62 views

Skip to first unread message

Feb 14, 2022, 6:34:18 PM2/14/22

to

Hello,

More of my philosophy about the poor local search ability of particle

swarm optimization (PSO) in artificial intelligence and about artificial

intelligence..

I am a white arab from Morocco, and i think i am smart since i have also

invented many scalable algorithms and algorithms..

I will explain something important about particle swarm optimization

(PSO) in artificial intelligence:

In many research papers, it is proved that particle swarm optimization

(PSO) in artificial intelligence could provide faster convergence and

could find better solutions when compared to GA(genetic algorithm). The

implementation of PSO is also simple. But the main disadvantage of PSO

is with its poor local search ability. But you have to understand that

the poor local search ability of PSO is much more compensated by its

"faster" convergence, this is why i think that PSO is really useful

since you can even guarantee the optimal convergence in PSO as i am

learning you in my below thoughts and writing, so read them carefully.

More of my philosophy about evolutionary algorithms and artificial

intelligence..

You can read more about my education and my way of doing here:

Here is more proof of the fact that i have invented many scalable

algorithms and algorithms:

https://groups.google.com/g/comp.programming.threads/c/V9Go8fbF10k

And you can take a look at my photo that i have just put

here in my website(I am 53 years old):

https://sites.google.com/site/scalable68/jackson-network-problem

I think i am smart and I will explain more evolutionary algorithms such

as particle swarm optimization (PSO) and the genetic algorithm(and also

don't forget to read carefully my below new interesting proverb):

I think that Modern trends in solving tough optimization problems tend

to use evolutionary algorithms and nature-inspired metaheuristic

algorithms, especially those based on swarm intelligence (SI), two major

characteristics of modern metaheuristic methods are nature-inspired, and

a balance between randomness and regularity. And notice that i am

talking smartly below about the powerful modern evolutionary algorithm

that we call particle swarm optimization (PSO), and i think that the

powerful modern evolutionary algorithm that we call particle swarm

optimization (PSO) is also a balanced use of randomness with a proper

combination with certain deterministic components that is in fact the

essence of making such algorithms so powerful and effective, and notice

that the randommness in a genetic algorithm (GA) comes from the

randomness of mutations of chromosomes or in PSO it comes from the size

of the population that is constituted with the members that search also

randomly, and this randomness in artificial intelligence like PSO

and Reinforcement learning permits to move forward towards a better

global optimum of efficiency, and if the randomness in an algorithm is

too high, then the solutions generated by the algorithm do not converge

easily as they could continue to "jump around" in the search space. If

there is no randomness at all, then they can suffer the same

disadvantages as those of deterministic methods (such as the

gradient-based search). Therefore, a certain tradeoff is needed.

More of my my philosophy about the Exploration/Exploitation trade off in

AI(artificial intelligence)..

In Reinforcement Learning in AI(artificial intelligence), for each

action (i.e. lever) on the machine, there is an expected reward. If this

expected reward is known to the Agent, then the problem degenerates into

a trivial one, which merely involves picking the action with the highest

expected reward. But since the expected rewards for the levers are not

known, we have to collate estimates to get an idea of the desirability

of each action. For this, the Agent will have to explore to get the

average of the rewards for each action. After, it can then exploit its

knowledge and choose an action with the highest expected rewards (this

is also called selecting a greedy action). As we can see, the Agent has

to balance exploring and exploiting actions to maximize the overall

long-term reward. So as you are noticing i am posting below my

just new proverb that talks about the Exploration/Exploitation trade off

in AI(artificial intelligence), and you also have to know how to build

correctly "trust" between you and the others so that to optimize

correctly, and this is why you are seeing me posting my thoughts like i

am posting.

You have to know about the Exploration/Exploitation trade off in

Reinforcement Learning and PSO(Particle Swarm Optimization) in AI by

knowing the following and by reading my below thoughts about artificial

intelligence:

Exploration is finding more information about the environment.

Exploitation is exploiting known information to maximize the reward.

This is why i have just invented fast the following proverb that also

talks about this Exploration/Exploitation trade off in AI (artificial

intelligence):

And here is my just new proverb:

"Human vitality comes from intellectual openness and intellectual

openness also comes from divergent thinking and you have to well balance

divergent thinking with convergent thinking so that to converge towards

the global optimum of efficiency and not get stuck on a local optimum of

efficiency, and this kind of well balancing makes the good creativity."

