Error-Robust Modes of the Retinal Population Code

Abstract

Across the nervous system, certain population spiking patterns are observed far more frequently than others. A hypothesis about this structure is that these collective activity patterns function as population codewords–collective modes–carrying information distinct from that of any single cell. We investigate this phenomenon in recordings of ∼150 retinal ganglion cells, the retina’s output. We develop a novel statistical model that decomposes the population response into modes; it predicts the distribution of spiking activity in the ganglion cell population with high accuracy. We found that the modes represent localized features of the visual stimulus that are distinct from the features represented by single neurons. Modes form clusters of activity states that are readily discriminated from one another. When we repeated the same visual stimulus, we found that the same mode was robustly elicited. These results suggest that retinal ganglion cells’ collective signaling is endowed with a form of error-correcting code–a principle that may hold in brain areas beyond retina.

A Biologically Plausible Mechanism to Learn Clusters of Neural Activity

“…Previous work has demonstrated that clusters of retinal ganglion cell activity exhibit error-correctionnamely, highly reliable activation by the same visual stimulus [12]. Thus, we wanted to test whether the activity of readout neurons in the WTA circuit could similar-ity exhibit error correction.…”

“…In prior work, we described the probability distribution over neural activity patterns with complex models, such as a hidden Markov model, and used machine learning methods to fit the parameters of these models [12]. While this analysis provided insight into the nature of the population code, downstream neural circuits are almost FIG.…”

“…In fact, depending on the characteristics of neural noise, redundant population codes can be optimal for encoding information [9,10]. In this context, an appealing principle for population codes is that correlations among neurons organize neural activity patterns into a discrete set of clusters, which can each be viewed as an error-robust population codeword [11][12][13][14][15].…”

“…Radiating out from the summit in different directions in the space of neural activity patterns are a set of "ridges". These ridges constitute neural activity clusters that are well-separated from each other and that can be identified as the latent states of a hidden Markov model [12]. Neural activity clusters have two other important properties.…”

“…First, they exhibit error correction, meaning that when the same stimulus is repeated, one or few clusters are activated, despite the fact that the detailed neural activity pattern was often different on every trial. Second, clus-ters have receptive fields that are qualitatively different from those of their constituent ganglion cells, including some clusters with orientation selectivity [12].…”