I noted two typos in this article:
- On p. 9, the computational complexity of the global maximization approach to state inference as suggested in Paninski et al. (2010), J Comput Neurosci, is somewhere between O(TxM²) and O(TxM³) (not O(TxM)), just as for the standard Kalman filter (i.e., linear in time, but each time step requires an M-dimensional matrix inversion).
- On p. 25, eqn. (15) is misprinted. The correct eqn. is given in the code as used and published on github. For clarity, I have also uploaded a pdf-file on github.com/durstdby/PLRNNstsp with the correct eqn.