And i will explain more my proverb so that you understand it:

I think that divergent thinking is thought process or method used to

generate creative ideas by exploring many possible solutions, but notice

that we even need openness in a form of economic actors that share ideas

across nations and industries (and this needs globalization) that make

us much more creative and that's good for economy, since you can easily

notice that globalization also brings a kind of optimality to divergent

thinking, and also you have to know how to balance divergent thinking

with convergent thinking, since if divergent thinking is much greater

than convergent thinking it can become costly in terms of time, and if

the convergent thinking is much greater than divergent thinking you can

get stuck on local optimum of efficiency and not converge to a global

optimum of efficiency, and it is related to my following thoughts about

the philosopher and economist Adam Smith, so i invite you to read them:

https://groups.google.com/g/alt.culture.morocco/c/ftf3lx5Rzxo

More philosophy about what is artificial intelligence and more..

I am a white arab, and i think i am smart since i have also invented

many scalable algorithms and algorithms, and when you are smart you will

easily understand artificial intelligence, this is why i am finding

artificial intelligence easy to learn, i think to be able to understand

artificial intelligence you have to understand reasoning with energy

minimization, like with PSO(Particle Swarm Optimization), but

you have to be smart since the Population based algorithm has to

guarantee the optimal convergence, and this is why i am learning

you how to do it(read below), i think that GA(genetic algorithm) is

good for teaching it, but GA(genetic algorithm) doesn't guarantee the

optimal convergence, and after learning how to do reasoning with energy

minimization in artificial intelligence, you have to understand what is

transfer learning in artificial intelligence with PathNet or such, this

transfer learning permits to train faster and require less labeled data,

also PathNET is much more powerful since also it is higher level

abstraction in artificial intelligence..

Read about it here:

https://mattturck.com/frontierai/

And read about PathNet here:

https://medium.com/@thoszymkowiak/deepmind-just-published-a-mind-blowing-paper-pathnet-f72b1ed38d46

More about artificial intelligence..

I think one of the most important part in artificial intelligence is

reasoning with energy minimization, it is the one that i am working on

right now, see the following video to understand more about it:

Yann LeCun: Can Neural Networks Reason?

https://www.youtube.com/watch?v=YAfwNEY826I&t=250s

I think that since i have just understood much more artificial

intelligence, i will soon show you my next Open source software project

that implement a powerful Parallel Linear programming solver and a

powerful Parallel Mixed-integer programming solver with Artificial

intelligence using PSO, and i will write an article that explain

much more artificial intelligence and what is smartness and what is

consciousness and self-awareness..

And in only one day i have just learned "much" more artificial

intelligence, i have read the following article about Particle Swarm

Optimization and i have understood it:

Artificial Intelligence - Particle Swarm Optimization

https://docs.microsoft.com/en-us/archive/msdn-magazine/2011/august/artificial-intelligence-particle-swarm-optimization

But i have just noticed that the above implementation doesn't guarantee

the optimal convergence.

So here is how to guarantee the optimal convergence in PSO:

Clerc and Kennedy in (Trelea 2003) propose a constriction coefficient

parameter selection guidelines in order to guarantee the optimal

convergence, here is how to do it with PSO:

v(t+1) = k*[(v(t) + (c1 * r1 * (p(t) – x(t)) + (c2 * r2 * (g(t) – x(t))]

x(t+1) = x(t) + v(t+1)

constriction coefficient parameter is:

k = 2/abs(2-phi-sqrt(phi^2-(4*phi)))

k:=2/abs((2-4.1)-(0.640)) = 0.729

phi = c1 + c2

To guarantee the optimal convergence use:

c1 = c2 = 2.05

phi = 4.1 => k equal to 0.729

w=0.7298

Population size = 60;

Also i have noticed that GA(genetic algorithm) doesn't guarantee the

optimal convergence, and SA(Simulated annealing) and Hill Climbing are

much less powerful since they perform only exploitation.

In general, any metaheuristic should perform two main searching

capabilities (Exploration and Exploitation). Population based algorithms

( or many solutions ) such as GA, PSO, ACO, or ABC, performs both

Exploration and Exploitation, while Single-Based Algorithm such as

SA(Simulated annealing), Hill Climbing, performs the exploitation only.

In this case, more exploitation and less exploration increases the

chances for trapping in local optima. Because the algorithm does not

have the ability to search in another position far from the current best

solution ( which is Exploration).

Simulated annealing starts in one valley and typically ends in the

lowest point of the same valley. Whereas swarms start in many different

places of the mountain range and are searching for the lowest point in

many valleys simultaneously.

And in my next Open source software project i will implement a powerful

Parallel Linear programming solver and a powerful Parallel Mixed-integer

programming solver with Artificial intelligence using PSO.

And read my following thoughts of my philosophy about what is smartness:

https://groups.google.com/g/alt.culture.morocco/c/Wzf6AOl41xs

And read my following thoughts of my philosophy about my new proverbs

and about dignity:

https://groups.google.com/g/alt.culture.morocco/c/72FdpcFe9Vk

Thank you,

Amine Moulay Ramdane.

