๐Ÿค– Computer Science/Artificial Intelligence

Introduction to Artificial Intelligence

yesolz 2024. 5. 1. 18:24
728x90

 

1. Basic

  • ์ธ๊ณต์ง€๋Šฅ > ๊ธฐ๊ณ„ํ•™์Šต > ๋”ฅ๋Ÿฌ๋‹ > ํŠธ๋žœ์Šคํฌ๋จธ
  • Machine Learning
    • Supervised Learning (Data-label)
      • classification: ํŠน์ • ํด๋ž˜์Šค ์˜ˆ์ธก
      • regression: ์ˆ˜์น˜ ์˜ˆ์ธก
    • Unsupervised Learning (unlabled data)
      • clustering - ํŠน์ง• ๊ณ ๋ คํ•ด์„œ ๋ฌถ์Œ
    • Reinforcement Learning: ๋‘ ๊ฐ€์ง€ ์„ ํƒ
      • Exploration: ์ƒˆ๋กœ์šด ํ–‰๋™
      • Exploitation: ๊ธฐ์กด ํ–‰๋™ ์ค‘ ๊ฐ€์žฅ ๋งŒ์กฑ๋„ ๋†’์€ ๊ฒƒ ์„ ํƒ

 

 

 

2. Machine Learning(๊ฐœ๋…)๊ณผ Data(์ข…๋ฅ˜, ๋ชจ์–‘)

  • Dataset
    • training data
    • validation data
    • test data
  • Target
  • Prediction
  • model parameter
  • model hyperparameter

 

  • Machine Learning Process
    1. Problem Identification
    2. Data Collection
    3. Data Pre-processing (filtering, transformation, integration)
    4. Model Selection, Hyper-parameter Configuration
    5. Model Training: Hyper-parameter tuning
    6. Evaluation
  • ์ •ํ˜•vs๋น„์ •ํ˜• ๋ฐ์ดํ„ฐ
    • ์ •ํ˜• ๋ฐ์ดํ„ฐ (structured data): ๊ตฌ์กฐ O. ex) ํ‘œ, json
    • ๋น„์ •ํ˜• data (unstructured data): ๊ตฌ์กฐ X. ex) ๋ฌธ์„œ, page, ์‚ฌ์ง„, ์˜์ƒ, ์†Œ๋ฆฌ
  • Data ํ˜•ํƒœ
    • 1์ฐจ์› ๋ฐ์ดํ„ฐ: ๋ฌธ์ž์—ด, ๋‹จ์ผ ์Œ์„ฑ
    • 2์ฐจ์› ๋ฐ์ดํ„ฐ: ๋‹ค์ฑ„๋„ ์Œ์„ฑ (Left/Right)
    • 3์ฐจ์› ๋ฐ์ดํ„ฐ: ์ด๋ฏธ์ง€ (RGB)
  • Data cleaning (์ •์ œ):
    • ๊ฒฐ์ธก์น˜ (missing data)
    • ํ‹€๋ฆฐ ๊ฐ’ (invalid data)
    • ์ด์ƒ์น˜ (outlier)
  • Data transformation
    • ๋ฒ”์ฃผํ˜•์œผ๋กœ
    • ์ผ๋ฐ˜ ์ •๊ทœํ™”(normalization): data ๋ฒ”์œ„ ๊ฐ™๊ฒŒ ๋ณ€ํ™˜ ex) 10์  ๋งŒ์ , 50์  ๋งŒ์  ํ†ต์ผ
    • z-score ์ •๊ทœํ™”: ํ‰๊ท  0, ํ‘œ์ค€ํŽธ์ฐจ 1 ๋˜๋„๋ก. z = x-ํ‰๊ท /ํ‘œ์ค€ํŽธ์ฐจ
    • log ๋ณ€ํ™˜: ๊ฐ’์ด ๋„ˆ๋ฌด ํฐ ๊ฒฝ์šฐ / ์ฆ๊ฐ€๋ฅผ ๊ณฑ์…ˆ์œผ๋กœ ์ฒ˜๋ฆฌํ•˜๋Š” ๊ฒŒ ๋” ํŽธํ•  ๋•Œ
  • EDA(Exploratory Data Analysis)
  • Data ํƒ์ƒ‰ - visualization(๊ฐ€์‹œํ™”), ํ†ต๊ณ„์  ์š”์†Œ
  • Discrete vs Continuous
  • expectation(๊ธฐ๋Œ“๊ฐ’), variance(๋ถ„์‚ฐ), standard Deviation(ํ‘œ์ค€ํŽธ์ฐจ)
  • Kurtosis(์ฒจ๋„) - outlier ์ •๋„ ํŒ๋‹จ
  • Skewness(์™œ๋„) - ๋ถ„ํฌ์˜ ๋น„๋Œ€์นญ ์ •๋„ ํŒ๋‹จ
  • Data reduction (์ถ•์†Œ)
  • Sampling - ๋ถ„์„์— ํ•„์š”ํ•œ data๋งŒ ์ทจํ•จ
  • Feature selection - ์œ ์šฉํ•˜๊ฒŒ ์‚ฌ์šฉ๋  ํŠน์ง•๋งŒ ์„ ํƒ

