๋ณธ๋ฌธ ๋ฐ”๋กœ๊ฐ€๊ธฐ

๐“ก๐“ธ๐“ธ๐“ถ5: ๐’ฆ๐‘œ๐“‡๐‘’๐’ถ ๐’ฐ๐“ƒ๐’พ๐“‹/Artificial Intelligence(COSE361)

(17)
[์ธ๊ณต์ง€๋Šฅ] 5. Adversarial Search(์ ๋Œ€์  ํƒ์ƒ‰) ์ด ๋‹จ์›์—์„œ๋Š” competitive multiagent environment, ์ฆ‰ ์„œ๋กœ ๊ฒฝ์Ÿ์ ์ธ ๋‘ agent๊ฐ€ ์žˆ๋Š” ์ƒํ™ฉ์„ ๋‹ค๋ฃฌ๋‹ค. ์ด๋Š” game์œผ๋กœ ์ž˜ ์•Œ๋ ค์ ธ ์žˆ๋Š” adversarial search problem์„ ๋งํ•œ๋‹ค. 1. Game 1) game์€ ๋‹ค์Œ๊ณผ ๊ฐ™์€ ์š”์†Œ๋“ค๋กœ ์ด๋ฃจ์–ด์ ธ ์žˆ๋‹ค. - S0 : The initial state. ๊ฒŒ์ž„์ด ์–ด๋–ค ์ƒํ™ฉ์—์„œ ์‹œ์ž‘๋˜๋Š”์ง€ - Player(s) : ๊ฐ state์—์„œ ์–ด๋Š player๊ฐ€ ์›€์ง์ผ ์ฐจ๋ก€์ธ์ง€ ์•Œ๋ ค์คŒ - Actions(s) : ๊ฐ state์—์„œ ์ทจํ•  ์ˆ˜ ์žˆ๋Š” move๋“ค์˜ ์ง‘ํ•ฉ์„ return - Result(s, a) : ์–ด๋–ค state s์—์„œ action a๋ฅผ ํ–ˆ์„ ๋•Œ result. transition model. - TerminalTest(s) :..
[์ธ๊ณต์ง€๋Šฅ] 4. Beyond Classical Search - 2 2. Local Search In Continuous Space : ์ง€๊ธˆ๊นŒ์ง€๋Š” ํ˜„์žฌ ๋‚ด๊ฐ€ ์–ด๋–ค state์— ์žˆ๋Š”์ง€, action์„ ์ทจํ•˜๋ฉด ์–ด๋–ป๊ฒŒ ๋˜๋Š”์ง€, ์ด์‚ฐ์ , ๊ทธ๋ฆฌ๊ณ  ์ด ๊ฒŒ์ž„์˜ ๊ทœ์น™์ด ๋ญ”์ง€ ์•„๋Š” ๊ฒฝ์šฐ์˜€๋‹ค! ์ฆ‰ fully observable, deterministic, discrete, and known environment! discrete์™€ continuousํ•œ ํ™˜๊ฒฝ์˜ ๊ตฌ๋ถ„์€ ์‹œ๊ฐ„์ด ๋‹ค๋ค„์ง€๋Š” ๋ฐฉ๋ฒ•๊ณผ agent์˜ action๊ณผ percept์— ๋”ฐ๋ผ ์ด๋ฃจ์–ด์ง„๋‹ค. ์ง€๊ธˆ๊นŒ์ง€ ์„ค๋ช…ํ–ˆ๋˜ ์•Œ๊ณ ๋ฆฌ์ฆ˜์œผ๋กœ๋Š” continuous state์™€ action space๋ฅผ ๋‹ค๋ฃฐ ์ˆ˜๊ฐ€ ์—†๋‹ค. (๋‹จ, first-choice hill climbing๊ณผ simulated annealing ์ œ์™ธ) ์™œ๋ƒ๋ฉด ์—ฐ์†์ ์ธ ๊ฒฝ์šฐ branchin..
