<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>Tae Hyun Kim (Lowell) — 노트</title><description>인과추론·불확실성 하 의사결정·개인화를 임상·산업 도메인에 적용하는 학습·연구 노트의 지식 정원.</description><link>https://lowellth.com/</link><language>ko-KR</language><item><title>Dunnhumby — Track 1: Latent-Factor Customer Segmentation</title><link>https://lowellth.com/ko/notes/dunnhumby-track1-segmentation/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/dunnhumby-track1-segmentation/</guid><description>NMF latent factors (92.44% explained variance) + K-Means yield 7 stable behavioral segments (Bootstrap ARI 0.77) with per-segment marketing actions. Illustrative case study on the public Dunnhumby retail dataset.</description><pubDate>Sat, 13 Jun 2026 00:00:00 GMT</pubDate></item><item><title>Dunnhumby — Track 2: Causal Targeting via Heterogeneous Treatment Effects</title><link>https://lowellth.com/ko/notes/dunnhumby-track2-causal-targeting/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/dunnhumby-track2-causal-targeting/</guid><description>Meta-learner / Causal Forest CATE under severe positivity violation (PS AUC 0.989); an OPE-validated policy targets ~31% of customers and surfaces counter-intuitive negative-CATE segments. Hypothesis-generating on public data.</description><pubDate>Sat, 13 Jun 2026 00:00:00 GMT</pubDate></item><item><title>Applied Causal Inference for Pricing — CATE &amp; SCM Across Public Datasets</title><link>https://lowellth.com/ko/notes/causal-pricing-cate-scm-framework/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/causal-pricing-cate-scm-framework/</guid><description>An applied case study using only public datasets (LendingClub, iPinYou) that combines CATE estimation for price-sensitivity heterogeneity with SCM-based moderator analysis to design individual-level, risk-based pricing and RTB bidding policies — all findings illustrative and projected, not proprietary.</description><pubDate>Fri, 12 Jun 2026 00:00:00 GMT</pubDate></item><item><title>Customer Segmentation &amp; Causal Targeting — An Applied Case Study</title><link>https://lowellth.com/ko/notes/customer-segmentation-causal-targeting-case-study/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/customer-segmentation-causal-targeting-case-study/</guid><description>An end-to-end applied case study on the public Dunnhumby dataset — NMF latent factors and K-Means segmentation feeding meta-learner / Causal Forest HTE and an OPE-validated optimal targeting policy, with a candid look at positivity violation and counter-intuitive &quot;sleeping dog&quot; segments.</description><pubDate>Fri, 12 Jun 2026 00:00:00 GMT</pubDate></item><item><title>Customer Segmentation</title><link>https://lowellth.com/ko/notes/customer-segmentation/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/customer-segmentation/</guid><description>Customer Segmentation은 고객을 행동·가치·선호의 유사성으로 유한개 세그먼트로 분할하는 비지도(unsupervised) 과제다. 흔히 잠재요인 분해 후 군집화: 행동 피처 → NMF(비음수 parts-based 분해) → factor score → K-Means → 세그먼트.</description><pubDate>Fri, 12 Jun 2026 00:00:00 GMT</pubDate></item><item><title>From Estimation to Action — How HTE Drives Personalized Policy Across Domains</title><link>https://lowellth.com/ko/notes/hte-to-policy-through-line/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/hte-to-policy-through-line/</guid><description>One methodological spine — estimate heterogeneous treatment effects and turn them into individual-level policies — powers both clinical sequential treatment decisions and industrial targeting, pricing, and recommendation.</description><pubDate>Fri, 12 Jun 2026 00:00:00 GMT</pubDate></item><item><title>LLM Multi-Layer Attribute Extraction for Cross-Domain Recommendation</title><link>https://lowellth.com/ko/notes/llm-multilayer-attributes-recsys/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/llm-multilayer-attributes-recsys/</guid><description>A case study on extracting a 3-layer attribute taxonomy (product / perceptual / theory-grounded) with LLM/VLM pipelines, turning it into user profiles and a mixture-of-experts adaptor, and plugging it into standard recommenders across two public domains (fashion + music).