 

 

 

3. Naive Bayes Classifier (๋‚˜์ด๋ธŒ ๋ฒ ์ด์ฆˆ ๋ถ„๋ฅ˜๊ธฐ)

  • Machine Learning Model ํ‰๊ฐ€ ์ง€ํ‘œ (binary classification)
    • Accuracy : ์˜ฌ๋ฐ”๋ฅธ / ์ „์ฒด ๋ฐ์ดํ„ฐ
    • Recall : TruePositive / TruePosi + FalseNega
    • Precision: TruePosi /TruePosi + FlasePosi
    • F1 score: 2 * recall * precision / (recall + precision)
  • ์กฐ๊ฑด๋ถ€ ํ™•๋ฅ  (conditional probability)
  • ์กฐ๊ฑด๋ถ€ ํ™•๋ฅ ์˜ ์—ฐ์‡„๋ฒ•์น™ (chain rule of conditional probability)
  • ํ†ต๊ณ„์  ์ถ”๋ก  (์ถ”๋ก  ํ†ต๊ณ„ํ•™)
    • ๋นˆ๋„์ฃผ์˜ ์ถ”๋ก (Frequentist inference)
    • ๋ฒ ์ด์ฆˆ ์ถ”๋ก  (Bayesian inference)
  • Bayes Theorem : P(A|B)๋ฅผ ์•Œ ๋•Œ P(B|A)๋ฅผ ์•Œ ์ˆ˜ ์žˆ๋‹ค
    • P(H|E) = P(E|H) * P(H) / P(E)
  • Naive Bayes Classifier
    • ์žฅ์ : ๊ฐ„๋‹จ, ๋น ๋ฆ„. ๋…๋ฆฝ์ ์ด๋ผ ๊ฐ€์ • -> ์ดํ•ด ํ•ด์„ ์‰ฌ์›€, ๋‹ค์ค‘ ํด๋ž˜์Šค ๋ถ„๋ฅ˜ ๊ฐ€๋Šฅ
    • ๋‹จ์ : ๋…๋ฆฝ์  -> ๋น„ํ˜„์‹ค์ . ๋ฒ”์ฃผํ˜• ๋ฐ์ดํ„ฐ ์•„๋‹Œ ์—ฐ์†์  ๋ฐ์ดํ„ฐ. ํด๋ž˜์Šค ๋ถˆ๊ท ํ˜•
  • (์ž์—ฐ์–ด์ฒ˜๋ฆฌ) Tokenization
    • ๋‹จ์–ด ๊ธฐ๋ฐ˜ ํ† ํฐํ™” (word-based Tokenization) -> ๋‹จ์ : Dog vs Dogs ๋‹ค๋ฅด๊ฒŒ ์ธ์‹
    • ๋ฌธ์ž ๊ธฐ๋ฐ˜ ํ† ํฐํ™” (character-based Tokenization) -> ๋‹จ์ : ์—ฐ์‚ฐ๋Ÿ‰ ๋งŽ์•„์ง
    • ํ•˜์œ„ ๋‹จ์–ด ํ† ํฐํ™” (Subword Tokenization) -> ์ ๊ฒŒ ์“ฐ๋Š” ๊ฑด ๋ถ„ํ• ! ex) token/ization

 

 

 

4. Decision Tree, Tree search Algorithms

์ค‘์š” ๊ฐœ๋… : decision tree, state space, game tree(max-min์›๋ฆฌ), A* algorithm(goal state ๊ณ ๋ ค heuristic ์ ์šฉ, Monte-Carlo Tree Search algorithm(MCTS)

