Notice
Recent Posts
Recent Comments
Link
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 | 31 |
Tags
- pytorch 데이터셋 나누기
- 판다스 조건
- Does GNN Pretraining Help Molecular Representation?
- 시계열 라이브러리
- pandas 조건
- pandas 특정 조건 열 제거
- 선형함수 딥러닝
- pytorch dataset split
- 일부 레이어 고정
- sktime tutorial
- weight 일부 고정
- 모델 freeze
- EDA 추천 파이썬
- Skip connection
- pandas row 제거
- sktime 예제
- pandas 행 제거
- 비선형함수 딥러닝
- molecular representation
- 경우의 수 파이썬
- Prompt Tuning for Graph Neural Networks
- EDA in python
- 비선형함수
- sktime
- Graph Theory
- python 경우의 수
- pretraining
- pandas
- sktime 튜토리얼
- layer 일부 freeze
Archives
- Today
- Total
목록비선형함수 (1)
MoonNote

Linear activation function은 다음과 같은 형태이다. A = cx input을 받고 각 뉴런의 weight를 곱하고 input에 비례하는 output을 생성한다. 선형 함수는 1 or 0이 아닌 여러 output을 허용하기 때문에 step function 보다는 낫다. 하지만 다음과 같은 문제점이 있다. 1. 모델을 훈련하기 위해 backpropagation (gradient descent)을 사용하는 것이 불가능하다. function의 derivative가 일정하기 때문에 input과 input X와 관련이 없다. 따라서 backpropagation을 적용해도 input 뉴런의 어떤 weight가 더 좋은 예측에 영향을 주는지 이해할 수 없다. 2. 신경망의 모든 layer는 하나의..
Study/Machine Learning
2021. 11. 11. 12:06