</description><pubDate>Fri, 12 Jun 2026 00:00:00 GMT</pubDate></item><item><title>Marketing Attribution at Scale — From Simulation to Causal Inference</title><link>https://lowellth.com/ko/notes/marketing-attribution-simulation-to-scale/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/marketing-attribution-simulation-to-scale/</guid><description>A case study comparing 10+ multi-touch attribution methods against a known-ground-truth simulator, then scaling them on the public Criteo dataset, closing the loop with budget off-policy evaluation for channel allocation.</description><pubDate>Fri, 12 Jun 2026 00:00:00 GMT</pubDate></item><item><title>Optimal Targeting Policy</title><link>https://lowellth.com/ko/notes/optimal-targeting-policy/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/optimal-targeting-policy/</guid><description>Optimal Targeting Policy는 covariate $x$를 처치 결정 $\pi(x)\in\{0,1\}$로 사상해 정책 가치를 극대화하는 규칙이다:</description><pubDate>Fri, 12 Jun 2026 00:00:00 GMT</pubDate></item><item><title>Causal Inference Under Partial Identification — Sensitivity and Evidence Hierarchies</title><link>https://lowellth.com/ko/notes/partial-id-sensitivity-bridge/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/partial-id-sensitivity-bridge/</guid><description>When real-world data fail strong ignorability, point identification gives way to bounds, proxies, and sensitivity analysis — an honest hierarchy of evidence that connects credible causal claims to semiparametric efficiency.</description><pubDate>Fri, 12 Jun 2026 00:00:00 GMT</pubDate></item><item><title>RTB Bidding Strategy via Causal ML — From Prediction to Optimization</title><link>https://lowellth.com/ko/notes/rtb-causal-bidding-strategy/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/rtb-causal-bidding-strategy/</guid><description>A five-stage case study on the public iPinYou RTB dataset that moves from pCTR/pCVR prediction through causal effect estimation (CATE, SCM) to budget-constrained optimal bidding and off-policy policy evaluation.</description><pubDate>Fri, 12 Jun 2026 00:00:00 GMT</pubDate></item><item><title>Sequential and Adaptive Decision-Making — From Bandits to Dynamic Treatment Regimes</title><link>https://lowellth.com/ko/notes/sequential-adaptive-decisions/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/sequential-adaptive-decisions/</guid><description>A synthesis essay tracing one methodological spine through sequential decision-making under uncertainty — exploration–exploitation in bandits, off-policy evaluation, and optimal/dynamic treatment regimes — that powers clinical adaptive trials and real-time bidding alike.</description><pubDate>Fri, 12 Jun 2026 00:00:00 GMT</pubDate></item><item><title>Targeting &amp; Profiling Overview</title><link>https://lowellth.com/ko/notes/targeting-overview/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/targeting-overview/</guid><description>Targeting &amp; Profiling = personalization의 산업 얼굴. 같은 방법론 코어(이질적 효과 추정 → 개인 수준 최적 정책; MOC-Personalization)가 임상에서는 &quot;환자별 최적 치료 배정&quot;으로, 산업에서는 &quot;고객별 최적 캠페인/노출 배정&quot;으로 나타난다. 이 도메인은 누구에게…</description><pubDate>Fri, 12 Jun 2026 00:00:00 GMT</pubDate></item><item><title>Uplift Modeling</title><link>https://lowellth.com/ko/notes/uplift-modeling/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/uplift-modeling/</guid><description>Uplift(증분효과)는 처치(캠페인 노출·쿠폰·추천)가 한 개인의 결과(구매·전환)에 미치는 인과적 증분이다. 이진 처치 $W\in\{0,1\}$, 결과 $Y$, covariate $X$에 대해</description><pubDate>Fri, 12 Jun 2026 00:00:00 GMT</pubDate></item><item><title>User Profiling</title><link>https://lowellth.com/ko/notes/user-profiling/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/user-profiling/</guid><description>User Profiling은 고객의 행동 이력으로부터 개인 선호 프로파일(취향·맥락·잠재 패턴)을 추론해 벡터로 표현하는 과제다. 