More of my philosophy about the poor local search ability of particle

swarm optimization (PSO) in artificial intelligence and about artificial

intelligence..

I am a white arab from Morocco, and i think i am smart since i have also

invented many scalable algorithms and algorithms..

I will explain something important about particle swarm optimization

(PSO) in artificial intelligence:

In many research papers, it is proved that particle swarm optimization

(PSO) in artificial intelligence could provide faster convergence and

could find better solutions when compared to GA(genetic algorithm). The

implementation of PSO is also simple. But the main disadvantage of PSO

is with its poor local search ability. But you have to understand that

the poor local search ability of PSO is much more compensated by its

"faster" convergence, this is why i think that PSO is really useful

since you can even guarantee the optimal convergence in PSO as i am

learning you in my below thoughts and writing, so read them carefully.

More of my philosophy about evolutionary algorithms and artificial

intelligence..

You can read more about my education and my way of doing here:

Here is more proof of the fact that i have invented many scalable

algorithms and algorithms:

https://groups.google.com/g/comp.programming.threads/c/V9Go8fbF10k

And you can take a look at my photo that i have just put

here in my website(I am 53 years old):

https://sites.google.com/site/scalable68/jackson-network-problem

I think i am smart and I will explain more evolutionary algorithms such

as particle swarm optimization (PSO) and the genetic algorithm(and also

don't forget to read carefully my below new interesting proverb):

I think that Modern trends in solving tough optimization problems tend

to use evolutionary algorithms and nature-inspired metaheuristic

algorithms, especially those based on swarm intelligence (SI), two major

characteristics of modern metaheuristic methods are nature-inspired, and

a balance between randomness and regularity. And notice that i am

talking smartly below about the powerful modern evolutionary algorithm

that we call particle swarm optimization (PSO), and i think that the

powerful modern evolutionary algorithm that we call particle swarm

optimization (PSO) is also a balanced use of randomness with a proper

combination with certain deterministic components that is in fact the

essence of making such algorithms so powerful and effective, and notice

that the randommness in a genetic algorithm (GA) comes from the

randomness of mutations of chromosomes or in PSO it comes from the size

of the population that is constituted with the members that search also

randomly, and this randomness in artificial intelligence like PSO

and Reinforcement learning permits to move forward towards a better

global optimum of efficiency, and if the randomness in an algorithm is

too high, then the solutions generated by the algorithm do not converge

easily as they could continue to "jump around" in the search space. If

there is no randomness at all, then they can suffer the same

disadvantages as those of deterministic methods (such as the

gradient-based search). Therefore, a certain tradeoff is needed.

More of my my philosophy about the Exploration/Exploitation trade off in

AI(artificial intelligence)..

In Reinforcement Learning in AI(artificial intelligence), for each

action (i.e. lever) on the machine, there is an expected reward. If this

expected reward is known to the Agent, then the problem degenerates into

a trivial one, which merely involves picking the action with the highest

expected reward. But since the expected rewards for the levers are not

known, we have to collate estimates to get an idea of the desirability

of each action. For this, the Agent will have to explore to get the

average of the rewards for each action. After, it can then exploit its

knowledge and choose an action with the highest expected rewards (this

is also called selecting a greedy action). As we can see, the Agent has

to balance exploring and exploiting actions to maximize the overall

long-term reward. So as you are noticing i am posting below my

just new proverb that talks about the Exploration/Exploitation trade off

in AI(artificial intelligence), and you also have to know how to build

correctly "trust" between you and the others so that to optimize

correctly, and this is why you are seeing me posting my thoughts like i

am posting.

You have to know about the Exploration/Exploitation trade off in

Reinforcement Learning and PSO(Particle Swarm Optimization) in AI by

knowing the following and by reading my below thoughts about artificial

intelligence:

Exploration is finding more information about the environment.

Exploitation is exploiting known information to maximize the reward.

This is why i have just invented fast the following proverb that also

talks about this Exploration/Exploitation trade off in AI (artificial

intelligence):

And here is my just new proverb:

"Human vitality comes from intellectual openness and intellectual

openness also comes from divergent thinking and you have to well balance

divergent thinking with convergent thinking so that to converge towards

the global optimum of efficiency and not get stuck on a local optimum of

efficiency, and this kind of well balancing makes the good creativity."

And i will explain more my proverb so that you understand it:

I think that divergent thinking is thought process or method used to

generate creative ideas by exploring many possible solutions, but notice

that we even need openness in a form of economic actors that share ideas

across nations and industries (and this needs globalization) that make

us much more creative and that's good for economy, since you can easily

notice that globalization also brings a kind of optimality to divergent

thinking, and also you have to know how to balance divergent thinking

with convergent thinking, since if divergent thinking is much greater

than convergent thinking it can become costly in terms of time, and if

the convergent thinking is much greater than divergent thinking you can

get stuck on local optimum of efficiency and not converge to a global

optimum of efficiency, and it is related to my following thoughts about

the philosopher and economist Adam Smith, so i invite you to read them:

https://groups.google.com/g/alt.culture.morocco/c/ftf3lx5Rzxo

More philosophy about what is artificial intelligence and more..

