Likelihood-based methods for spike train analysis

Much of our recent work has involved statistical techniques for analyzing neural spike trains given high-dimensional inputs (e.g., visual stimuli, or complex movements). Recently, we've been interested in likelihood-based methods for modeling spike trains, especially methods which allow us to model the detailed, temporally-precise spiking statistics of neurons. A major focus is on computational tractability; for example, we have emphasized models for which the loglikelihood is a concave function, which makes optimization and other computations much more feasible.

Paninski, L., Pillow, J., & Simoncelli, E. (2004). Maximum likelihood estimation of a stochastic integrate-and-fire neural encoding model. Neural Computation 16: 2533-2561.

Paninski, L. (2004). Maximum likelihood estimation of cascade point-process neural encoding models. Network: Computation in Neural Systems 15: 243-262.

Paninski, L. (2006). The most likely voltage path and large deviations approximations for integrate-and-fire neurons. Journal of Computational Neuroscience 21: 71-87.

Paninski, L., Pillow, J. and Lewi, J. (2007). Statistical models for neural encoding, decoding, and optimal stimulus design. (Invited review.)

Paninski, L., Ahmadian, Y., Ferreira, D., Koyama, S., Rahnama, K., Vidne, M., Vogelstein, J. & Wu, W. (2009). A new look at state-space models for neural data. In press, Journal of Computational Neuroscience (special issue on statistical analysis of neural data).

Ahmadian, Y., Pillow, J. & Paninski, L. (2010). Efficient Markov Chain Monte Carlo methods for decoding population spike trains. In press, Neural Computation.

Pillow, J., Ahmadian, Y. & Paninski, L. (2010). Model-based decoding, information estimation, and change-point detection in multi-neuron spike trains. In press, Neural Computation.

We are currently focused on applying these methods to a variety of physiological systems; see here for applications to simultaneous population recordings in primary motor cortex in awake behaving primates, and here for analysis of dynamic light responses in retina.

Spike-triggered averaging

"Spike-triggered" methods are quite popular in neural data analysis, due to their computational convenience and relative interpretability; there are close connections with the likelihood-based methods summarized above.

Simoncelli, Paninski, Pillow & Schwartz. (2004). Characterization of neural responses with stochastic stimuli. In The New Cognitive Neurosciences, ed. Gazzaniga, M.

Paninski, L. (2003). Convergence properties of three spike-triggered analysis techniques. Network: Computation in Neural Systems 14: 437-464. (Special issue on natural scene statistics and neural codes.) A beta version of the code described in this paper is available here.

Paninski (2006). The spike-triggered average of the integrate-and-fire cell driven by Gaussian white noise. Neural Computation 18: 2592-2616.

Liam Paninski's home