타겟팅·세그멘테이션·추천의 공통 입력층 — 임상에서 환자 covariate/multimodal 표현(Multimodal Clinical Data)에 대응하는 산업 측 표현.</description><pubDate>Fri, 12 Jun 2026 00:00:00 GMT</pubDate></item><item><title>Anytime-Valid OPE</title><link>https://lowellth.com/ko/notes/anytime-valid-ope/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/anytime-valid-ope/</guid><description>임의의 정지 시점에서도 유효한(time-uniform) off-policy value 신뢰열을 제공하는 anytime-valid off-policy evaluation; e-process/confidence sequence 기반.</description><pubDate>Thu, 11 Jun 2026 00:00:00 GMT</pubDate></item><item><title>Anytime-Valid Inference Overview</title><link>https://lowellth.com/ko/notes/anytime-valid-inference-overview/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/anytime-valid-inference-overview/</guid><description>고정 표본 가설검정의 &quot;peeking&quot; 문제를 푸는 game-theoretic statistics. 식별-타당성 drift를 실시간 모니터링하는 안전 추론의 수학적 기초.</description><pubDate>Thu, 11 Jun 2026 00:00:00 GMT</pubDate></item><item><title>Confidence Sequence</title><link>https://lowellth.com/ko/notes/confidence-sequence/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/confidence-sequence/</guid><description>confidence sequence(CS) $(Ct){t\ge1}$는 time-uniform 커버리지를 갖는 신뢰구간 열:</description><pubDate>Thu, 11 Jun 2026 00:00:00 GMT</pubDate></item><item><title>Decision-Making Overview</title><link>https://lowellth.com/ko/notes/decision-making-overview/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/decision-making-overview/</guid><description>불확실성하 (순차) 의사결정의 방법론 — bandit regret부터 RL, off-policy evaluation, dynamic/optimal treatment regimes까지. 임상(DTR/OTR)과 산업(targeting·bidding) personalization을 모두 받친다.</description><pubDate>Thu, 11 Jun 2026 00:00:00 GMT</pubDate></item><item><title>Dynamic Treatment Regimes (DTR / OTR)</title><link>https://lowellth.com/ko/notes/dynamic-treatment-regimes/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/dynamic-treatment-regimes/</guid><description>DTR은 누적 이력 $Ht$(공변량·이전 처치·중간결과)를 처치로 사상하는 결정규칙 열 $\{dt(Ht)\}{t=1}^T$. optimal treatment regime(OTR) 은 기대 장기결과 $E[Y^{d}]$를 최대화. 추정:</description><pubDate>Thu, 11 Jun 2026 00:00:00 GMT</pubDate></item><item><title>e-process (e-value)</title><link>https://lowellth.com/ko/notes/e-process/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/e-process/</guid><description>e-value $E$는 귀무가설 $H0$ 하 $EP[E]\le 1$ ($\forall P\in H0$)인 비음 확률변수. e-process $(Et)$는 임의의 정지시각 $\tau$에 대해 $E\tau$가 e-value인 비음 과정($E[E\tau]\le1$) — 보통 귀무 하 비음 supermartingale.…</description><pubDate>Thu, 11 Jun 2026 00:00:00 GMT</pubDate></item><item><title>Efficient Influence Function</title><link>https://lowellth.com/ko/notes/efficient-influence-function/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/efficient-influence-function/</guid><description>(준)모수 모형의 regular asymptotically linear(RAL) 추정량들 중, 분산이 가장 작은 IF가 efficient influence function(EIF)이며 그 분산은 semiparametric efficiency bound(모든 parametric submodel의 Cramér-Rao…</description><pubDate>Thu, 11 Jun 2026 00:00:00 GMT</pubDate></item><item><title>Influence Function</title><link>https://lowellth.com/ko/notes/influence-function/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/influence-function/</guid><description>함수형 모수 $\psi:\mathcal{P}\to\mathbb{R}$의 추정량 $\hat\psi$이 asymptotically linear이면 영향함수(IF) $\phi$가 존재하여</description><pubDate>Thu, 11 Jun 2026 00:00:00 GMT</pubDate></item><item><title>Multi-Armed Bandits</title><link>https://lowellth.com/ko/notes/multi-armed-bandits/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/multi-armed-bandits/</guid><description>$K$개 arm, 매 라운드 $t$에 $At$를 당겨 보상 관측. cumulative regret 최소화:</description><pubDate>Thu, 11 Jun 2026 00:00:00 GMT</pubDate></item><item><title>Negative Control Outcome (NCO)</title><link>https://lowellth.com/ko/notes/negative-control-outcome/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/negative-control-outcome/</guid><description>NCO는 처치의 인과적 영향을 받지 않는다고 사전 보장되지만 같은 교란 $U$의 그림자를 받는 결과변수. 