일 | 월 | 화 | 수 | 목 | 금 | 토 |
---|---|---|---|---|---|---|
1 | 2 | 3 | 4 | 5 | 6 | |
7 | 8 | 9 | 10 | 11 | 12 | 13 |
14 | 15 | 16 | 17 | 18 | 19 | 20 |
21 | 22 | 23 | 24 | 25 | 26 | 27 |
28 | 29 | 30 |
- meta-learning
- Artificial Intelligence
- 인공지능
- Cam
- 시계열 분석
- 설명가능한 인공지능
- 코딩 테스트
- 메타러닝
- 코딩테스트
- AI
- 설명가능한
- Interpretability
- cs231n
- Unsupervised learning
- xai
- 기계학습
- python
- 백준
- Machine Learning
- SmoothGrad
- Class activation map
- coding test
- GAN
- 딥러닝
- Explainable AI
- keras
- Score-CAM
- 머신러닝
- grad-cam
- Deep learning
- Today
- Total
목록Deep learning/Tensorflow (2)
iMTE
Restricted Boltzmann Machine back propagation을 사용해, deep한 neural network를 학습시킬 수 있다. 하지만, deep networks의 cost function은 non-convex function이므로, 어디서 최적화를 시작하는지는 모델의 성능을 크게 좌우한다. (Weights initialization) 다양한 initializer들이 제시되었다. Xavier initializer, He initializer 등이 그렇다. 이 initializer는 들어오는 neurons의 수와 (fan-in) 출력되는 neurons의 수 (fan-out)을 사용해서 normal distribution의 variance를 결정한다. 이 방법은 대부분의 neuron들..
Sparse regularization Autoencoder는 입력된 데이터를 저차원으로 encoding하고 다시 고차원으로 decoding하는 unsupervised learning이다. 만약 hidden layer가 input layer의 dimension보다 작고, activation function으로 linear function을 사용하면, 이 autoencoder는 PCA와 동일한 역할을 한다. 즉, 분산이 큰 factor를 찾아내고 그 factor로 데이터를 transformation하는 것이다. 만약, hidden layer와 input layer의 차원이 같고 activation function이 linear function이라면, hidden layer는 identity matrix와 ..