Bayesian binning beats approximate alternatives: 
estimating peristimulus time histograms. Poster ID W13
D. Endres, M. Oram, J. Schindelin and P. Földiák Objective: model instantaneous  firing rates from neural spike trains  as a function of time.
predictive power: 

Ours The alternatives

Fixed boundary histogram approach 
(Shimazaki & Shinomoto, Neu. Comp., 2007)



Spike density function by smoothing  spike trains with a Gaussian kernel.
predictive power: 



predictive power: 


 

 

Bayesian binning for peristimulus time  histogram (PSTH):  iterates over all  possible binnings.  Computes Bayesian expectations  Yields error bars on predictions  Provides complexity control via model  comparison,  Only cubic effort for exact inference !!  Better predictor of real neural data than  alternative approaches above.