Conclusions - Computational Explorations in Cognitive Neuroscience - page 403

Chapter 12

Conclusions

field at both the general level of the computational mod-
eling endeavor and in terms of more specific issues, and
speculate about how these might be addressed and what
the next generation of models will bring. We conclude
by highlighting the ways in which computational mod-
els have contributed toward the development of cogni-
tive neuroscience.

Contents

12.1Overview ......................411
12.2 Fundamentals ....................411
12.3 General Challenges for Computational Modeling 413
12.3.1 ModelsAreTooSimple ...........414
12.3.2 ModelsAreTooComplex ..........417
12.3.3 ModelsCanDoAnything ..........418
12.3.4 ModelsAreReductionistic .........418
12.3.5 Modeling Lacks Cumulative Research . . . 419
12.4SpecificChallenges.................419
12.4.1 AnalyticalTreatmentsofLearning .....419
12.4.2 Error Signals ................420
12.4.3 Regularities and Generalization .......420
12.4.4 Capturing Higher-Level Cognition .....421
12.5 Contributions of Computation to Cognitive Neu-
roscience ......................421
12.5.1 Models Help Us to Understand Phenomena 421
12.5.2 ModelsDealwithComplexity........422
12.5.3 ModelsAreExplicit.............423
12.5.4 ModelsAllowControl............423
12.5.5 ModelsProvideaUnifiedFramework....423
12.6ExploringonYourOwn ..............4 24

12.2 Fundamentals

We covered quite a broad range of topics in this text,
from the movements of ions at the individual neuron
level up through learning in the service of complex cog-
nitive functions such as planning. Our goal has been to
understand findings across these different levels through
an interactive, balanced approach that emphasizes con-
nections between neurobiological, cognitive, and com-
putational considerations.
In chapter 2, we saw that individual neurons act
like detectors, constantly monitoring their inputs and
responding when something matches their pattern of
weights (synaptic efficacies). The ion channels of the
neuron compute a balance of excitatory, inhibitory, and
leak currents reflected in the membrane potential. When
this potential exceeds a threshold, the neuron fires and
sends inputs to other neurons in the network. The rate
of firing can be encoded as a continuous variable, acti-
vation.
Chapter 3 showed how individual detectors, when
wired together in a network, can exhibit useful collec-
tive properties that provide the building blocks of cog-
nition.

12.1 Overview

In this chapter we revisit some of the fundamental issues
in computational cognitive neuroscience covered in pre-
vious chapters, with an eye toward integration across
chapters. Then we explore remaining challenges for the

These network properties build on the ability

411