High Impact Presentation (Metaheuristics)
Transcript: High Impact Presentations By: Emad Abdalghaffar Metaheuristic Optimization Algorithms ITI Intake 42 - AI 2021 Let us DIVE Inspiration Optimization is the heart of many natural processes: - Natural Selection in the biological evolution, which is based on the the survival-of-the-fittest principle. - Social Swarming of birds or foraging strategies of ants. - Other Natural-Inspired behaviors. Simulating such natural behaviors into computational mechanics in form of computer algorithms, may lead to promising approaches in the nowadays computer-aided engineering designs. Definition Classification of The Optimization Approaches Exploration and Exploitation Balance in MOAs “Optimization” Simple as billiards,... To find an optimal value of a function. > Corresponding to pre-defined variables (DV) > Within a feasible space that bounds those DV. Exploration: The optimizer scans and explores the entire search domain, looking for some promising spots. While Exploitation: The optimizer scans each of these promising spots. Considering the space & time complexity as KPIs. ....derivative-based VS derivative-free Enabling a suitable balance between exploration and exploitation highly increases the optimizer efficiency. Let us EXPLORE MO's State of The Art Trendy & Promising MOs Evolutionary Evolutionary Genetic Algorithm Genetic Algorithm (GA): Is an adaptive heuristic search algorithm based on the evolutionary principles of natural selection and genetics. "survival of the fittest" Swarm Swarm Grey Wolf Optimizer Grasshopper Optimization Algorithm Krill Herd Algorithm It mimics the leadership and hunting mechanism of grey wolves in nature. the prey may be either a randomly chosen solution or the best obtained solution so far, in case of exploration or exploitation, respectively The switch between the exploration and exploitation components is controlled by linear reduction of the parameter “a” from 2 to 0. As, when 0 < a < 1, the value of |A| < 1 which activates the exploitation mechanism, and the opposite occurs when 1 < a < 2. The GWO has only two adaptive parameters: “a” and “C” “a” guarantees the exploration-exploitation balance “C” emphasizes the random search It mimics the swarming behavior of grasshoppers. The grasshopper movement is based mainly on the social interaction with its neighbors. Which is described in terms of attraction, repulsion, or comfort zone state. Next position is based on: - current position (unlike PSO) - target position - other grasshoppers position C is a shrinking coefficient of each zone.. (override the local optimial traps) Decreasing the factor c ensures the gradual convergence (balance between explor. and exloit.). Inspired by the krill (fish) swarming. The random walk depends on: - the swarm movement - seeking food - random motion the krill's objective is a minimum distances of both the center of the herd density and from the food location. It holds the GA operators: crossover and mutation Other Nature-Inspired Other Nature-Inspired Ant Lion Optimizer Imperialist Competitive Algorithm Whale Optimization Algorithm It mimics the smart hunting mechanism of the Antlions. The ALO mimics the interaction between antlions and ants during the hunting process Ants and antlions position: - random walk - building traps - catching preys the ant replaces the antlion if provides better fitness - re-building traps ICA is an evolutionary sociopolitical algorithm. possible solutions are considered as countries. The evolutionary improvement is done through these operators: - Initial population, the best fittneses are imperialists, the rest are colonies. Each imperialist takes over a portion of the colonies according to its power. - Assimilation: where the imperialist influeces the colony country. (like the French and British cultures among their colonies) - Revolution - Title Exchange - Empires survival/collapse - Convergence: when just one empire is survived (called grand empire) The WOA is inspired by the bubble-net hunting strategy. The Arithmetic Optimization Algorithm Inspired by the main arithmetic operators in mathematics: addition (A) subtraction (S) multiplication (M) division (D). Exploration: D & M > high dispersion Exploitation: A & S > low dispersion Let us SPOT few trends Trends: Compatible with ML and DL: - The combination between MOAs and NN is a trending direction in the modern literature that shows high capability in reducing the computational time and efforts. - Equipping AI features (e.g., Fuzzy Logic) within an optimizer, is also trendy as it gets the process done faster and more open to handle general families of optimization problems. - It is trending recently to use a random-search optimization algorithm during the exploration phase just to spot the most promising areas within the search space. Then to employ a deterministic or derivative-based optimization algorithm in order to exploit that promising areas. Reference: www.seyedalimirjalili.com Thanks!
Prezi Present
Prezi Video
Prezi Design
