Unit 10 2 Successes.mp4 knowitvideos Japanese English

Unit 10 2 Successes.mp4YouTube↗ IA↗(long shot)
Film & Animation 6.4K views 16 likes 0 dislikes 08 Nov 2011

knowitvideos 11.8K subscribers N/A YouTube↗

Click to Play Video

例えば4×3のグリッドワールドで

For example, in the 4 by 3 GridWorld,

出発時に＋1と－1の報酬がある場所を知らなかったとしたら？

what if we don't know where the plus 1 and minus 1 rewards are when we start out?

強化学習のエージェントは領域を探索して学び

A reinforcement learning agent can learn to explore the territory,

どこに報酬があるかを見つけて最適ポリシーを学びます

find where the rewards are, and then learn an optimal policy.

一方MDP解法は報酬の場所を正確に知っているだけでいいのです

Whereas, an MDP solver can only do that once it knows exactly where the rewards are.

この＋1と－1を動き回って見つける考えは

Now, this idea of wandering around and then finding a plus 1 or a minus 1

バックギャモンのような多くのゲームの形と似ています

is analogous to many forms of games, such as backgammon--

バックギャモンは確率的なゲームで

and here's an example: backgammon is a stochastic game;

最後は勝つか負けるかです

and at the end, you either win or lose.

1990年代にIBMでゲイリー・テザウロが

And in the 1990s, Gary Tesauro at IBM

バックギャモンで遊ぶプログラムを書きました

wrote a program to play backgammon.

彼の最初の試みはバックギャモンが得意なプレーヤーがラベル付けした例を使って

His first attempt tried to learn the utility of a Game state, U of S,

ゲームの状態の効用を学ぶことでした

using examples that were labelled by human expert backgammon players.

しかし上級のプレーヤーには面白みがないので小さい数字の状態だけをラベル付けしました

But this was tedious work for the experts, so only a small number of states were labelled.

プログラムは教師あり学習を使い一般化しようするので

The program tried to generalize from that, using supervised learning,

うまくいきませんでした

and was not able to perform very well.

そこで彼は二度目の試みでプレーヤーや人間や学習を使わずに

So Tesauro's second attempt used no human expertise and no supervision.

プログラムのコピーに対戦させました

Instead, he had 1 copy of his program play against another;

ゲームの最後に勝者は正報酬を手にしました

and at the end of the game, the winner got a positive reward,

敗者は負の報酬です

and the loser, a negative.

そして彼は強化学習を利用しました

So he used reinforcement learning;

ゲームの状態すべての知識の記録を取り人間からの入力がないにも関わらず

he backed up that knowledge throughout the Game states, and he was able to arrive at a function

世界で一番のプレーヤーと

that had no input from human expert players,

同じ水準のプレイをすることができる機能までたどりつきました

but, still, was able to perform at the level of the very best players in the world.

約20万のゲーム例から学んだあとに実行できたのです

He was able to do this, after learning from examples of about 200,000 games.

数が多く見えるかもしれませんが

Now, that may seem like a lot--

実際はバックギャモンのすべての状態空間の

but it really only covers about 1 trillionth

約1兆分の1をカバーしているだけです

of the total state space of backgammon.

さて他の例です

Now, here's another example:

これはリモートコントロールのヘリコプターです

This is a remote controlled helicopter

スタンフォード大学のイング教授が強化学習を使い訓練しました

that Professor Andrew Ng at Stanford trained, using reinforcement learning;

このヘリコプターは…失礼しました

and the helicopter--oh--oh, sorry--

間違えました　この絵を逆さまにしましょう

I made a mistake--I put this picture upside down

イング教授はヘリコプターを高度な方法で飛ばせるよう訓練しました

because--really, Ng trained the helicopter

例えば逆さまでの飛行です

to be able to fly fancy maneuvers--like flying upside down.

彼はプロの操縦士の訓練データを数時間だけ見て

And he did that by looking at only a few hours

この逆さまでの飛行を成功させたのです

of training data from expert helicopter pilots

操縦士が操縦かんを次々に引き継いでヘリを操縦しそれをすべて記録します

who would take over the remote controls, pilot the helicopter--and those would all be recorded--

いい事をした時や悪い事をした時は見返りがあります

and then, you would get rewards from when it did something good, or when it did something bad;

イング教授は自動操縦のパイロットを作るため

and Ng was able to use reinforcement learrning

この訓練データから強化学習を利用しました

to build an automated helicopter pilot, just from those training examples.

そして自動操縦のパイロットも

And that automated pilot, too, can perform tricks

一握りの人間しかできないトリックを使えます

that only a handful of humans are capable of performing.

