ٷΰ ޴
ٷΰ
ָ޴ ٷΰ

AI Academy

Ŀŧ ֽϴ.
, ǥ
Deep Learning 1
2019-04-22 ~ 2019-04-24
ð 09:30~17:30(1 7ð, 3 21ð)
۷ι  
Ѵ
20
900,000 [VAT ]
û


Ұ
ۿ ȸͺм ͺм ۿ ϸ ϰ(gradient descent) ϱ Ѵ. α׷ ְ tensorflow, keras, pytorch α׷ . ׿ ͺм α׷ ʿ ̽ Ű numpy, pandas  ȴ.
ǥ
 ۿ ȭ Ѵ.
 ͸ ó ִ.
 α׷ ִ.
 Ͽ ֵ Ķ Ʃ ִ.
 ǹƮ ϴ
 Ȱϰ ϴ ͺм(̾ƼƮ)
 ̿ ǰ ȹ ȹ, Product Manager

1

 

(ȸͺм, regression analysis)

 

l  Ǯ - linear  regression : python ǽ

l  ()̿Ͽ Ǯ - linear algebra

l  Normal Equation vs Gradient Descent

l  vs ӽŷ׹

l  ) ޿ ī뷮 踦 1Լ(1 ȸͺм) 𵨸ϱ

l  ) ޿ ī뷮 踦 Լ( ȸͺм) 𵨸ϱ

 ϰ(Gradient Descent), н

 

l  Gradient Descent ۿ

l  ) ޿ ī뷮 踦 ϰ Ǯ - python code ǽ

l  Universal Approximator

l  ȭ, Global Optima, Local Optima,

l  Optimizer : Momentum, RMSProp, Adam, ...

l  Convex, Non-Convex

 

l 

l  н

l  loss Լ: MSE

l  optimizer

l  ) Non-linear regression (ﰢԼ) 𵨸 ϱ

2

  α׷: numpy, tensorflow, keras

 

l  ̽ Ű numpy

l  tensorflow α׷

l  keras α׷

  Ȱ1:  

Classification (з)

 

l  ) Ҹ(sound) з : Dog, Car, Child Ҹ(*.wav ) ϱtensorflow ϱ

l  keras ϱ

l  , ε

l  softmax, argmax

l  cross-entropy loss

l  feature extraction

  ǹ & Ķ Ʃ

(hyper-parameter)

 

l  н, Validation, Test

l  (over-fitting)

l  Cross Validation

l  Ķ

l  Ķ Ʃ

l  Dropout

l  Batch normalization

l  ) Ķ Ʃ

3

  ó

 

l  pandas ̿ ó

l  ¥ ó

l  matplotlib - ⺻ ðȭ

l  sns - ðȭ Ұ

  Ȱ2: Anomaly Detection ͺм

 

l  Anomaly Detection

l  keras ϱ

l  keras callback ̿

l  confusion matrix , ROC curve

) Card Fraud Detection Ǯ: kaggle.com

 

پ ӿũ Ұ

l  tensorflow, keras

l  pytorch

l  Azure Ұ

l  AWS Ұ

l  mxnet Ұ

 

п  ܱ ɿ Ḧ ȯص帳ϴ.

  1. ȯ : ȯ ߻ ȯؿ ȯ ߻Ϸκ 5 ̳ ȯ 帮ڽϴ.
  2. ȯ
    • 1) ų Ҹ ̹ ݾ ȯ 帮ڽϴ.
    • 2) Ⱓ 1 ̳ , ȯ 帮ڽϴ.
      - : ̹
      - ð 1/3 : ̹ 2/3 شϴ ݾ
      - ð 1/2 : ̹ 1/2 شϴ ݾ
      - ð 1/2 : ȯ
    • 3) Ⱓ 1 ʰϴ , ȯ 帮ڽϴ.
      - : ̹
      - : ȯ ߻ ش ȯ (Ⱓ 1 ̳ ݾ) ջ ݾ