Human-Centered Tools for Coping with Imperfect Algorithms During Medical Decision-Making
첫번째 논문은 이미지 binary classfication을 정확히 할수 있게
refine-by-region
crop images
refine-by-example
pin examples from search results
refine-by-concept
sliding sliders, control weight
CAV(Concept Activation Vectors)
CBIR (content-based image retrieval)
- reduce semantic gap
- not algorithmic capabilities, increase user experience
Generating Clarifying Questions for Informative Retrieval
두번째 논문은 검색을 잘 할 수있게
5.1 Query Aspect Generation
- query reformulative data
- query specialization
5.2 Rule based model
- generate clarifying question
- data generation
- produce weak supervision data to train question generation model
- template
- query aspects, query entity type information
5.3 QLM (Question Likelihood Maximization)
question generation model
5.4 QCM (Query Clarification Maximization)
clarification utility
이 두가지를 어떻게 잇느냐가 중요
video --> (frame, image) --> 검색 --> query correlation
video 에 어떤 interaction들이 있는지
교수님 논문