[์ธ๊ณต์ง€๋Šฅ] 4. Beyond Classical Search - 1 1. Local Search algorithm : Local Search algoritms์€ single current node์™€ ์ผ๋ฐ˜์ ์œผ๋กœ ๊ทธ ๋…ธ๋“œ์˜ neighbors node๋กœ ์›€์ง์ด๋Š” operation์„ ์‚ฌ์šฉํ•œ๋‹ค. ์ „ํ˜•์ ์œผ๋กœ search์— ์˜ํ•ด ๋”ฐ๋ผ์˜ค๋Š” path๋Š” ์œ ์ง€๋˜์ง€ ์•Š๋Š”๋‹ค. -> ๊ทธ๋ž˜์„œ little memory๋งŒ ์‚ฌ์šฉ (๋ณดํ†ต constant amount) -> ์•„์ฃผ ํฌ๊ฑฐ๋‚˜ infinite(continuous)ํ•œ state space์—์„œ ๋‚ฉ๋“ํ• ๋งŒํ•œ solution์„ ์ฐพ์„ ์ˆ˜ ์žˆ๋‹ค. (systematic algorithms์€ ์ ํ•ฉํ•˜์ง€ ์•Š๊ฑธ๋ž‘) -> ๋˜ํ•œ local search algorithm์€ optimization problem(objective funtion์— ๋”ฐ๋ผ ๊ฐ€์žฅ best state..
[์ธ๊ณต์ง€๋Šฅ] 3. Solving problems by searching - 3 5. Heuristic Search * Best-First Search : Best-first search๋Š” ์ผ๋ฐ˜์ ์ธ TreeSearch๋‚˜ GraphSearch algorithm์˜ instance์ด๋‹ค. ์ด ๋•Œ node๋Š” evaluation function์ธ f(n)์— ๊ธฐ์ดˆํ•ด expansion๋œ๋‹ค. Evaluation function์€ cost estimate์™€ ๊ฐ™์€ ๊ฒƒ์œผ๋กœ ์ดํ•ดํ•  ์ˆ˜ ์žˆ์œผ๋ฉฐ, ๊ทธ๋ž˜์„œ ๊ฐ€์žฅ ์ž‘์€ evaluation์„ ๊ฐ€์ง„ node๊ฐ€ ๊ฐ€์žฅ ๋จผ์ € expand๋œ๋‹ค. ์ฆ‰, Best-First Search๋Š” ๋Œ€๋ถ€๋ถ„์˜ TreeSearch๋‚˜ GraphSearch ์•Œ๊ณ ๋ฆฌ์ฆ˜์—์„œ ์“ฐ์ด๊ณ , node๋ฅผ ํ™•์žฅํ•ด ๋‚˜๊ฐˆ ๋•Œ ์–ด๋– ํ•œ evaluation function์„ ๊ธฐ์ค€์œผ๋กœ ํ•œ๋‹ค. ์ด ๋•Œ ์ด Evaluation f..
[์ธ๊ณต์ง€๋Šฅ] 3. Solving problems by searching - 2 4. Blind Search 1) Breadth-First Search (๋„ˆ๋น„์šฐ์„  ํƒ์ƒ‰) - FIFO(First in first out) : root node๋ฅผ ์ฒ˜์Œ์œผ๋กœ expandํ•˜๊ณ , ๋ชจ๋“  successors(์ž์† ๋…ธ๋“œ)๋ฅผ ๋‹ค์Œ์œผ๋กœ expandํ•œ ๋‹ค์Œ์— ๋˜ ๋‹ค์Œ successors...... - Performance of Breadth-First Search : complete ํ•˜์ง€๋งŒ optimal ํ•˜์ง€๋Š” ์•Š๋‹ค. (* complete : ๋ฌด์กฐ๊ฑด goal์„ ์ฐพ์„ ์ˆ˜ ์žˆ์Œ) ๋งŒ์•ฝ ๊ฐ€์žฅ ์–•์€ goal node๊ฐ€ depth d์— ์žˆ๋‹ค๋ฉด, ๋ชจ๋“  ๋” ์–•์€ node๋“ค์„ ํƒ์ƒ‰ํ•œ ๋‹ค์Œ์— ์ฐพ์„ ์ˆ˜ ์žˆ์„ ๊ฒƒ. ๊ทธ๋Ÿฌ๋‚˜ ๊ทธ ๊ฐ€์žฅ ์–•์€ goal node๊ฐ€ ํ•ญ์ƒ optimalํ•˜๋‹ค๊ณ ๋Š” ๋ณผ ์ˆ˜ ์—†์Œ. (๋‹จ, ๋ชจ๋“  cost๊ฐ€ ๊ฐ™์„ ..
