Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > comp.programming.threads > #6664 > unrolled thread

Boston Dynamics' Robots Can Do Parkour. Should We Worry?

Back to article view | Back to comp.programming.threads

Contents

  Boston Dynamics' Robots Can Do Parkour. Should We Worry? Amine Moulay Ramdane <aminer68@gmail.com> - 2021-08-17 13:30 -0700

#6664 — Boston Dynamics' Robots Can Do Parkour. Should We Worry?

Hello,


Boston Dynamics' Robots Can Do Parkour. Should We Worry?

Read more here:

https://interestingengineering.com/boston-dynamics-robots-parkour

I am a white arab, and i think i am smart since i have also
invented many scalable algorithms and algorithms..

You can read more of my thoughts about artificial intelligence here:

https://groups.google.com/g/alt.culture.morocco/c/5Xmlx8NS8F8

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

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.


Thank you,
Amine Moulay Ramdane.

[toc] | [standalone]

Back to top | Article view | comp.programming.threads

csiph-web