“…Examples include analytical calculations of the information flow through models of neural circuits with parameters fit to anatomical and physiological data (Treves and Panzeri, 1995;Schultz and Treves, 1998;Schultz and Rolls, 1999), and parametric fitting of probabilistic models of neural spike firing (Paninski, 2004;Pillow et al, 2005Pillow et al, , 2008, of spike count distributions (Gershon et al, 1998;Clifford and Ibbotson, 2000), or of neural mass signals (Magri et al, 2009). It should come as no surprise that "there is no free lunch", and, although in principle information quantities can be computed exactly under such circumstances, the problem is moved to one of assessing model validity -an incorrect model can lead to a bias in either direction in the information computed.…”