  • Decision Tree
    • Regression tree
    • Classification tree - feature๋ฝ‘์•„์„œ ๋ถ„๋ฅ˜
  • Tree Search algorithm - DFS, min-max tree, MCTS
  • State space: set of all possible configurations of a system.
  • Search problem: state, state space, starting state, goal state, actions, solution, cost function(๋ชจ๋“  ์•ก์…˜์— cost value ํ• ๋‹น -> cost-optimal solution)

 

  • Types of search
    • uninformed search (brute force): DFS, BFS, uniformed cost search
    • informed search: greedy search, A* search. graph search
    • Adversarial search: game tree, minimax algorithm, alpha-beta pruning
  • Minimax Search algorithm
    • max node: child node์˜ evaluation ๊ฐ’ ์ค‘ ๊ฐ€์žฅ ํฐ ๊ฐ’(alpha)๋กœ ์ด๋™
    • min node: child node์˜ evaluation ๊ฐ’ ์ค‘ ๊ฐ€์žฅ ์ž‘์€ ๊ฐ’(beta)๋กœ ์ด๋™
  • A* algorithm - 3๊ฐœ์˜ ํŒŒ๋ผ๋ฏธํ„ฐ
    • g: ์‹œ์ž‘~ํ˜„์žฌ๋…ธ๋“œ ์‹ค์ œ ๋น„์šฉ
    • h: ํœด๋ฆฌ์Šคํ‹ฑ ๊ฐ’. ํ˜„์žฌ~๋ชฉํ‘œ๋…ธ๋“œ ์ถ”์ • ๊ฒฝ๋กœ ๋น„์šฉ. (h(n) <= ์‹ค์ œ ๊ฒฝ๋กœ ๋น„์šฉ)
    • f: g+h. ํ˜„์žฌ ๋…ธ๋“œ์˜ ์ „์ฒด ์˜ˆ์ƒ ๋น„์šฉ -> A* ์•Œ๊ณ ๋ฆฌ์ฆ˜์€ f ๊ฐ’์ด ์ž‘์€ ๋…ธ๋“œ ์šฐ์„  ํƒ์ƒ‰
  • Monte Carlo Tree Search (MCTS): ๊ฐ€๋Šฅํ•œ ๋ชจ๋“  ํ–‰๋™์„ ํƒ์ƒ‰ํ•˜๊ณ , ๊ฐ ํ–‰๋™์˜ ์„ฑ๋Šฅ์„ ํ‰๊ฐ€ํ•˜์—ฌ ๊ฐ€์žฅ ์œ ๋งํ•œ ํ–‰๋™์„ ์„ ํƒ
    • AlphaGo Zero: MCTS + ์‹ ๊ฒฝ๋ง(๊ฐ•ํ™”ํ•™์Šต). ์ธ๊ฐ„๋ฐ์ดํ„ฐx, ๋” ๋›ฐ์–ด๋‚œ ์„ฑ๋Šฅ
    • 4๋‹จ๊ณ„ : selection -> expansion -> simulation -> backpropagation

 

 

 

5. Nearest Neighbors

์ค‘์š” ๊ฐœ๋… : Data distance, K-nearest neighbors, Collaborative filtering

  • Instance-based learning : ๋ชจ๋ธ ํ•™์Šต ์—†์ด ๊ฐ€๊นŒ์šด ๋ฐ์ดํ„ฐ ์ด์šฉํ•ด ๋ถ„๋ฅ˜/์˜ˆ์ธก cf. Model-vased learning
  • Data distance
    • Euclidean distance : ์ง์„ ๊ฑฐ๋ฆฌ
    • Manhattan distance : ์ง๊ฐ ๊ฑฐ๋ฆฌ
    • Jaccard distance : ๋‘ ์ง‘ํ•ฉ ์‚ฌ์ด ์œ ์‚ฌ๋„
    • Hamming distance : XOR ํ•ด์„œ ์–ป์€ 1์˜ ๊ฐœ์ˆ˜ (๋‹ค๋ฅธ ๊ฐœ์ˆ˜)
    • Cosine similarity : -1~1์˜ ๊ฐ’, 1์— ๊ฐ€๊นŒ์šธ์ˆ˜๋ก ์œ ์‚ฌ

