Trends in Cognitive Sciences
OpinionNeural computations that underlie decisions about sensory stimuli
Section snippets
The likelihood ratio
Consider the following problem. You are given a single measurement from a light detector, which you must use to decide whether a dim light is on or off at a specified location. Unfortunately, the detector does not indicate with certainty whether or not light is present. Instead, it indicates a value from 0 to 9 in the presence or absence of light, with some values being more likely than others when light is present (see Box 1). How do you use the value from the detector to decide if the light
The difference between two opposing sensors is proportional to the logLR
Having discussed the usefulness of the logLR, we will now demonstrate that neurons can compute an approximation to this quantity. Consider the light-detection problem, but instead of a mechanical device you are given the output of a light-sensitive neuron (measured in spikes per second). Like the device, the neural response can vary considerably but tends to be higher when the light is present. As shown in Fig. 2a, the neuron's output in the presence and absence of light can be characterized
Equivalent decision rules
A simpler approach is to implement an equivalent decision rule that does not depend on an evaluation of the LR from a representation of the PDFs. Recall that equivalent rules can be found using quantities that are monotonically related to the LR, as we showed for the output of the light-detection device (Box 1). We next develop the idea that the brain can easily approximate such a quantity – the logLR – from the activity (measured as a rate, in spikes per second) of certain sensory neurons.
Where in the brain are decisions formed?
According to our computational framework, neurons form decisions by calculating the difference in spike rates from appropriately chosen neurons. This difference approximates the logLR, a quantity that allows sensory information, prior probabilities and reward expectation to be combined into a single decision variable. Below, we review experimental evidence that the neurons that compute this kind of decision variable are found in brain structures involved in planning for action. First, we show
Representation of psychological factors that contribute to decision formation
An advantage of forming decisions by calculating the logLR is the ability to incorporate information from numerous sources. Simple addition can be used to accumulate both sensory information and psychological factors like prior probability and anticipated value Eq. (6). This idea predicts that the action-oriented circuits thought to be involved in interpreting sensory information toward a perceptual decision also reflect psychological factors that influence decision formation.
For tasks
Outstanding questions
- •
Where are perceptual decisions formed in the brain?
- •
How does the brain learn to route the information about a particular stimulus to the response process appropriate for the given task? Does attention affect this routing?
- •
How are psychological factors such as prior probability and reward expectation incorporated into the decision process?
- •
How and when does a signal representing the accumulation of information used to reach a decision actually indicate the decision?
- •
What neural computations are
Conclusion
We have presented a framework that describes how the brain makes decisions about simple perceptual stimuli. A categorical decision arises through the evaluation of a decision variable that approximates the log of the likelihood ratio favoring one hypothesis over another. We noted that the logLR is a natural currency for combining sensory evidence obtained from multiple sources – or from multiple samples in time – with prior probability and anticipated costs and benefits. We showed that under a
Acknowledgements
The authors thank Matt Leon, Mark Mazurek, John Palmer and Fred Rieke for helpful comments on the manuscript.
References (61)
Sensing without touching: psychophysical performance based on cortical microstimulation
Neuron
(2000)- et al.
Neural correlates of visual and motor decision processes
Curr. Opin. Neurobiol.
(1998) Visuomotor neurons: ambiguity of the discharge or ‘motor’ perception?
Int. J. Psychophysiol.
(2000)- et al.
Effect of expected reward magnitude on the response of neurons in the dorsolateral prefrontal cortex of the macaque
Neuron
(1999) - et al.
The Theory of Games and Economic Behavior
(1944) - Savage, L.J. (1972) The Foundations of Statistics (2nd revised edn),...
- et al.
Signal Detection Theory and Psychophysics
(1966) - et al.
Neural computation of log likelihood in control of saccadic eye movements
Nature
(1995) - et al.
Neural control of voluntary movement initiation
Science
(1996)