• The Main Problem: Semantic Gap(to computer, it’s just a gigantic grid(matrix) of numbers)
• Visual recognition algorithm must be robust with
  • Lighting conditions
  • Location of camera
  • Brightness
  • Deformation
  • Occlusion
  • Background
  • Intraclass variation
  • Various labels
• Data-driven approach: Train bunch of data, then predict given test data for pre-knowledge from train data
• Nearest Neighbor
• K-Nearest Neighbor