Paper by Sun et al. 2015 Nature Neuroscience.
Understanding the functions of a brain region requires knowing the neural representations of its myriad inputs, local neurons and outputs. Primary visual cortex (V1) has long been thought to compute visual orientation from untuned thalamic inputs, but very few thalamic inputs have been measured in any mammal.
We determined the response properties of ~28,000 thalamic boutons and ~4,000 cortical neurons in layers 1–5 of awake mouse V1. Using adaptive optics that allows accurate measurement of bouton activity deep in cortex, we found that around half of the boutons in the main thalamorecipient L4 carried orientation-tuned information and that their orientation and direction biases were also dominant in the L4 neuron population, suggesting that these neurons may inherit their selectivity from tuned thalamic inputs. Cortical neurons in all layers exhibited sharper tuning than thalamic boutons and a greater diversity of preferred orientations. Our results provide data-rich constraints for refining mechanistic models of cortical computation.
Paper by Ganmor et al. 2015 eLife.
Information is carried in the brain by the joint spiking patterns of large groups of noisy, unreliable neurons. This noise limits the capacity of the neural code and determines how information can be transmitted and read-out. To accurately decode, the brain must overcome this noise and identify which patterns are semantically similar.
We use models of network encoding noise to learn a thesaurus for populations of neurons in the vertebrate retina responding to artificial and natural videos, measuring the similarity between population responses to visual stimuli based on the information they carry. This thesaurus reveals that the code is organized in clusters of synonymous activity patterns that are similar in meaning but may differ considerably in their structure. This organization is highly reminiscent of the design of engineered codes. We suggest that the brain may use this structure and show how it allows accurate decoding of novel stimuli from novel spiking patterns.
Figure 3. The population code of the retina is comprised of clusters of responses with highly similar meaning. (A) Top: Similarity matrices of the population responses of representative groups of 20 neurons to an artificial (left) and natural (right) video. Each entry in the matrix corresponds to the similarity between two population responses observed in the test data (responses shown at bottom). Matrix rows (and columns) are ordered by total spike count in the population responses. Bottom: The population responses corresponding to the entries in the matrix; black ticks represent spikes. Each column is a population activity pattern corresponding to the matrix column directly above. Blue lines mark borders between different clusters. The lack of structure in the matrices implies that population responses with similar spike counts do not carry similar meanings. (B) Same as A, only here the matrix is clustered into 120 clusters. Matrix rows (and columns) are ordered such that responses from the same cluster appear together. A clustered organization of the population code is clearly evident. (C) Same as B, but using the Hamming distance between population responses, instead of the similarity measure d. A simple measure of syntactic similarity does not reveal the underlying clustered organization of the code.
Paper by Hires et al. 2015 eLife.
Cortical spike trains often appear noisy, with the timing and number of spikes varying across repetitions of stimuli. Spiking variability can arise from internal (behavioral state, unreliable neurons, or chaotic dynamics in neural circuits) and external (uncontrolled behavior or sensory stimuli) sources. The amount of irreducible internal noise in spike trains, an important constraint on models of cortical networks, has been difficult to estimate, since behavior and brain state must be precisely controlled or tracked.
We measured the encoding of information by L4 neurons in the somatosensory cortex during active tactile sensation. Spike rates were low except for several milliseconds after touch onset. During object localization, the majority (>70%) of spikes were temporally coupled to touch onset. Whisker movements organized the remaining spikes so that they aligned with particular phases of the whisk cycle. Based on observations of whisker behavior it is possible to predict brief time windows when a neuron will fire a single or small number of spikes, as well as time periods when the spike probability is zero. Touch times could be reliably and precisely decoded by pooling activity from a handful of L4 neurons (Panzeri et al., 2014). Spike count variance after touch, measured using the FF, was close to the binomial limit, the theoretical minimum. Based on these criteria we conclude that L4 responses encode touch with millisecond timescale precision and minimal noise.
This article includes the (very interesting) discussion between the authors and the reviewers.
Review by Ince et al. 2010 Neural Networks.
In this Review, we will consider one particular mathematical analysis approach to population coding, based on information theory (Quian Quiroga & Panzeri, 2009). One advantage of this approach is that information theory quantifies stimulus discriminability based on single trials (rather than on an average across trials), and this makes it biologically relevant to characterizing population codes, because (as discussed above) the brain recognizes sensory stimuli and takes decisions on single trials.