  • K-NN (K-Nearest Neighbors): supervised learning - classification.
    • K๊ฐœ์˜ ์ด์›ƒ์ด ๊ฐ€์ง„ ๊ฐ’ ์ฐธ๊ณ  -> target value ์„ค์ •.
    • learning ์‹œ๊ฐ„ x, answering.
    • ๋ช‡๋ช…์˜ ์ด์›ƒ์œผ๋กœ ํŒ๋‹จํ•ด์•ผํ•˜๋Š”์ง€์˜ ๋ฌธ์ œ
  • Weighted nearest classifier (K-NN์˜ ๋ณ€ํ˜•, ๊ฐ€์ค‘์น˜ ๋ถ€์—ฌ)
    • ๊ฑฐ๋ฆฌ ๊ธฐ๋ฐ˜์œผ๋กœ weight ๋ถ€์—ฌ.
    • k๊ฐœ nearest neighbor๋“ค์˜ class ๋ณ„ weight sum์ด ๊ฐ€์žฅ ํฐ class๋กœ ๊ฒฐ์ •
  • Collaborative filtering - ์ถ”์ฒœ์‹œ์Šคํ…œ : ๋‚˜์™€ ์„ฑํ–ฅ์ด ๋น„์Šทํ•œ ์‚ฌ๋žŒ๋“ค์ด ์‚ฌ์šฉํ•œ ์•„์ดํ…œ ์ถ”์ฒœ
    • -> Cosine similarity ์‚ฌ์šฉ! - (๊ฒฝํ–ฅ์„ฑ์„ ๋น„๊ตํ•˜๊ธฐ ์œ„ํ•ด์„œ๋Š” Cosine similarity๋ฅผ ์‚ฌ์šฉํ•œ๋‹ค.)

 

 

 

6. Clustering

clustering - unsupervised learning

์ค‘์š” ๊ฐœ๋…: k-means algorithm, association analysis(apriori algorithm)

  • k-means algorithm
    1. k๊ฐœ์˜ ์ค‘์‹ฌ์  (centroid) ๋ฐฐ์น˜
    2. ๊ฐ data๋“ค ๊ฐ€๊นŒ์šด centroid์— ํ• ๋‹น (cluster ํ˜•์„ฑ)
    3. cluster ์•ˆ data ๊ธฐ๋ฐ˜์œผ๋กœ centroid ๋ณ€๊ฒฝ
    4. centroid ๋ณ€๊ฒฝ ์—†์„ ๋•Œ๊นŒ์ง€ 2, 3 ๋ฐ˜๋ณต 
  • k-means algorithm ๋ฌธ์ œ์ : 1์—์„œ ํ•œ๋ฒˆ์— k๊ฐœ centroid random ์„ ํƒ -> k-means++
  • k-means ++ : ์ดˆ๊ธฐ ์ค‘์‹ฌ ์„ค์ • ๋ฐฉ๋ฒ•!
    1. ๋ฌด์ž‘์œ„ 1๊ฐœ ์„ ํƒ - ์ฒซ๋ฒˆ์งธ centroid
    2. ๋‚˜๋จธ์ง€ data์™€ centroid ๊ฑฐ๋ฆฌ ๊ณ„์‚ฐ
    3. 2์—์„œ ๊ณ„์‚ฐํ•œ ๊ฑฐ๋ฆฌ ๋น„๋ก€ํ•˜์—ฌ ์ƒˆ๋กœ์šด ์ค‘์‹ฌ ์„ ํƒ (๋ฉ€๋ฆฌ ๋–จ์–ด์ง„ ๊ฑฐ ์„ ํƒ)
    4. k๊ฐœ ๊ณ ๋ฅผ ๋•Œ๊นŒ์ง€ 2, 3 ๋ฐ˜๋ณต
  • k ๊ฐ’ ์„ ํƒํ•˜๊ธฐ
    • Elbow ๋ฐฉ๋ฒ•: ์ œ๊ณฑ์˜ค์ฐจํ•ฉ ๊ทธ๋ž˜ํ”„ -> ๊ธฐ์šธ๊ธฐ ๊ธ‰๊ฒฉํžˆ ๊ฐ์†Œํ•˜๋Š” ์ง€์ 
    • Silhouette analysis: ์‹ค๋ฃจ์—ฃ ๊ณ„์ˆ˜์˜ ํ‰๊ท ๊ฐ’
  • Association analysis : ๋ฐ์ดํ„ฐ ์ง‘ํ•ฉ ์•„์ดํ…œ๋“ค ๊ฐ„ ๊ด€๊ณ„ - Apriori algorithm
  • Apriori algorithm
    • Frequent item ์„ ์ • (๋นˆ๋„ minimum support ์ด์ƒ)
    • Frequent item set ํ™•์žฅ (์ด์ „ ์„ ํƒ ์•„์ดํ…œ ์ง‘ํ•ฉ -> ํฐ ์•„์ดํ…œ ์ง‘ํ•ฉ)
    • ๋” ์ด์ƒ ์ƒˆ๋กœ์šด frequent item set ์ƒ์„ฑํ•  ์ˆ˜ ์—†์„ ๋•Œ๊นŒ์ง€ 2.
  • Apriori algorithm ๊ด€๋ จ ๊ฐœ๋…
    • confidence(์‹ ๋ขฐ๋„)
    • lift(ํ–ฅ์ƒ๋„): ๋‘ ์•„์ดํ…œ ์ง‘ํ•ฉ ๊ฐ„ ๊ด€๋ จ์„ฑ ์ธก์ •. 1์„ ๊ธฐ์ค€์œผ๋กœ 1๋ณด๋‹ค ํฌ๋ฉด ๊ธ์ •/ ๊ฐ™์œผ๋ฉด ์˜ํ–ฅx/ ์ž‘์œผ๋ฉด ๋ถ€์ •์˜ํ–ฅ์œผ๋กœ ํŒ๋‹จ

