The input rider allows to keep the input signal (the signal that is seen by the compressor) at a relatively constant level.
The parameters computed by smart:comp 2 are based on the signal "seen" during the learning pahse. Therefore, the computed compression parameters will work very well as long as the level of you signal does deviate too much from this level. By activaing the input rider, smart:comp 2 can make sure that big level differences of different signal sections are evened out.
An additional benefit of the input rider is a relatively constant level of the output signal. So while the main compressor stage of smart:comp 2 takes care of all micro-dynamics (dynamic changes over short periods of time), the input rider can take care of the macro-dynamics of your track.