[์ธ๊ณต์ง€๋Šฅ] 3. Solving problems by searching - 1 1. Problem-Solving Agents 1) Problem Formulation - Problem์€ initial state, actions, transition model, goal test, path cost๋กœ ์ •์˜๋œ๋‹ค. - Problem formulation์€ goal์ด ์ฃผ์–ด์ง„ ์ƒํ™ฉ์—์„œ, ๋ฌด์Šจ action ๊ณผ states๋ฅผ ๊ณ ๋ คํ•  ๊ฒƒ์ธ์ง€ ๊ฒฐ์ •ํ•˜๋Š” ๊ณผ์ •์„ ๋งํ•œ๋‹ค. - Problem์˜ solution์€ initial state๊ฐ€ goal state๋กœ ๊ฐ€๊ฒŒ ๋งŒ๋“œ๋Š” action๋“ค์˜ sequence์ด๋‹ค. - goal์— ๋‹ค๋‹ค๋ฅด๊ธฐ ์œ„ํ•œ sequence of actions์„ ์ฐพ์•„๋‚˜๊ฐ€๋Š” ๊ณผ์ •์ด search์ด๋‹ค. - solution์ด ์ฐพ์•„์ง€๋ฉด action์ด ์ˆ˜ํ–‰๋˜๋„๋ก ์ถ”์ฒœ๋œ๋‹ค. ์ด๊ฒƒ์ด ๋ฐ”๋กœ execution์ด..
[์ธ๊ณต์ง€๋Šฅ] 1.1 What is AI? &2.3 The Nature of Environments 1. AI(์ธ๊ณต์ง€๋Šฅ) ์ด๋ž€? : ์‚ฌ๋žŒ์˜ Intelligent๋ฅผ ๋ชจ๋ฐฉํ•˜๋Š” ๊ธฐ๊ณ„๋ฅผ ๋งŒ๋“ค์–ด๋ณด์ž! - Thinking Humanly : ์‚ฌ๋žŒ๋‹ต๊ฒŒ ์ƒ๊ฐํ•˜๋Š”๊ฒŒ ๋ญ”๋ฐ? - Thinking Rationally : ๋…ผ๋ฆฌํ•™๊ณผ ๊ด€๋ จ๋จ(3๋‹จ๋…ผ๋ฒ•) - Acting Humanly : Turing test - Acting Rationally : ํ•ฉ๋ฆฌ์ ์œผ๋กœ ํ–‰๋™ํ•˜๋Š” ๊ฒƒ๊ณผ ํ•ฉ๋ฆฌ์ ์œผ๋กœ ์ƒ๊ฐํ•˜๋Š” ๊ฒƒ์€ ๋‹ค๋ฆ„ (ํ–‰๋™ํ•œ๋‹ค๊ณ  ์ƒ๊ฐํ•˜๋Š” ๊ฑด ์•„๋‹˜) EX) ๋„ค๋น„๊ฒŒ์ด์…˜ ์ฆ‰, AI = Science(์ƒ๊ฐ) & Engineering(ํ–‰๋™) (๊ด„ํ˜ธ์•ˆ์€ task environment for an automated taxi) 2. The Nature Of Environments 1) Specifying the task environment : agent๊ฐ€ ์žˆ์„..