대조적으로 NCE(negative control exposure)는 결과에 인과효과가 없는 노출. NCO에 대한 &quot;겉보기 효과&quot;가 0이 아니면 → 비관측 교란의 신호(탐지) → proximal로 보정.</description><pubDate>Thu, 11 Jun 2026 00:00:00 GMT</pubDate></item><item><title>Off-Policy Evaluation (OPE)</title><link>https://lowellth.com/ko/notes/off-policy-evaluation/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/off-policy-evaluation/</guid><description>다른 behavior policy $\pib$로 수집한 로그로 target policy $\pie$의 가치 $V(\pie)=E{\pie}[\sum r]$를 추정.</description><pubDate>Thu, 11 Jun 2026 00:00:00 GMT</pubDate></item><item><title>One-step Estimator</title><link>https://lowellth.com/ko/notes/one-step-estimator/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/one-step-estimator/</guid><description>플러그-인 $\psi(\hat P)$에 추정된 EIF의 경험평균을 더해 1차 편향을 교정:</description><pubDate>Thu, 11 Jun 2026 00:00:00 GMT</pubDate></item><item><title>Partial Identification</title><link>https://lowellth.com/ko/notes/partial-identification/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/partial-identification/</guid><description>점식별(point identification)이 가정 부족으로 불가능할 때, 모수는 데이터 + 가정과 양립하는 identified set $\ThetaI$(흔히 구간 $[\thetaL,\thetaU]$)에만 속함을 안다. Manski의 무가정/worst-case bounds가 출발점. sharp bounds =…</description><pubDate>Thu, 11 Jun 2026 00:00:00 GMT</pubDate></item><item><title>Proximal Causal Inference</title><link>https://lowellth.com/ko/notes/proximal-causal-inference/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/proximal-causal-inference/</guid><description>비관측 교란 $U$가 있을 때, 두 종류의 proxy로 인과효과를 식별:</description><pubDate>Thu, 11 Jun 2026 00:00:00 GMT</pubDate></item><item><title>TMLE (Targeted Maximum Likelihood Estimation)</title><link>https://lowellth.com/ko/notes/tmle/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/tmle/</guid><description>플러그-인 추정량을 표적 모수 방향으로 보정(targeting) 하는 절차:</description><pubDate>Thu, 11 Jun 2026 00:00:00 GMT</pubDate></item><item><title>ESCM² (Entire Space Counterfactual Multi-Task Model)</title><link>https://lowellth.com/ko/notes/escm2/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/escm2/</guid><description>ESMM의 두 가지 이론적 한계 — Inherent Estimation Bias (IEB)와 Potential Independence Priority (PIP) — 를 해결하기 위해, Inverse Propensity Score (IPS) 및 Doubly Robust Estimator 기반…</description><pubDate>Wed, 25 Mar 2026 00:00:00 GMT</pubDate></item><item><title>ESMM (Entire Space Multi-Task Model)</title><link>https://lowellth.com/ko/notes/esmm/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/esmm/</guid><description>CVR의 Sample Selection Bias와 Data Sparsity 문제를 동시에 해결하기 위해, $\text{impression} \to \text{click} \to \text{conversion}$의 순차적 사용자 행동을 활용하여 전체 impression space에서 CVR을 간접 학습하는…</description><pubDate>Wed, 25 Mar 2026 00:00:00 GMT</pubDate></item><item><title>DeepFM</title><link>https://lowellth.com/ko/notes/deepfm/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/deepfm/</guid><description>DeepFM (Guo et al., 2017)은 FM component와 Deep component를 병렬로 결합하여, low-order feature interaction(explicit)과 high-order feature interaction(implicit)을 동시에 학습하는 CTR 예측 모델이다.</description><pubDate>Fri, 13 Feb 2026 00:00:00 GMT</pubDate></item><item><title>Factorization Machine</title><link>https://lowellth.com/ko/notes/factorization-machine/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/factorization-machine/</guid><description>Factorization Machine (FM)은 Rendle (2010)이 제안한 범용 예측 모델로, 모든 feature 쌍 간의 상호작용을 latent factor vector의 내적으로 모델링한다.</description><pubDate>Fri, 13 Feb 2026 00:00:00 GMT</pubDate></item><item><title>PNN</title><link>https://lowellth.com/ko/notes/pnn/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/pnn/</guid><description>PNN (Qu et al., 2016)은 embedding layer와 DNN hidden layer 사이에 product layer를 도입하여, feature embedding 간의 interaction을 명시적으로 포착한 후 DNN으로 전달하는 CTR 예측 모델이다.