Download presentation
Presentation is loading. Please wait.
Published byArlene Daniels Modified over 6 years ago
1
Advanced Artificial Intelligence Evolutionary Search Algorithm
Chung-Ang University, Wangduk Seo The original version of this content is available from the Wikipedia Hello, my name is Wangduk Seo and my presentation title is an Evolutionary Search Algorithm
2
Introduction Through a history of computer science and engineering, many algorithms are developed and proposed to solve various difficult problems. Advanced Artificial Intelligence / Chung-Ang University / Wangduk Seo
3
However… However, this cartoon describes the difficulty of real world problem to adopt an algorithm to problem. In similar cases, there are gap between a theoretic approach and real world problem. Advanced Artificial Intelligence / Chung-Ang University / Wangduk Seo
4
However… Global Optima
Assuming there is a global optima which we want to find, one major reason may be hard to find global optima in proper time. Advanced Artificial Intelligence / Chung-Ang University / Wangduk Seo
5
as good as GLOBAL OPTIMA
New Approach Find good solution as good as GLOBAL OPTIMA Applicable to various problems So new approach was proposed. First, rather find a global optima, try to find approximate answer enough to compromise. Second, develop a general method that can be applicable to various problems. Advanced Artificial Intelligence / Chung-Ang University / Wangduk Seo
6
Heuristic and Metaheuristic
𝑺𝒐𝒍𝒖𝒕𝒊𝒐𝒏 𝟏 𝑺𝒐𝒍𝒖𝒕𝒊𝒐𝒏 𝟐 𝑺𝒐𝒍𝒖𝒕𝒊𝒐𝒏 𝟑 ... Available Information Better This new approach is called heuristic and metaheuristic. In greek, heuristic means “find” or “discover”. A heuristic is any approach to problem solving that employs method, not guaranteed to be optimal, but find approximate answers and improve them using available information in problem. Advanced Artificial Intelligence / Chung-Ang University / Wangduk Seo
7
Metaheuristic Metaheuristic ( = Higher level of heuristic) Heuristic
Applicable Metaheuristic is a higher-level heuristic designed to find, generate, or select a heuristic that may provide a sufficiently good solution to an optimization problem, especially with incomplete or imperfect information or limited computation capacity. Advanced Artificial Intelligence / Chung-Ang University / Wangduk Seo
8
Evolutionary Search Algorithm
Evolutionary Algorithm = Nature Inspired Metaheuristics Evolutionary algorithm (EA) is a subset of evolutionary computation, a generic population-based metaheuristic optimization algorithm. An EA uses mechanisms inspired by biological evolution, such as reproduction, mutation, recombination, and selection. Advanced Artificial Intelligence / Chung-Ang University / Wangduk Seo
9
Various Evolutionary Algorithms
Genetic Algorithm Particle Swarm Optimization Ant Colony Optimization Cuckoo Search Algorithm These are popular evolutionary algorithms. Genetic Algorithm is inspired by the process of natural selection. Particle swarm optimization is inspired the movement of organisms in a bird flock. Ant colony optimization is a probabilistic technique to find good path through graphs imitating a behavior of ant. Cuckoo search is inspired by the obligate brood parasitism of some cuckoo species by laying their eggs in other birds’ nests. Advanced Artificial Intelligence / Chung-Ang University / Wangduk Seo
10
General Steps General Steps are described in the picture. First, initialize the population. Then, evaluate each individual in the population. Finally, improve them by various operators such as selection, crossover, and mutation. Advanced Artificial Intelligence / Chung-Ang University / Wangduk Seo
11
2 Keypoints of Evolutionary Algorithm
“How can we find good approximate solution?” Exploration Exploitation There are 2 keypoints in evolutionary algorithms to find good approximated solution, which are exploration and exploitation. Advanced Artificial Intelligence / Chung-Ang University / Wangduk Seo
12
Exploration We should search solution space as large as possible!
Since search space is usually large, exploration is important that search large space can increase the chance to find good solutions. Advanced Artificial Intelligence / Chung-Ang University / Wangduk Seo
13
Exploitation We should search around good solutions!
