KAIST XAI Tutorial 2025 | Introduction to Local Explanation Methods | Artyom Tashyan (SAIL, KAIST)
Key Takeaways
This tutorial focuses on local explanation methods for quantifying feature contributions to model predictions
Original Description
Local explanation methods clarify on what basis a model makes each individual prediction. In this talk, we focus on techniques that quantify how much each feature of an input contributes to its prediction, with particular attention to SHAP, LIME, and LRP. Specifically, we will discuss SHAP and LIME as representative model-agnostic methods, and LRP as a representative model-specific method tailored to deep learning models.
지역적 설명 방법은 모델의 개별 예측이 어떤 근거로 나왔는지를 국소적인 관점에서 설명하는 기법입니다. 본 발표에서는 개별 입력 데이터의 각 특징이 해당 예측에 얼마나 기여하는지를 정량적으로 계산하는 방법을 중심으로, SHAP, LIME, LRP 등의 기법을 다룰 예정입니다. 구체적으로, SHAP·LIME과 같은 (1) 모델 불가지론적(model-agnostic) 방법과 LRP와 같은 (2) 딥러닝 모델 특화(model-specific) 방법을 대표 사례로 소개합니다.
Watch on YouTube ↗
(saves to browser)
Sign in to unlock AI tutor explanation · ⚡30
Related Reads
📰
📰
📰
📰
Unlocking Open-Weight LLMs: A Developer's Guide to API Integration
Dev.to AI
Meet “Miss GPT”: The Polite Face of Confident Misinformation
Medium · LLM
Open-Weight LLM API Integration: A Practical Guide for Developers
Dev.to AI
A 26Million Parameter Model That Can Call Your Tools Without Waking the Cloud
Medium · Data Science
🎓
Tutor Explanation
DeepCamp AI