . . . . . . . . . . . . . . 189
7.3
Algorithms for online optimization . . . . . . . . . . . . . . . . . . . . . . . . 193
7.3.1 Testing of traditional algorithms . . . . . . . . . . . . . . . . . . 193
7.3.2 RCDS algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
7.3.3 RSimplex algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
7.3.4 Performance comparison for single-objective
algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
7.3.5 Testing of multi-objective optimization with
stochastic algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
Beam-based optimization is the approach of adjusting the control parameters
of the accelerator to optimize its performance, using the real-time, measured
beam performance as the guide for choosing the trial settings. Manual tuning is
a type of beam-based optimization in its original form. While manual tuning is
an indispensable approach and is widely used, its application is usually limited
to simple, small-scale problems. Automated tuning is an advanced form of
beam-based optimization, in which the computer takes the role of a human
being to make decisions. By closely interacting with the control system to
169
C H A P T E R 7
Online optimization
algorithms
CONTENTS
7.1
General considerations of online optimization . . . . . . . . . . . . . 170
7.1.1 Need for online optimization of accelerators . . . . . . . 170
7.1.2 Formulating online optimization problem . . . . . . . . . 171
7.1.3 Practical considerations for online optimization
implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
7.2
Review of optimization algorithms . . . . . . . . . . . . . . . . . . . . . . . . 178
7.2.1 Deterministic optimization algorithms . . . . . . . . . . . . . 178
7.2.1.1
Gradient-based methods . . . . . . . . . . . . . . . 178
7.2.1.2
Gradient-free methods . . . . . . . . . . . . . . . . . 181
7.2.2 Stochastic optimization algorithms . . . . . . . . . . . . . . . . 185
7.2.3 Machine learning (ML) methods . . . . . . . . . . . . . . . . . . . 189
7.3
Algorithms for online optimization . . . . . . . . . . . . . . . . . . . . . . . . 193
7.3.1 Testing of traditional algorithms . . . . . . . . . . . . . . . . . . 193
7.3.2 RCDS algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
7.3.3 RSimplex algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
7.3.4 Performance comparison for single-objective
algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
7.3.5 Testing of multi-objective optimization with
stochastic algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
Beam-based optimization is the approach of adjusting the control parameters
of the accelerator to optimize its performance, using the real-time, measured
beam performance as the guide for choosing the trial settings. Manual tuning is
a type of beam-based optimization in its original form. While manual tuning is
an indispensable approach and is widely used, its application is usually limited
to simple, small-scale problems. Automated tuning is an advanced form of
beam-based optimization, in which the computer takes the role of a human
being to make decisions. By closely interacting with the control system to
169
C H A P T E R 7
Online optimization
algorithms
CONTENTS
7.1
General considerations of online optimization . . . . . . . . . . . . . 170
7.1.1 Need for online optimization of accelerators . . . . . . . 170
7.1.2 Formulating online optimization problem . . . . . . . . . 171
7.1.3 Practical considerations for online optimization
implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
7.2
Review of optimization algorithms . . . . . . . . . . . . . . . . . . . . . . . . 178
7.2.1 Deterministic optimization algorithms . . . . . . . . . . . . . 178
7.2.1.1
Gradient-based methods . . . . . . . . . . . . . . . 178
7.2.1.2
Gradient-free methods . . . . . . . . . . . . . . . . . 181
7.2.2 Stochastic optimization algorithms . . . . . . . . . . . . . . . . 185
7.2.3 Machine learning (ML) methods . . . . . . . . . . . . . . . . . . . 189
7.3
Algorithms for online optimization . . . . . . . . . . . . . . . . . . . . . . . . 193
7.3.1 Testing of traditional algorithms . . . . . . . . . . . . . . . . . . 193
7.3.2 RCDS algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
7.3.3 RSimplex algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
7.3.4 Performance comparison for single-objective
algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
7.3.5 Testing of multi-objective optimization with
stochastic algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
Beam-based optimization is the approach of adjusting the control parameters
of the accelerator to optimize its performance, using the real-time, measured
beam performance as the guide for choosing the trial settings. Manual tuning is
a type of beam-based optimization in its original form. While manual tuning is
an indispensable approach and is widely used, its application is usually limited
to simple, small-scale problems. Automated tuning is an advanced form of
beam-based optimization, in which the computer takes the role of a human
being to make decisions. By closely interacting with the control system to
169
C H A P T E R 7
Online optimization
algorithms
CONTENTS
7.1
General considerations of online optimization . . . . . . . . . . . . . 170
7.1.1 Need for online optimization of accelerators . . . . . . . 170
7.1.2 Formulating online optimization problem . . . . . . . . . 171
7.1.3 Practical considerations for online optimization
implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
7.2
