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
- fastapi
- 3d medical image
- clip intensity values
- model-free control
- shadowing
- thresholding
- freebooze
- pulloff
- 자료구조
- sidleup
- normalization
- domain adaptation
- REINFORCE
- Knowledge Distillation
- objective functions for machine learning
- Excel
- resample
- loss functions
- Inorder Traversal
- remove outliers
- checkitout
- MRI
- noise contrast estimation
- scowl
- Actor-Critic
- rest-api
- non parametic softmax
- straightup
- Policy Gradient
- sample rows
Archives
- Today
- Total
Let's Run Jinyeah
Entropy and Cross-Entropy 본문
Entropy는 일반적으로 불확실성을 나타내는 지표이다. 딥러닝에서는 이를 정보량으로 볼 수 있다.
Rationale
Information (𝑿=𝒙i) = -㏒ 𝑷(𝑿=𝒙i)
- The degree of information delivered by an event xi
- is low if 𝑷(𝑿=𝒙i) is close to 1
- is high if 𝑷(𝑿=𝒙i) is close to 0
- 즉, 확률이 클수록 정보량이 적다
- logarithm: -logx = x가 1에 가까울수록 작다
Expectation
E [𝑿] = ∑ 𝒙·𝑷(𝑿=𝒙)
- Summation of (특정 outcome x 그 outcome이 나올 확률)
Entropy
H [𝑿] = E [Information(𝑿=𝒙)] = - ∑ 𝑷(𝑿=𝒙) · ㏒𝑷(𝑿=𝒙)
- (특정 값에 대한 정보량)의 기댓값
- Expected value of the uncertainty of its outcomes
Cross-Entropy
H [Xt, Xp] = E [Information(𝑿=𝒙)] = - ∑ 𝑷t(𝑿t=𝒙) · ㏒𝑷p(𝑿p=𝒙)
𝑷t: true probability distribution
𝑷p: approximate probability distribution
- 정답 분포와 예측 분포의 outcome이 동일하고 정답 분포를 알 경우 활용
- minimize when 𝑷t≈𝑷p and 𝑷 is large
- 둘 다 높은 확률이면 0에 수렴하여 cross-entropy loss가 작아짐
- 둘 다 낮은 확률이면 무한대에 수렵하여 cross-entropy loss 커짐
- 하나는 높고, 하나는 낮은 확률이면(𝑷t≉𝑷p), 한 값이 크기 때문에 cross-entropy loss 커짐
'Deep Learning > Theory' 카테고리의 다른 글
Transfer Learning and Domain Adaptation (0) | 2022.08.10 |
---|---|
Normalization (0) | 2022.06.18 |
Bayes Decision Theory (0) | 2022.05.15 |
Objective function/Loss function (0) | 2022.05.10 |
Variance & Bias (0) | 2021.08.30 |
Comments