</description><pubDate>Fri, 13 Feb 2026 00:00:00 GMT</pubDate></item><item><title>Wide and Deep</title><link>https://lowellth.com/ko/notes/wide-and-deep/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/wide-and-deep/</guid><description>Wide &amp; Deep (Cheng et al., 2016)은 Linear wide component(memorization)와 DNN deep component(generalization)를 결합한 CTR 예측 모델이다. Google Play 앱 추천에서 처음 적용되었다.</description><pubDate>Fri, 13 Feb 2026 00:00:00 GMT</pubDate></item><item><title>Multi-Task Learning</title><link>https://lowellth.com/ko/notes/multi-task-learning/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/multi-task-learning/</guid><description>여러 관련 태스크를 동시에 학습하여 공유 표현(shared representation)을 통해 일반화 성능을 향상시키는 학습 패러다임</description><pubDate>Thu, 29 Jan 2026 00:00:00 GMT</pubDate></item><item><title>AIPW (Augmented Inverse Probability Weighting)</title><link>https://lowellth.com/ko/notes/aipw/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/aipw/</guid><description>- $\hat{\mu}t(X)$: Outcome 모델 ($E[Y|T=t, X]$)</description><pubDate>Wed, 29 Jan 2025 00:00:00 GMT</pubDate></item><item><title>ATT (Average Treatment Effect on the Treated)</title><link>https://lowellth.com/ko/notes/att/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/att/</guid><description>실제로 처치를 받은 그룹에 대한 평균 처치 효과</description><pubDate>Tue, 28 Jan 2025 00:00:00 GMT</pubDate></item><item><title>Back-door Criterion</title><link>https://lowellth.com/ko/notes/back-door-criterion/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/back-door-criterion/</guid><description>Back-door Criterion (Pearl, 1993)은 observational data에서 causal effect를 식별하기 위한 graphical criterion. 변수 집합 $Z$가 $X \rightarrow Y$의 causal effect 식별에 충분한지 판단.</description><pubDate>Tue, 28 Jan 2025 00:00:00 GMT</pubDate></item><item><title>BART (Bayesian Additive Regression Trees)</title><link>https://lowellth.com/ko/notes/bart/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/bart/</guid><description>여러 트리의 합으로 결과를 모델링하는 Bayesian 앙상블 방법</description><pubDate>Tue, 28 Jan 2025 00:00:00 GMT</pubDate></item><item><title>CATE (Conditional Average Treatment Effect)</title><link>https://lowellth.com/ko/notes/cate/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/cate/</guid><description>Conditional Average Treatment Effect (CATE)는 covariate $X=x$가 주어졌을 때의 평균 처치 효과:</description><pubDate>Tue, 28 Jan 2025 00:00:00 GMT</pubDate></item><item><title>A/B Testing</title><link>https://lowellth.com/ko/notes/a-b-testing/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/a-b-testing/</guid><description>A/B 테스트는 무작위 대조 실험(RCT)의 온라인 응용으로, 두 가지 이상의 변형(variants)을 무작위로 사용자에게 노출시켜 인과 효과를 추정하는 방법입니다.</description><pubDate>Tue, 28 Jan 2025 00:00:00 GMT</pubDate></item><item><title>Causal Forest</title><link>https://lowellth.com/ko/notes/causal-forest/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/causal-forest/</guid><description>Causal Forest는 Athey, Tibshirani, Wager (2019)가 제안한 일반화 랜덤 포레스트(GRF)의 인과추론 응용으로, 처리 효과의 이질성을 최대화하도록 분할합니다.</description><pubDate>Tue, 28 Jan 2025 00:00:00 GMT</pubDate></item><item><title>CEVAE (Causal Effect Variational Autoencoder)</title><link>https://lowellth.com/ko/notes/cevae/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/cevae/</guid><description>VAE를 활용하여 잠재 교란변수를 추론하고 인과 효과를 추정하는 방법</description><pubDate>Tue, 28 Jan 2025 00:00:00 GMT</pubDate></item><item><title>CFR (Counterfactual Regression)</title><link>https://lowellth.com/ko/notes/cfr/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/cfr/</guid><description>IPM(Integral Probability Metric) 정규화로 균형 잡힌 표현을 학습하는 딥러닝 방법</description><pubDate>Tue, 28 Jan 2025 00:00:00 GMT</pubDate></item><item><title>Collider</title><link>https://lowellth.com/ko/notes/collider/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/collider/</guid><description>Collider는 treatment (X)와 outcome (Y) 모두로부터 영향을 받는 변수 (common effect). X → C ← Y 구조에서 C가 collider.