Review of optimization algorithms . . . . . . . . . . . . . . . . . . . . . . . . 178
7.2.1 Deterministic optimization algorithms . . . . . . . . . . . . . 178
7.2.1.1
Gradient-based methods . . . . . . . . . . . . . . . 178
7.2.1.2
Gradient-free methods . . . . . . . . . . . . . . . . . 181
7.2.2 Stochastic optimization algorithms . . . . . . . . . . . . . . . . 185
7.2.3 Machine learning (ML) methods . . . . . . . . . . . . . . . . . . . 189
7.3
Algorithms for online optimization . . . . . . . . . . . . . . . . . . . . . . . . 193
7.3.1 Testing of traditional algorithms . . . . . . . . . . . . . . . . . . 193
7.3.2 RCDS algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
7.3.3 RSimplex algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
7.3.4 Performance comparison for single-objective
algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
7.3.5 Testing of multi-objective optimization with
stochastic algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
Beam-based optimization is the approach of adjusting the control parameters
of the accelerator to optimize its performance, using the real-time, measured
beam performance as the guide for choosing the trial settings. Manual tuning is
a type of beam-based optimization in its original form. While manual tuning is
an indispensable approach and is widely used, its application is usually limited
to simple, small-scale problems. Automated tuning is an advanced form of
beam-based optimization, in which the computer takes the role of a human
being to make decisions. By closely interacting with the control system to
169
C H A P T E R 7
Online optimization
algorithms
CONTENTS
7.1
General considerations of online optimization . . . . . . . . . . . . . 170
7.1.1 Need for online optimization of accelerators . . . . . . . 170
7.1.2 Formulating online optimization problem . . . . . . . . . 171
7.1.3 Practical considerations for online optimization
implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
7.2
Review of optimization algorithms . . . . . . . . . . . . . . . . . . . . . . . . 178
7.2.1 Deterministic optimization algorithms . . . . . . . . . . . . . 178
7.2.1.1
Gradient-based methods . . . . . . . . . . . . . . . 178
7.2.1.2
Gradient-free methods . . . . . . . . . . . . . . . . . 181
7.2.2 Stochastic optimization algorithms . . . . . . . . . . . . . . . . 185
7.2.3 Machine learning (ML) methods . . . . . . . . . . . . . . . . . . . 189
7.3
Algorithms for online optimization . . . . . . . . . . . . . . . . . . . . . . . . 193
7.3.1 Testing of traditional algorithms . . . . . . . . . . . . . . . . . . 193
7.3.2 RCDS algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
7.3.3 RSimplex algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
7.3.4 Performance comparison for single-objective
algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
7.3.5 Testing of multi-objective optimization with
stochastic algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
Beam-based optimization is the approach of adjusting the control parameters
of the accelerator to optimize its performance, using the real-time, measured
beam performance as the guide for choosing the trial settings. Manual tuning is
a type of beam-based optimization in its original form. While manual tuning is
an indispensable approach and is widely used, its application is usually limited
to simple, small-scale problems. Automated tuning is an advanced form of
beam-based optimization, in which the computer takes the role of a human
being to make decisions. By closely interacting with the control system to
169
C H A P T E R 7
Online optimization
algorithms
CONTENTS
7.1
General considerations of online optimization . . . . . . . . . . . . . 170
7.1.1 Need for online optimization of accelerators . . . . . . . 170
7.1.2 Formulating online optimization problem . . . . . . . . . 171
7.1.3 Practical considerations for online optimization
implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
7.2
Review of optimization algorithms . . . . . . . . . . . . . . . . . . . . . . . . 178
7.2.1 Deterministic optimization algorithms . . . . . . . . . . . . . 178
7.2.1.1
Gradient-based methods . . . . . . . . . . . . . . . 178
7.2.1.2
Gradient-free methods . . . . . . . . . . . . . . . . . 181
7.2.2 Stochastic optimization algorithms . . . . . . . . . . . . . . . . 185
7.2.3 Machine learning (ML) methods . . . . . . . . . . . . . . . . . . . 189
7.3
Algorithms for online optimization . . . . . . . . . . . . . . . . . . . . . . . . 193
7.3.1 Testing of traditional algorithms . . . . . . . . . . . . . . . . . . 193
7.3.2 RCDS algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
7.3.3 RSimplex algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
7.3.4 Performance comparison for single-objective
algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
7.3.5 Testing of multi-objective optimization with
stochastic algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
Beam-based optimization is the approach of adjusting the control parameters
of the accelerator to optimize its performance, using the real-time, measured
beam performance as the guide for choosing the trial settings. Manual tuning is
a type of beam-based optimization in its original form. While manual tuning is
an indispensable approach and is widely used, its application is usually limited
to simple, small-scale problems. Automated tuning is an advanced form of
beam-based optimization, in which the computer takes the role of a human
being to make decisions. By closely interacting with the control system to
169