After introducing the main concepts of information theory in the context of sensory neuroscience, we will discuss ways to reduce the limited sampling bias which plagues estimation of information measures from experimentally recorded neural populations, extending the feasibility of such analysis to larger populations. We will then discuss how to quantify the contribution of the interactions between groups of neurons to the overall information carried by the neuronal population. We will focus in particular on evaluating what is the contribution of interactions up to any given order to the total information transmitted by the population, and how this contribution scales with population size. We will validate and demonstrate this formalism by applying it to simulated data with realistic neuronal statistics, with the aim of exploring the robustness of the methods to data sampling. We will also illustrate the methodology by computing the information about whisker stimuli carried by real simultaneously recorded populations from the rat somatosensory cortex in order to demonstrate what type of neurophysiological conclusion can be reached with it.
Despite the progress with the bias correction procedures described above, when the neuronal population is large it becomes impossible to compute the information in neural responses directly because the number of possible responses r grows exponentially with the population size (this is sometimes called the curse of dimensionality). A promising approach to the information analysis of larger populations is the use of information theory coupled to decoding approaches (Quian Quiroga & Panzeri, 2009). These procedures use a stimulus-decoding procedure to predict the most likely stimulus elicited from a single trial population response, and this makes it possible to compress the population response space into the space of ‘predicted stimuli’ (Quian Quiroga & Panzeri, 2009).
Review by Kumar et al. 2010, Nat. Reviews
The brain is a highly modular structure. To exploit modularity, it is necessary that spiking activity can propagate from one module to another while preserving the information it carries. Therefore, reliable propagation is one of the key properties of a candidate neural code. Surprisingly, the conditions under which spiking activity can be propagated have received comparatively little attention in the experimental literature. By contrast, several computational studies in the last decade have addressed this issue. Using feedforward networks (FFNs) as a generic network model, they have identified two dynamical activity modes that support the propagation of either asynchronous (rate code) or synchronous (temporal code) spiking. Here, we review the dichotomy of asynchronous and synchronous propagation in FFNs, propose their integration into a single extended conceptual framework and suggest experimental strategies to test our hypothesis.
Review by Tiesinga et al. 2008, Nature Rev.
Current technologies are progressing to the point where it is possible to record the simultaneous spiking activity of hundreds of neurons, as well as to manipulate their spike timing. However, without a theoretical framework for understanding cortical information processing, such data might not be easily interpretable. A key to cortical computations is the integration of feedforward and top-down information, which occurs at the level of the single cortical neuron. In order to fully understand this process we need to determine the computational role of precise and reliable spike times.
This Review focuses on precisely emitted spike patterns and their theoretical implications, and aims to set the stage for the large-scale study of cortical information processing. We review the biophysical mechanisms that are responsible for generating spike patterns and describe methods for uncovering spike patterns in the presence of cortical background activity. Finally, we link the integration of temporally precise synaptic inputs in active dendrites to communication, using spike volleys, within and between cortical areas.
Review by Salinas and Sejnowski 2001, Nature Rev.
Recent theoretical and experimental findings have brought the study of temporally correlated neuronal activity into a new perspective. This has emerged by investigating two related questions.
How is a postsynaptic neuron affected by the presence of correlated activity in its inputs? The response of a neuron depends on the rates at which excitatory and inhibitory input spikes impinge on it, but the temporal pattern of those spikes can also modulate postsynaptic activity. When and how does this happen? This is a biophysical problem.
What is the relationship between such temporal patterning and behaviour? It is important to understand how networks generate and react to oscillatory signals, but it is just as important to determine their function. This is a systems-level problem.
Here, we review recent findings on these two issues. First, we briefly discuss some examples of the traditional interpretation in which correlation is viewed as an additional coding dimension for building internal representations. Then we discuss some common correlation patterns and review results which show that neurons can be highly sensitive to their presence. We propose that correlations could be controlled independently of firing rate and that this would serve to regulate the flow of information rather than its meaning. Finally, we discuss several experiments in which changes in correlations have been measured and reported to be independent of changes in mean firing rate. In these cases, correlations covary with expectation, attention, response latency or rivalry — all processes that affect the transit of information but not how sensory stimuli are represented. The idea that correlations can gate the flow of neural information is a recent viewpoint that could give rise to new theoretical and experimental studies