Also, since good solutions are likely to be around other good solutions, exploitation is that search around good solutions. Advanced Artificial Intelligence / Chung-Ang University / Wangduk Seo
14
Exploration vs Exploitation – Trade off
However, these two concepts are conflicted to each other in terms of search space, and balancing them is one of the important problem of designing an evolutionary algorithms. Advanced Artificial Intelligence / Chung-Ang University / Wangduk Seo
15
Exploration vs Exploitation – Trade off
These graph shows that various algorithm balance 2 concepts variously because of different algorithm design purpose. Advanced Artificial Intelligence / Chung-Ang University / Wangduk Seo
16
Genetic Algorithm A genetic algorithm is the most popular algorithm in the evolutionary algorithms and one of the most studied algorithm. A genetic algorithm is famous for its wide variety of problems. Advanced Artificial Intelligence / Chung-Ang University / Wangduk Seo
17
Genetic Algorithm Genetic algorithm’s process is Initialization, Evaluation, Selection, Genetic Operators. Genetic operators are three steps that crossover, mutation and reproduction. Advanced Artificial Intelligence / Chung-Ang University / Wangduk Seo
18
Initialization and Evaluation
Initialization make individuals randomly or guided to start with good individuals. In evaluation step, all individuals are scored and ranked by its own fitness function such as path cost, accuracy. Advanced Artificial Intelligence / Chung-Ang University / Wangduk Seo
19
Selection In selection steps, algorithm select individuals to reproduce new individuals by genetic operators. This picture shows the Random Wheel Selection that individuals are selected biased by its evaluated score. Advanced Artificial Intelligence / Chung-Ang University / Wangduk Seo
20
Crossover Single-point Crossover Uniform Crossover Two-point Crossover Genetic operators make new individuals from selected individuals. Crossover produce new individuals such as swapping selected individuals. For example, single-point crossover choose one crossover point randomly and swap individuals based on the crossover point. Advanced Artificial Intelligence / Chung-Ang University / Wangduk Seo
21
Mutation Mutation is a genetic operator used to maintain genetic diversity from one generation of a population to the next. This mutation is swap mutation that swap entities randomly in a individual. Advanced Artificial Intelligence / Chung-Ang University / Wangduk Seo
22
Example “THISISANEGG” “THIS IS AN EGG”
Let’s see how genetic algorithm works. There is one sentence without any space. Our goal is insert space properly. Advanced Artificial Intelligence / Chung-Ang University / Wangduk Seo
23
Representation “THISISANEGG” “THIS IS AN EGG” Evaluation Score : 6 1 1
1 Search Space : 𝟐 𝟏𝟎 Evaluation Sample 1 Score : 6 Ground Truth Answer (Hard to know in real world) 1 These sentences can be represent by binary string that 0 means now space, and 1 means space. Choosing 0 or 1 in each entities, the search space of this problem is 2 power 10. To evaluate individuals, the ground truth answer are compared with the individuals counting the correct entities. Usually, the ground truth answer are blinded. Since this example’s purpose is to know how GA works, we assume that we know the ground truth answer for special case. Advanced Artificial Intelligence / Chung-Ang University / Wangduk Seo
24
Initialization and Evaluation
Randomly Initialized (Population size = 4) Crossover Score 1 1 6 Evaluation 1 1 7 1 1 4 Mutation 1 1 5 4 individuals are randomly initialized. Each individuals are evaluated by comparing the answer. Let’s assume two best individuals are selected for crossover and one individuals are selected randomly for mutation. Advanced Artificial Intelligence / Chung-Ang University / Wangduk Seo
25
Crossover Score 1 Crossover 1 5 1 1 8 Crossover point
1 Crossover 1 5 1 1 8 Crossover point By single point crossover, two new individuals are generated. Crossover point are randomly selected and swap them. The generated individuals’ scores are 5 and 8. Advanced Artificial Intelligence / Chung-Ang University / Wangduk Seo
26
Mutation Score Mutation 1 1 6
1 6 Swap mutation is operated to generate new individuals. Two points are randomly selected and swapped. The new generated individual’s score is 6. Advanced Artificial Intelligence / Chung-Ang University / Wangduk Seo
27
Make New Generation Score New Generation Score 1 1 1 1 1 6 1 1 1 1 7 1
1 6 New Generation Score 1 7 1 6 1 4 1 7 1 5 1 8 1 5 Crossover 1 6 1 8 1 6 Mutation Through the genetic operators, 3 individuals are generated and choose individuals order by its score as many as the size of population. Repeat until the stop criterion is satisfied. Advanced Artificial Intelligence / Chung-Ang University / Wangduk Seo
28
Design Issue in GA 1. Problem Modeling 2. Evaluation Method
3. Algorithm Selection in each stages There 3 algorithm design issues in Genetic Algorithm. First, Problem modeling such as data representation, hyper-parameter setting. Second, design evaluation method to identify good solutions and bad solutions. Finally, since there are several algorithms in each stages of Genetic Algorithm, appropriate algorithms must be selected for good performance. Advanced Artificial Intelligence / Chung-Ang University / Wangduk Seo
29
In real world… In real world, Evolutionary algorithm is applied various field. Left picture is NASA ST5 spacecraft antenna. This complicated shape was found by evolutionary computer design program to create the best radiation pattern. Right picture is find walking mechanism by Genetic Algorithm through generations. Advanced Artificial Intelligence / Chung-Ang University / Wangduk Seo
30
Q & A Advanced Artificial Intelligence / Chung-Ang University / Wangduk Seo
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.