</description><pubDate>Tue, 28 Jan 2025 00:00:00 GMT</pubDate></item><item><title>Confounder</title><link>https://lowellth.com/ko/notes/confounder/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/confounder/</guid><description>Confounder는 treatment (X)와 outcome (Y) 모두에 영향을 주는 변수 (common cause)로, X와 Y 사이에 spurious (non-causal) association을 생성.</description><pubDate>Tue, 28 Jan 2025 00:00:00 GMT</pubDate></item><item><title>Constraint-Based Methods Overview</title><link>https://lowellth.com/ko/notes/constraint-based-methods-overview/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/constraint-based-methods-overview/</guid><description>Constraint-based methods는 데이터의 conditional independence (CI) 관계를 테스트하여 인과 그래프를 복원하는 방법. Faithfulness 가정 하에서 CI 관계와 d-separation의 대응을 활용.</description><pubDate>Tue, 28 Jan 2025 00:00:00 GMT</pubDate></item><item><title>Contextual Bandits</title><link>https://lowellth.com/ko/notes/contextual-bandits/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/contextual-bandits/</guid><description>Contextual Bandits는 맥락(context)에 따라 최적의 행동(arm)이 달라지는 다중 슬롯 머신 문제입니다.</description><pubDate>Tue, 28 Jan 2025 00:00:00 GMT</pubDate></item><item><title>CUPED</title><link>https://lowellth.com/ko/notes/cuped/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/cuped/</guid><description>CUPED (Controlled-experiment Using Pre-Experiment Data)는 사전 실험 데이터를 활용하여 A/B 테스트의 분산을 줄이는 기법입니다.</description><pubDate>Tue, 28 Jan 2025 00:00:00 GMT</pubDate></item><item><title>DAG (Directed Acyclic Graph)</title><link>https://lowellth.com/ko/notes/dag/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/dag/</guid><description>DAG (Directed Acyclic Graph)는 변수 간의 causal relationship을 시각적으로 표현하는 그래프. Causal inference에서 confounding 구조를 파악하고 identification strategy를 결정하는 핵심 도구.</description><pubDate>Tue, 28 Jan 2025 00:00:00 GMT</pubDate></item><item><title>Design Effect</title><link>https://lowellth.com/ko/notes/design-effect/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/design-effect/</guid><description>설계 효과(Design Effect, DEFF)는 복잡한 표집 설계가 단순 무작위 표집에 비해 분산에 미치는 영향을 측정합니다.</description><pubDate>Tue, 28 Jan 2025 00:00:00 GMT</pubDate></item><item><title>do-operator</title><link>https://lowellth.com/ko/notes/do-operator/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/do-operator/</guid><description>do-연산자는 Pearl이 제안한 개입(intervention)을 형식화하는 방법입니다.</description><pubDate>Tue, 28 Jan 2025 00:00:00 GMT</pubDate></item><item><title>d-separation</title><link>https://lowellth.com/ko/notes/d-separation/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/d-separation/</guid><description>d-separation (directional separation)은 DAG에서 두 변수 집합이 세 번째 집합에 조건부로 독립인지 판단하는 graphical criterion.</description><pubDate>Tue, 28 Jan 2025 00:00:00 GMT</pubDate></item><item><title>Double/Debiased Machine Learning (DML)</title><link>https://lowellth.com/ko/notes/double-debiased-ml/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/double-debiased-ml/</guid><description>고차원 nuisance parameter $\eta0$ 존재 하에서 저차원 관심 parameter $\theta0$에 대한 유효한 통계적 추론을 수행하기 위한 방법론.</description><pubDate>Tue, 28 Jan 2025 00:00:00 GMT</pubDate></item><item><title>Doubly Robust Estimator</title><link>https://lowellth.com/ko/notes/doubly-robust-estimator/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/doubly-robust-estimator/</guid><description>Doubly Robust (DR) Estimator는 outcome regression과 propensity score 모델을 결합하여, 둘 중 하나만 올바르게 specified되어도 consistent한 추정량.</description><pubDate>Tue, 28 Jan 2025 00:00:00 GMT</pubDate></item><item><title>DR-Learner</title><link>https://lowellth.com/ko/notes/dr-learner/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/dr-learner/</guid><description>DR-Learner는 CATE 추정을 위한 2단계 doubly robust estimator로, Pseudo-outcome을 covariate에 대해 regression하는 방식.