2024. 1. 17. 08:02ㆍ코딩 도구/LG Aimers
LG Aimers: AI전문가과정 4차
Module 5. 『인과추론』
ㅇ 교수 : 서울대학교 이상학 교수
ㅇ 학습목표
본 모듈은 인과성에 대해 추론하고 경험적 데이터를 사용하여 인과 관계를 결정하는 방법을 익힘으로써 데이터를 생성한 프로세스에 대해
만들어야 하는 필수 가정과 이러한 가정이 합리적인지 평가하는 방법, 마지막으로 추정되는 양을 해석하는 방법에 대해 학습하고자 합니다.
Part 3. 인과추론의 다양한 연구 방향 제시
-Generalized Identifiability
-Generalized Identifiability: Drug-Drug Interactions
Px1,x2(y) = ∑(b)Px2(y|b)Px1(b)
Y cardiovascular disease; B blood pressure; X1 taking an antihypertensive drug; and X2 the use of an anti-diabetic drug.
Goal: assess the effect of prescribing both treatments on the risk of cardiovascular diseases from individual drug experiments, either or .
-Summary for General Identifiability
The identifiability of any expression of the form
P(y | do(x), z) can be determined given any causal graph G and an arbitrary combination of observational and experimental studies.
If the query is identifiable, then its estimand can be derived in polynomial time.
-Transportability
Is it possible to compute the effect of X on Y in a target environment , using observational and experimental findings from different populations?
e.g., applying education policies of U.S. to South Korea.
-Transportability: the Spectrum
-Transportability: Formulas Depend on the Causal Story
-Summary for Transportability
▶ Non parametric transportability can be determined provided that the problem instance is encoded in selection diagrams (= G + ).
▶ When transportability is feasible, the transport formula can be derived in polynomial time.
▶ The causal calculus and the corresponding transportation algorithm are complete.
-Identification under Selection
▶ Selection bias, caused by preferential inclusion S of samples from the data, is a major obstacle to valid causal and statistical inferences;
-Selection Bias without External Information
Theorem
Q = P(y|x) is recoverable from selection biased data if and only if (S ⊥⊥ Y | X).
-Summary for Selection Bias
▶ Nonparametric recoverability of selection bias from causal and statistical settings can be determined provided that an augmented causal graph (w/ the selection mechanism S ) is available.
▶ When recoverability is feasible, the estimand can be derived in polynomial
time.
▶ The result is complete for pure recoverability, and sufficient for recoverability
with external information.
Recovering from Missing Data
-Identification under Missing Data: Example
Consider a study conducted in a school with Age (A), Gender (G) and Obesity (O).
▶ Age and Gender are fully observed (obtained from school records).
▶ Obesity however is corrupted by missing values due to some students not reporting their weight.
-Identification under Missing Data: Proxy Variable
-Identification under Missing Data: Reasons for Missingness
Missingness can be caused by random processes or can depend on other variables.
▶ Students accidentally losing their questionnaires.
▶ Teenagers rebelling and not reporting their weight.
▶ Obese students who are embarrassed of their obesity and hence reluctant to reveal their weight.
Summary for Part 3
Modern Identification
1. General Identification: combining data sets of different experimental conditions
2. Transportability: combining data sets from different sources
3. Identification under Selection S
4. Identification under Missingness RO
'코딩 도구 > LG Aimers' 카테고리의 다른 글
LG Aimers 4기 Deep Neural Networks (2) | 2024.01.18 |
---|---|
LG Aimers 4기 Phase 1 온라인 교육 후기 (0) | 2024.01.17 |
LG Aimers 4기 인과추론 수행을 위한 기본 방법론 (0) | 2024.01.16 |
LG Aimers 4기 인과성과 기본개념 (1) | 2024.01.15 |
LG Aimers 4기 그리고 Ensemble Learning (0) | 2024.01.14 |