The first step in location-based addressing is called interpolation. It is used to decide whether we should use the weights we obtained at the previous time step ,, or use the weights obtained through content-based addressing, . But how do we decide that? We use a new scalar parameter, , which is used for determining which weights we should use. The value of can be either 0 or 1.
We can express our weight vector calculation as follows:
- When the value of is 0, our equation becomes , which implies our weight vector is the one we obtained at the previous time step.
- When the value of is 1, our equation becomes , which implies our weight vector is the one we obtained through content-based addressing.
So the value of acts as a gate to switch between which weights we have to use.