Group Meetings (KR)

• Demystifying Black-box Models with Symbolic Metamodels
  • 일시: 22.07.26
  • https://papers.nips.cc/paper/2019/hash/567b8f5f423af15818a068235807edc0-Abstract.html
  • 요약
    • Complex black-box model을 interpertable하게 만들 수 있는 metamodel 형성
    • Metamodel 형성 시, Meijer G function 활용
    • Meatmodel optimization에는 gradient descent 방식 사용

• Offline RL Policies Should be Trained to be Adaptive
  • 일시: 22.08.01
  • 저자발표: https://icml.cc/virtual/2022/oral/18004
  • 요약
    • Bayesian approach를 통해 offline RL을 adaptive하게 train하는 방식 제안
    • Offline data에서 추론한 Q-function들과 belief를 통해 unseen data에 대해서 적절한 모델 적용 가능

• Patten Recognition and Machine Learning: Kernel Method
  • 일시: 22.08.22
  • https://www.microsoft.com/en-us/research/uploads/prod/2006/01/Bishop-Pattern-Recognition-and-Machine-Learning-2006.pdf
  • 요약
    • What is kernel? - 왜 feature space mapping의 내적으로 정의 되었는가?
    • What is Hilbert space?
    • What is Reproducing Kernel Hilbert Space(RKHS)? - ML에서 자주 활용되는 이유

• Dual Representation of SVM(Support Vector Machine)?
  • 일시: 22.08.29
  • https://en.wikipedia.org/wiki/Support-vector_machine
  • 요약
    • Optimization with Lagrangian multiplier
    • Constrained quadratic programming
    • Dual representaion of SVM with KKT condition