Analyzing Neural Responses to Natural Signals: Maximally Informative Dimensions

By Tatyana O. Sharpee, Nicole C. Rust, William Bialek. Neural Computation, 2002

333 citations recorded by the source corpus.

We propose a method that allows for a rigorous statistical analysis of neural responses to natural stimuli that are nongaussian and exhibit strong correlations. We have in mind a model in which neurons are selective for a small number of stimulus dimensions out of a high-dimensional stimulus space, but within this subspace the responses can be arbitrarily nonlinear. Existing analysis methods are based on correlation functions between stimuli and responses, but these methods are guaranteed to work only in the case of gaussian stimulus ensembles. As an alternative to correlation functions, we maximize the mutual information between the neural responses and projections of the stimulus onto low-dimensional subspaces. The procedure can be done iteratively by increasing the dimensionality of this subspace. Those dimensions that allow the recovery of all of the information between spikes and the full unprojected stimuli describe the relevant subspace. If the dimensionality of the relevant subspace indeed is small, it becomes feasible to map the neuron's input-output function even under fully natural stimulus conditions. These ideas are illustrated in simulations on model visual and auditory neurons responding to natural scenes and sounds, respectively.

Subfields: Cognitive Neuroscience, Signal Processing

View Paper: https://www.semanticscholar.org/paper/88ddb7205b70a2f53b4972366d1d8634712c064e

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Cognitive Neuroscience

In Signal Processing

Citations in the corpus

Nonlinear multivariate analysis of neurophysiological signals. (2005, result): “In this latter case, interestingly, the results obtained have led several authors to the conclusion that nonlinear encoding models are superior to traditional linear ones (Borst and Theunissen, 1999; Buracas et al., 1998; de Ruyter van Steveninck et al., 1997; Sharpee et al., 2004).”
Visual adaptation: neural, psychological and computational aspects. (2007, methodology): “As well as an appreciation of the statistics of natural images, a complete model of sensory adaptation will require detailed consideration of the uses to which the encoded sensory information is to be put.”
Highly Selective Receptive Fields in Mouse Visual Cortex (2008, methodology): “In this case, we corrected the STA to account for the non-white Gaussian spectrum of the stimuli, via normalization by the power spectrum of the stimulus set (Theunissen et al., 2001; Sharpee et al., 2004).”
How Close Are We to Understanding V1? (2005, result): “For example, given the results of Knierim and Van Essen (1992) using bar stimuli, or Sillito et al. (1995) using gratings, what should we reasonably expect to result from the sorts of context seen in the natural scene of Figure 3? Indeed, it is not even clear whether one can answer the question sinc”
Maximum likelihood estimation of cascade point-process neural encoding models (2004, background): “…of the STA, the bias is always dependent on the true g (ﬁgure 1), and to obtain a bias-free estimator for k in general we must consider the slower but more general class of semiparametric estimators which adapt to different g as well (Weisberg and Welsh 1994, Paninski 2003, Sharpee et al 2004).”
Neural processing of natural sounds (2014, methodology): “Fortunately, advances in regression techniques and machine learning have enabled the estimation of STRFs using natural sounds (FIG.”
Do We Know What the Early Visual System Does? (2005, methodology): “, 2004), maximally informative dimensions (Sharpee et al., 2004), and kernel regression methods such as the support vector machine (Wu and Gallant, 2004).”
Equitability, mutual information, and the maximal information coefficient (2013, background): “Although substantial difficulties with estimating mutual information on continuous data remain, such estimates have proved useful in a variety of realworld problems in neuroscience (14, 15, 25), molecular biology (16, 17, 26–28), medical imaging (29), and signal processing (13).”
Spike-triggered neural characterization. (2006, methodology): “This can be 241 done, for instance, using measures of information theory 242 (Paninski, 2003; Sharpee et al., 2003, 2004).”