Instance Based Learning and Others

Instance Based Learning and Others


  • This is the learning as before, where given data inputs, make a function/model that generalize, map the output are.
  • Instead IBL is just a look-up table. Store all input to database and spit out the output with corresponding input.
  • The downside, is no learning(generalization), and highly overfitting. How to fix this?




  • Now in Georgia Tech map in background, there's 3 houses with different kind of price. The red-expensive, blue-moderate, green-cheap.
  • the one on the left and the on the right can be decided what's color are based on nearest neighbor.
  • But the one in the middle as it reasonably closer from three color points, we have to take a circle rather than choosing the nearest one.
  • That is we need some sort of distance, radius that we could take, and take into account of the nearest neighbors.
  • Based on this test case, we can safely assume that the one in the middle as red-ish color.
  • This one called kNN, k nearest neighbors.