deCharms, R. C., & Zador, A. (2000). Neural representation and the cortical code. Annual Review of Neuroscience, 23, 613–647. https://doi.org/10.1146/annurev.neuro.23.1.613
Summary
Abstract
The principle function of the central nervous system is to represent and transform information and thereby mediate appropriate decisions and behaviors. The cerebral cortex is one of the primary seats of the internal representations maintained and used in perception, memory, decision making, motor control, and subjective experience, but the basic coding scheme by which this information is carried and transformed by neurons is not yet fully understood. This article defines and reviews how information is represented in the firing rates and temporal patterns of populations of cortical neurons, with a particular emphasis on how this information mediates behavior and experience. .
The article opens by explaining what it means for a neural signal to be a representation. The authors emphasize that representations have both content, or carry meaningful information, and function, in that they are somehow used to affect adaptive behaviors.
The authors outline a general procedure to determine whether a particular neural area is involved in encoding representations for a transformation of information. First, a task must be “operationally constructed” to make a direct measurement of an experience or stimulus that involves the transformation of information. Then, we must decide on a measure of neuronal activity that we hypothesize to be related to representation. Finally, we use the statistical relationship between the input to the representation, the neuronal signal, and the output (i.e., behavior or subjective report) to determine whether the representational signal carries content related to the input and whether the representational function affects the output.
Key terms
- Neural code = “a system of rules and mechanisms by which a signal carries information” (i.e., generate representations that have meaning and function).
- Representation = a neural signal that is used in the process of transforming information; neuronal representations can be defined wherever the activity of encoding neurons is measurable and a defined transformation takes place.
- Computation = a process that transforms representations.
- Decoding = also known as stimulus reconstruction; an approach to determining the stimulus that caused a neuronal response by trying to reconstruct the feature that preceded the response, rather than directly analyzing what response the stimulus leads to.
- Intentionality = the property that internal mental representations have content (e.g., of sensory percepts, memories, or plans).
Reading notes
Neural representation
- Representations are specified by both content and function: “Neuronal representation is ideally defined with respect to both the content of neuronal signals and their functional relevance to an organism in a particular behavioral context.”
- “This is not captured by any measure that reflects only the relationship of a neuronal signal to stimuli being presented, for such measures ignore how (and even if) the neuronal signal is being decoded and used.”
- Decoding is more naturally motivated than direct stimulus-response analysis: “This decoding is analogous to the task a neuron downstream might perform when ‘reading out’ the spike trains that are its inputs. This approach reflects the organism’s point of view, because it requires the experimenter to interpret the neuronal activity in the way that the organism itself is posited to interpret it.”
- Neurons and neural representations enable high-level behavioral adaptation: “The function of neurons or neural representations is not just to provide a highly correlated and information-rich mirror of the environment (Churchland et al 1994), except perhaps at the earliest stages of sensory processing, but to lead to adaptive behavioral results.”
The most faithful copy of sensory stimuli is on the sensory surface—from here on, information is only transformed or lost, it can never reflect external stimuli more accurately (Shannon 1949).
- Spike timing can indicate either representation or computation.
The cortical code
- Two views of how to choose an interval for analyzing an individual spike train:
- Rate-coding hypothesis = we can replace data from an entire spike train with a single value, the mean rate.
- Temporal-coding hypothesis = the temporal structure of the spike train carries additional information about stimulus characteristics that are not included in the mean rate.
- Two hypotheses about population coding:
- Independent coding = neurons carry signals independently; “all of the information that can be obtained from one neuron can be obtained from that neuron alone, without reference to the activities of others.”
- Coordinated coding = the relationships between neurons carry information; “the signal must be derived from the relations between multiple neurons in a population, whether this relation is spike synchrony or any other pattern.”
Linking mental and neural representations
Attempting to understand neuronal representation in the absence of cognition and behavior is akin to analyzing the mathematical patterns in a musical score without ever listening to the music—it misses entirely the reasons that particular patterns exist.
- Neural representations, like cognitive representations, have measurable intentionality in their content.
What is the purpose of a particular mental phenomenon, experience, or representation? What exactly is awareness for? What behavioral abilities does subjective experience confer that are impossible without it? What is its adaptive function, and how can it be linked with its underlying neuronal signals?