• Useful for reduce the computation and less data input making faster process for learning algorithm  • We may not be the one that handle all the features. Often another member in team give additional features. Maybe many member gather theirs to one big chunk of features
• This make it highly redundant features that often risk being the same features
• The example talks about cm and inch, which is the same parameter, length.
• Or maybe the example of the pilot in helicopter
• Example the correlaction between engagment and skill is aptitude
• If we try to visualize the graph between 2 similar parameters, we can see the linear projection betweent the 2 input(Graph shown above, linear increase).
• What we need is the line that visualize the projection and makes that as 1D input
• output is the position of each point in projection (green line)
• Later discussed being better than the memory reduce or data reduce • Reduce 3D to 2D
• Often faced about 1000-D, try to reduce to 100-D
• Ng said that thousand-D impossible to simulate in slide, so he is doing 3D
• The example show about 3D structure. Again, if we observe closely, the 3D makes some linearity increase in graph.
• If we can somehow make a 2D-projection (plane) that represent the 3D-structure
• try to project 3-D to a plane, and record the 3D position (x,y,z) of a plane. Then we can plot the data based on the 2D graph
• Convert 2-D in matrix as z1 and z2
• Include 3 process data structure
• 3D data Structure
• 2D Plane in 3D data structure
• 2D graph     • Later, make the resulting dimensional resulting to faster process of learning algorithm