静止画は十分ですね　動画でイング教授のヘリコプターを見ましょう

But enough of this still picture--let's watch a video of Ng's helicopters in action.

スタンフォード大学自動操縦ヘリコプター

[Stanford University Autonomous Helicopter] [sound of helicopter flying] [Chaos] [Stanford University Autonomous Helicopter]

Similar videos:

WeeklyIdol with SuperJunior مترجم عربي
usir96
45.7K views 187 likes 16 dislikes 15 Sep 2012100% match

Unit 2, Topic 34, Note on Implementation

11.7K views 26 likes 1 dislikes 05 Oct 2011100% match

Unit 2, Topic 28, Optimistic Heuristics

17.7K views 26 likes 3 dislikes 05 Oct 2011100% match

Unit 1w, 4 Terminology

75.1K views 114 likes 4 dislikes 05 Oct 2011100% match

Unit 3 9c Answer

11.4K views 20 likes 9 dislikes 08 Oct 2011100% match

Unit 2, Topic 14, Uniform Cost Search

62.2K views 42 likes 24 dislikes 05 Oct 2011100% match

CEOS Chatter
Fernando Garcia
1.3K views 2 likes 1 dislikes 14 Mar 2012100% match

Unit 3 5b Question

16.8K views 3 likes 0 dislikes 08 Oct 2011100% match

Unit 3 6d Question

16.4K views 1 likes 0 dislikes 08 Oct 2011100% match

Gundarr (Ep 14): RainBro Down Show Down (subtitulado al castellano)
daniborr
29.9K views 122 likes 18 dislikes 08 Jul 2012100% match

Unit 2, Topic 8, Graph Search Answer

13.7K views 6 likes 0 dislikes 05 Oct 2011100% match

Mr. Peabody & Sherman
TheCaptionedInternet
43.6K views 74 likes 9 dislikes 03 Mar 2014100% match

Instrucciones para llorar, de Julio Cortázar
Inés Andrés
5.9K views 10 likes 0 dislikes 10 Jan 2012100% match

Unit 10 10 Passive Temporal Difference Learning.mp4

21.8K views 228 likes 8 dislikes 05 Nov 2011100% match

Unit 2, Topic 5, Tree Search Answer

17.8K views 19 likes 3 dislikes 05 Oct 2011100% match

Unit 3 8 More Complex Bayes Networks

13K views 26 likes 3 dislikes 08 Oct 2011100% match

Gundarr (Ep 25): Fish and Chumps (subtitulado al castellano)
daniborr
14.5K views 51 likes 3 dislikes 04 Oct 2012100% match

Kingitus kõige nõrgemale
Tuuli Koitjärv
21.3K views 60 likes 4 dislikes 11 Jan 2014100% match

Thomas Jefferson's Warning about Banks
zilenthunderz
16.9K views 239 likes 4 dislikes 29 Sep 2012100% match

Recomendación kuentalibros - pizarra digital y niños con discapacidad auditiva
anitamc20
933 views 1 likes 0 dislikes 21 Dec 2012100% match

حقوقية امريكية تفضح جون برينان و الامن يقوم باعتقالها
عين اليمن
13.8K views 120 likes 5 dislikes 08 May 2012100% match

(ENG/CHI SUB) 140216 SM True Thailand Facebook EXO
Baby Xiumin
783 views 10 likes 0 dislikes 16 Feb 2014100% match

Dialog im Modegeschäft
Maria Ubieto
15.2K views 37 likes 19 dislikes 15 Jun 2013100% match

Unit 2, Topic 27, A-Star Fifth Question

22.4K views 32 likes 7 dislikes 05 Oct 2011100% match

Gundarr (Ep 21): Tele Vision Quest (subtitulado al castellano)
daniborr
16.8K views 54 likes 4 dislikes 30 Aug 2012100% match

Unit 3 8b Answer

10.9K views 18 likes 10 dislikes 08 Oct 2011100% match

Unit 11 04.mp4 Localization Problem Examples.mp4

10.3K views 32 likes 1 dislikes 11 Nov 2011100% match

ABR terapie pentru copiii cu paralizie cerebrală
Vesna Gabeloska
12.2K views 27 likes 6 dislikes 03 Sep 2011100% match

Unit 2, Topic 25, A-Star Third Question

26.6K views 9 likes 17 dislikes 05 Oct 2011100% match

Homework 1, Question 7, A* Search

11.3K views 11 likes 5 dislikes 08 Oct 2011100% match

Settings Privacy About Contact Donate