 

 

 

7. Perceptron

perceptron -> supervised learning

์ค‘์š” ๊ฐœ๋…: Artificial neural network model, Perceptron, Multilayer perceptron, CNN, RNN

  • ์ž์ฃผ ์“ฐ๋Š” Activation function
    • Sigmoid: ๊ธฐ์šธ๊ธฐ ์ค„์–ด๋“ฆ. output scale ์ œํ•œ
    • ReLU
    • tanh

  • Multi-layer perceptron
    • input layer - hidden layer - output layer
  • Softmax Function: ๋‹ค์ค‘ ํด๋ž˜์Šค ๋ถ„๋ฅ˜ - ์ถœ๋ ฅ ์ธต์˜ ํ™œ์„ฑํ™” ํ•จ์ˆ˜
    • ๋ชจ๋“  ํด๋ž˜์Šค์˜ ํ™•๋ฅ ์˜ ํ•ฉ์ด 1์ด ๋˜๋„๋ก ๋งŒ๋“ ๋‹ค
    • Prediction Error: ์–ผ๋งˆ๋‚˜ ์ž˜ ์˜ˆ์ธกํ–ˆ๋Š”์ง€ ์ธก์ •
  • Error backpropagation

 

  • epoch(์—ํฌํฌ): ์‹ ๊ฒฝ๋ง์ด ์ „์ฒด ํ•™์Šต ๋ฐ์ดํ„ฐ์…‹ ํ•œ๋ฒˆ ํ†ต๊ณผ
  • batch: ์ „์ฒด ๋ฐ์ดํ„ฐ์…‹์„ ์ผ์ • ํฌ๊ธฐ๋กœ ๋‚˜๋ˆ”
    • iteration: ๊ฐ ๋ฐฐ์น˜์— ๋Œ€ํ•ด ํ•™์Šต๊ณผ์ • ํ•œ๋ฒˆ ํ†ต๊ณผ
    • backpropagation์€ ๊ฐ ๋ฐฐ์น˜์— ๋Œ€ํ•ด ์ˆ˜ํ–‰
  • Gradient Decent algorithm
  • ANN(artificial neural network) learning process
    • ANN ๋‚ด์˜ weight ๊ฐฑ์‹ 
    • ๊ฐ batch ๋งˆ์นœ ์‹œ์ ์— ์ง„ํ–‰
    • ํ•˜๋‚˜์˜ epoch - batch ๊ฐœ์ˆ˜๋งŒํผ weight update
    • ํ›„ ์ƒˆ๋กœ์šด batch feedforward
  • CNN(Convolutional Neural Network)
  • RNN (Recurrent Neural Network)

 

