The blindness of modeling and how to break through. Training models from data is just like painting or capturing pictures from the world. From different angles, you will see the different pictures. The pictures are just the approximations of the world! So do the models! There must be some blindnesses behind the approximations. In this talk, we will point out several types of blindnesses in modeling procedures and introduce the solution of breaking through them.
from the real world problem to the machine learning problem
modeling types and modeling procedures
the blindness in modeling procedures ? and how to break through the blindness ?
the blindness coming from data
the blindness coming from coordinates of data
the blindness coming from frameworks of solvers (Supervised, Unsupervised, Semi-supervised)
the blindness coming from algorithms and objective function of solvers
the blindness coming from parameters of solvers
How to control your model moving forward through the darkness ?