I am a white arab, and i think i am smart since i have also invented

many scalable algorithms and algorithms, and when you are smart you will

easily understand artificial intelligence, this is why i am finding

artificial intelligence easy to learn, i think to be able to understand

artificial intelligence you have to understand reasoning with energy

minimization, like with PSO(Particle Swarm Optimization), but

you have to be smart since the Population based algorithm has to

guarantee the optimal convergence, and this is why i am learning

you how to do it(read below), i think that GA(genetic algorithm) is

good for teaching it, but GA(genetic algorithm) doesn't guarantee the

optimal convergence, and after learning how to do reasoning with energy

minimization in artificial intelligence, you have to understand what is

transfer learning in artificial intelligence with PathNet or such, this

transfer learning permits to train faster and require less labeled data,

also PathNET is much more powerful since also it is higher level

abstraction in artificial intelligence..

Read about it here:

https://mattturck.com/frontierai/

And read about PathNet here:

https://medium.com/@thoszymkowiak/deepmind-just-published-a-mind-blowing-paper-pathnet-f72b1ed38d46

More about artificial intelligence..

I think one of the most important part in artificial intelligence is

reasoning with energy minimization, it is the one that i am working on

right now, see the following video to understand more about it:

Yann LeCun: Can Neural Networks Reason?

https://www.youtube.com/watch?v=YAfwNEY826I&t=250s

I think that since i have just understood much more artificial

intelligence, i will soon show you my next Open source software project

that implement a powerful Parallel Linear programming solver and a

powerful Parallel Mixed-integer programming solver with Artificial

intelligence using PSO, and i will write an article that explain

much more artificial intelligence and what is smartness and what is

consciousness and self-awareness..

And in only one day i have just learned "much" more artificial

intelligence, i have read the following article about Particle Swarm

Optimization and i have understood it:

Artificial Intelligence - Particle Swarm Optimization

https://docs.microsoft.com/en-us/archive/msdn-magazine/2011/august/artificial-intelligence-particle-swarm-optimization

But i have just noticed that the above implementation doesn't guarantee

the optimal convergence.

So here is how to guarantee the optimal convergence in PSO:

Clerc and Kennedy in (Trelea 2003) propose a constriction coefficient

parameter selection guidelines in order to guarantee the optimal

convergence, here is how to do it with PSO:

v(t+1) = k*[(v(t) + (c1 * r1 * (p(t) – x(t)) + (c2 * r2 * (g(t) – x(t))]

x(t+1) = x(t) + v(t+1)

constriction coefficient parameter is:

k = 2/abs(2-phi-sqrt(phi^2-(4*phi)))

k:=2/abs((2-4.1)-(0.640)) = 0.729

phi = c1 + c2

To guarantee the optimal convergence use:

c1 = c2 = 2.05

phi = 4.1 => k equal to 0.729

w=0.7298

Population size = 60;

Also i have noticed that GA(genetic algorithm) doesn't guarantee the

optimal convergence, and SA(Simulated annealing) and Hill Climbing are

much less powerful since they perform only exploitation.

In general, any metaheuristic should perform two main searching

capabilities (Exploration and Exploitation). Population based algorithms

( or many solutions ) such as GA, PSO, ACO, or ABC, performs both

Exploration and Exploitation, while Single-Based Algorithm such as

SA(Simulated annealing), Hill Climbing, performs the exploitation only.

In this case, more exploitation and less exploration increases the

chances for trapping in local optima. Because the algorithm does not

have the ability to search in another position far from the current best

solution ( which is Exploration).

Simulated annealing starts in one valley and typically ends in the

lowest point of the same valley. Whereas swarms start in many different

places of the mountain range and are searching for the lowest point in

many valleys simultaneously.

And in my next Open source software project i will implement a powerful

Parallel Linear programming solver and a powerful Parallel Mixed-integer

programming solver with Artificial intelligence using PSO.

And read my following thoughts of my philosophy about what is smartness:

https://groups.google.com/g/alt.culture.morocco/c/Wzf6AOl41xs

And read my following thoughts of my philosophy about my new proverbs

and about dignity:

https://groups.google.com/g/alt.culture.morocco/c/72FdpcFe9Vk

Thank you,

Amine Moulay Ramdane.

Reply all

Reply to author

Forward

0 new messages

Search

Clear search

Close search

Google apps

Main menu