  • Word embedding: ๋‹จ์–ด๋ฅผ vector space์˜ vector๋กœ ๋งคํ•‘
    • One hot encoding: ์ด์ง„ ๋ฒกํ„ฐ. ์–ด๋–ค ๋‹จ์–ด ์žˆ๋Š” ์œ„์น˜์—๋งŒ 1, ๋‚˜๋จธ์ง€ 0
    • Customized encoding: ์‚ฌ์šฉ์ž๊ฐ€ ์œ ์‚ฌ์„ฑ ๊ณ ๋ คํ•ด ์ง์ ‘ ๋‹จ์–ด ๋งคํ•‘
    • Word2Vec: ๋‹จ์–ด๋ฅผ ๊ณ ์ • ๊ธธ์ด ์‹ค์ˆ˜๋กœ ๋งคํ•‘

 

 

 

8. Deep Learning

์ค‘์š” ๊ฐœ๋…:์ธ๊ณต์‹ ๊ฒฝ๋ง/Deep Learning, Application - Object Detection/Image Generation, GAN(Generative Adversarial Networks), CycleGAN

 

  • Alexnet: ๋”ฅ๋Ÿฌ๋‹ ํ•ซํ•˜๊ฒŒ ๋งŒ๋“ ! GPU ์ด์šฉ, ์—ฌ๋Ÿฌ์ธต
  • Gradient Vanishing Problem (๊นŠ์€ ์ธต์œผ๋กœ ๊ฐˆ์ˆ˜๋ก ๊ทธ๋ž˜๋””์–ธํŠธ ์†Œ์‹ค, ๊ฐ€์ค‘์น˜ ์—…๋ฐ์ดํŠธ ์ž˜ X)
    • -> ํ•ด๊ฒฐ ์œ„ํ•ด, AlexNet์€ ReLU ์ด์šฉ! ReLU: max(0, x)
  • Dropout: ๋žœ๋ค์œผ๋กœ drop์‹œํ‚ด. -> training data์˜ overfitting ๋ฐฉ์ง€
  • ์‘์šฉ - Computer Vision - Object Detection: ์›๋ž˜๋Š” ์œ„์น˜, ๋ถ„๋ฅ˜ ๋‘๋‹จ๊ณ„. -> 'YOLO' ๋ชจ๋ธ์€ ํ•œ ๊ณผ์ •์œผ๋กœ ์ˆ˜ํ–‰

 

  • Generative Adversarial Network(GAN): ์ƒ์„ฑ๋ชจ๋ธ ์„ฑ๋Šฅ ์˜ฌ๋ฆผ
    • Generative Models: ์‹ค์ œ ๋ฐ์ดํ„ฐ ํ•™์Šต -> ๊ฐ€์งœ ๋ฐ์ดํ„ฐ ์ƒ์„ฑ
    • Generator ์™€ Discriminator ์กด์žฌ, Adversarial Traning(์ ๋Œ€์ ํ•™์Šต). ๋‘˜์ด ๊ฒฝ์Ÿ!
    • Backpropagation ์‹œ, Discrimator ์—…๋ฐ์ดํŠธ ์•ˆ ํ•˜๊ณ  Generator ๋งŒ ์—…๋ฐ์ดํŠธ ํ•˜๋Š” ์‹์œผ๋กœ ์ˆ˜๋ ด ์†๋„ ๋งž์ถค
  • ์‘์šฉ) Image Generation
    • Pix2Pix
  • Imate-to-Image Translation
    • dataset: {Edge, Photo}
    • Edge์ด๋ฏธ์ง€ -> Generator -> ์ƒ์„ฑ
    • L1 Loss: ์‹ค์ œ ์ด๋ฏธ์ง€์™€ ์ƒ์„ฑ ์ด๋ฏธ์ง€ Pixel ์ฐจ -> Error
  • Cycle GAN
    • paired dataset ๊ตฌ์ถ• ํ•œ๊ณ„, unpaired dataset ๋”ฐ๋กœ ๊ตฌ์ถ•ํ•˜๋Š” ๊ฒŒ ๋” ํŽธํ•จ
    • Unpaired Image-to-Image Translation
    • 2๊ฐœ์˜ generator ์ด์šฉ -> Cycle-consistency -> Adversarial Training

 

 

 

 

'์ธ๊ณต์ง€๋Šฅ์ž…๋ฌธ' ์ˆ˜์—… ํ•™์Šต ๋‚ด์šฉ์„ ์Šค์Šค๋กœ ์ •๋ฆฌํ•˜์˜€์Šต๋‹ˆ๋‹ค.

 

728x90