</description><pubDate>Tue, 28 Jan 2025 00:00:00 GMT</pubDate></item><item><title>Endogeneity</title><link>https://lowellth.com/ko/notes/endogeneity/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/endogeneity/</guid><description>내생성(Endogeneity)은 설명 변수가 오차항과 상관될 때 발생하는 문제입니다.</description><pubDate>Tue, 28 Jan 2025 00:00:00 GMT</pubDate></item><item><title>Fundamental Problem of Causal Inference</title><link>https://lowellth.com/ko/notes/fundamental-problem-of-causal-inference/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/fundamental-problem-of-causal-inference/</guid><description>동일한 개인에 대해 처치(W=1)와 대조(W=0)의 결과를 동시에 관측할 수 없는 문제</description><pubDate>Tue, 28 Jan 2025 00:00:00 GMT</pubDate></item><item><title>HTE (Heterogeneous Treatment Effects)</title><link>https://lowellth.com/ko/notes/hte/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/hte/</guid><description>처치 효과가 개인의 특성에 따라 달라지는 현상</description><pubDate>Tue, 28 Jan 2025 00:00:00 GMT</pubDate></item><item><title>Instrumental Variables</title><link>https://lowellth.com/ko/notes/instrumental-variables/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/instrumental-variables/</guid><description>도구변수(Instrumental Variables, IV)는 내생성 문제를 해결하기 위해 사용하는 외생적 변수입니다.</description><pubDate>Tue, 28 Jan 2025 00:00:00 GMT</pubDate></item><item><title>IPW (Inverse Propensity Weighting)</title><link>https://lowellth.com/ko/notes/ipw/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/ipw/</guid><description>Propensity Score의 역수를 가중치로 사용하여 처치 효과 추정</description><pubDate>Tue, 28 Jan 2025 00:00:00 GMT</pubDate></item><item><title>ITE (Individual Treatment Effect)</title><link>https://lowellth.com/ko/notes/ite/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/ite/</guid><description>개인 $i$에 대한 처치 효과</description><pubDate>Tue, 28 Jan 2025 00:00:00 GMT</pubDate></item><item><title>MDP (Markov Decision Process)</title><link>https://lowellth.com/ko/notes/mdp/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/mdp/</guid><description>마르코프 결정 과정(Markov Decision Process, MDP)은 순차적 의사결정 문제의 수학적 프레임워크입니다.</description><pubDate>Tue, 28 Jan 2025 00:00:00 GMT</pubDate></item><item><title>Mediator</title><link>https://lowellth.com/ko/notes/mediator/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/mediator/</guid><description>Mediator는 treatment (X)가 outcome (Y)에 영향을 주는 causal pathway 상에 있는 중간 변수. X → M → Y 구조에서 M이 mediator.</description><pubDate>Tue, 28 Jan 2025 00:00:00 GMT</pubDate></item><item><title>Meta-learners</title><link>https://lowellth.com/ko/notes/meta-learners/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/meta-learners/</guid><description>Meta-learners는 기존 supervised learning 방법 (base learner)을 활용하여 CATE를 추정하는 알고리즘의 총칭.</description><pubDate>Tue, 28 Jan 2025 00:00:00 GMT</pubDate></item><item><title>Policy Trees</title><link>https://lowellth.com/ko/notes/policy-trees/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/policy-trees/</guid><description>정책 트리(Policy Trees)는 Athey &amp; Wager (2021)가 제안한 해석 가능한 정책 학습 방법입니다.</description><pubDate>Tue, 28 Jan 2025 00:00:00 GMT</pubDate></item><item><title>Positivity (Overlap)</title><link>https://lowellth.com/ko/notes/positivity/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/positivity/</guid><description>모든 공변량 값에 대해 처치를 받을 확률이 0과 1 사이에 존재</description><pubDate>Tue, 28 Jan 2025 00:00:00 GMT</pubDate></item><item><title>Propensity Score Matching (PSM)</title><link>https://lowellth.com/ko/notes/propensity-score-matching/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/propensity-score-matching/</guid><description>Propensity Score가 유사한 처치군과 대조군 개인을 매칭</description><pubDate>Tue, 28 Jan 2025 00:00:00 GMT</pubDate></item><item><title>R-Learner</title><link>https://lowellth.com/ko/notes/r-learner/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/r-learner/</guid><description>R-Learner (Residualized Learner)는 Robinson Transformation을 기반으로 residualized outcome과 residualized treatment를 사용하여 CATE를 추정하는 Meta-learners.