Working memory is the ability to maintain
and manipulate information "on-line" in the absence
of external stimulus. For example, when there is a delay
between stimulus and response, an animal's behavior
relies on the active short-term memory of the sensory
stimulus across the delay time. Perceptual decision-making
is a very general process through which the brain evaluates
and discriminates sensory inputs and makes a categorical
choice of perception or action. Interestingly, neural
correlates of working memory and perceptual decision
making have been found in the same association cortical
areas, such as posterior parietal cortex and pref rontal
cortex. A major challenge is to understand the cellular
and synaptic mechanisms of such "cognitive" cortical
circuits.
Over the last eight years, we have used
biophysically realistic modeling to investigate cortical
mechanisms of working memory and decision-making. Our
models are based on new advances in cortical neurophysiology
and are sufficiently quantitative so that comparison
between the models and physiological data becomes possible.
Our work gave rise to testable hypotheses about the
cellular mechanisms of working memory, such as the critical
role of NMDA receptors at recurrent synapses and differential
functions of diverse subtypes of inhibitory neurons.
Conceptually, our work supports the idea that working
memory and decision-making can be conceptualized in
terms of time integration by attractor network dynamics.
In order to maintain the memory of a stimulus, neural
activity in the brain has to convert a pulse-like transient
input into a persistent activity that is self-sustained
for many seconds. Thus, the output activity is like
the time integral of the input. Similarly, to subserve
a decision- making process, neural activity has to integrate
the stimulus over time, so that the brain can accumulate
and weigh evidence for choice alternatives. Our work
suggests that such attractor dynamics should be implemented
by slow, not fast, synaptic or cellular mechanisms.
Slow reverberation may be a characteristic of "cognitive"
cortical microcircuits.