</description><pubDate>Tue, 28 Jan 2025 00:00:00 GMT</pubDate></item><item><title>Representation Learning Overview</title><link>https://lowellth.com/ko/notes/representation-learning-overview/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/representation-learning-overview/</guid><description>처치와 독립적이면서 결과 예측에 유용한 표현(representation)을 학습하는 방법</description><pubDate>Tue, 28 Jan 2025 00:00:00 GMT</pubDate></item><item><title>S-Learner</title><link>https://lowellth.com/ko/notes/s-learner/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/s-learner/</guid><description>S-Learner (Single Learner)는 treatment indicator를 feature로 포함하는 단일 모델로 response function을 추정한 후 CATE를 계산하는 Meta-learners.</description><pubDate>Tue, 28 Jan 2025 00:00:00 GMT</pubDate></item><item><title>SCM (Structural Causal Model)</title><link>https://lowellth.com/ko/notes/scm/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/scm/</guid><description>SCM (Structural Causal Model)은 변수들 간의 인과 관계를 수학적으로 표현하는 framework. Pearl의 causal inference framework의 핵심.</description><pubDate>Tue, 28 Jan 2025 00:00:00 GMT</pubDate></item><item><title>Score-Based Methods Overview</title><link>https://lowellth.com/ko/notes/score-based-methods-overview/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/score-based-methods-overview/</guid><description>Score-based methods는 그래프에 score function을 부여하고, 데이터에 가장 적합한 그래프를 탐색. Constraint-based와 달리 CI tests 없이 model fit 최적화.</description><pubDate>Tue, 28 Jan 2025 00:00:00 GMT</pubDate></item><item><title>Statistical Power</title><link>https://lowellth.com/ko/notes/statistical-power/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/statistical-power/</guid><description>통계적 검정력(Statistical Power)은 효과가 실제로 존재할 때 그것을 탐지할 확률입니다.</description><pubDate>Tue, 28 Jan 2025 00:00:00 GMT</pubDate></item><item><title>Strong Ignorability</title><link>https://lowellth.com/ko/notes/strong-ignorability/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/strong-ignorability/</guid><description>Ignorability와 Positivity를 결합한 가정</description><pubDate>Tue, 28 Jan 2025 00:00:00 GMT</pubDate></item><item><title>SUTVA (Stable Unit Treatment Value Assumption)</title><link>https://lowellth.com/ko/notes/sutva/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/sutva/</guid><description>한 단위의 잠재 결과는 다른 단위의 처치 할당에 영향받지 않으며, 각 처치 수준에 대해 단일 버전만 존재한다.</description><pubDate>Tue, 28 Jan 2025 00:00:00 GMT</pubDate></item><item><title>T-Learner</title><link>https://lowellth.com/ko/notes/t-learner/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/t-learner/</guid><description>T-Learner (Two Learner)는 treatment group과 control group에 대해 별도의 모델을 학습하여 CATE를 추정하는 Meta-learners.</description><pubDate>Tue, 28 Jan 2025 00:00:00 GMT</pubDate></item><item><title>Thompson Sampling</title><link>https://lowellth.com/ko/notes/thompson-sampling/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/thompson-sampling/</guid><description>Thompson Sampling은 탐색과 활용의 균형을 맞추는 베이지안 접근법입니다.</description><pubDate>Tue, 28 Jan 2025 00:00:00 GMT</pubDate></item><item><title>Treatment Effects Overview</title><link>https://lowellth.com/ko/notes/treatment-effects-overview/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/treatment-effects-overview/</guid><description>Potential Outcome Framework에서 추정 대상(estimand)이 되는 처치 효과(treatment effects)를 체계적으로 정리한다.</description><pubDate>Tue, 28 Jan 2025 00:00:00 GMT</pubDate></item><item><title>X-Learner</title><link>https://lowellth.com/ko/notes/x-learner/</link><guid isPermaLink="true">https://lowellth.com/ko/notes/x-learner/</guid><description>X-Learner는 imputed treatment effect를 활용한 3단계 알고리즘으로, 그룹 간 불균형과 CATE의 구조적 특성을 효과적으로 활용하는 Meta-learners.</description><pubDate>Tue, 28 Jan 2025 00:00:00 GMT</pubDate></item></channel></rss>