Download Automatic and Voluntary Shifts of Attention in a Dynamic

A Dynamic Neural Field Model of the Hemodynamics
Associated with Response Selection and Dual-Task Performance
Aaron T. Buss, John P. Spencer, Tim Wifall, & Eliot Hazeltine
University of Iowa, Department of Psychology, Delta Center
www.delta-center.org
Response Selection & Dual-Task
The Dual-Task Model
Discussion
Response selection (RS): process of
binding stimulus and response
information allowing for meaningful, goaldirected behavior; thus, RS is a
fundamental and pervasive aspect of
everyday functioning
The model is composed of modality-specific networks of fields (Vis-Man, Aud-Voc, Vis-Voc, Aud-Man).
Sensory input fields receive stimuli on a given trial. Once activation peaks build, these fields project activation
to modality-specific 2-Dimensional SR fields. The SR fields form associations between stimuli and responses.
SR peaks build in these fields and then project activation to motor output fields which drive behavior.
The DNF model successfully captured
changes in the neural dynamics of RS
while also capturing decreases in RT and
dual-task costs over learning
As peaks of activation are built (reflecting the selection of responses), Hebbian traces accumulate for
particular SR associations. This makes peaks build more quickly on subsequent trials; this is the source of
flexible associations between stimuli and responses that can be rapidly established through task instruction.
This is the first model of RS to
quantitatively simulate both behavior and
hemodynamic responses simultaneously.
The same dynamics underlying behavior
were used to directly compute the
hemodynamics associated with behavior
on a trial-by-trial basis
Traditional view: amodal central
processor, operating over abstract
symbols binds stimuli and responses1,2
Why the dual-task paradigm? It
reveals fundamental aspects of RS
such as resource limitations by pushing
RS to the limit
Bi-stable frontal neurons modulate, prioritize, or coordinate activity between ‘tasks’; these neurons are
‘dumb’ in that they do not care about the particular task, only the modalities of the stimulus and response. The
neurons are activated when a peak is built in an associated sensory input field and are turned off when a peak
is built in an associated motor output field.
Critically, frontal networks for different modalities are mutually competitive, which leads to slower activation
of these networks with simultaneous task presentation; this is the sources of dual-task costs.
Subjects are presented with two stimuli
simultaneously Large dual-task costs
early in learning which are reduced, in
some cases eliminated, with practice
Neural basis of RS in the Inferior
Frontal Junction (IFJ): sensitive to
demands on RS; larger activation early in
learning on dual-task trials, reduces to
single task levels after learning
+
Thus, RS has distinct behavioral and
neural signatures that are revealed by
the dual-task paradigm. A deeper
understanding of RS, then, can be
achieved by a theory that is able to
bridge between brain and behavior,
accounting for both types of evidence
Dynamic Field Theory
DFT moves beyond traditional
accounts by grounding RS behaviors in
the integration of perception and action
using neural population dynamics in a
real-time, process-based model.
What is RS in DFT? It is a dynamic,
real-time binding of stimulus and
response information achieved by
activating neural populations tuned to
each dimension and associating them in
bi-modal (2D) fields.
Because DFT simulates behavior
through neural population dynamics,
we can simultaneously capture
behavioral and neural dynamics. Here,
we present the first model that is able
to quantitatively simulate behavioral
and fMRI data in a dual-task paradigm.
Different parameters for the different
modalities of input and output fields produce
differences in the rate of learning for different
task pairings: visual and manual fields have
higher resting level and a slower timescale of
activation than the auditory and vocal fields.
Thus, Hebbian traces have a smaller effect
within the visual and manual fields. These
parameter differences reflect differences in
speed of processing and peak stability within
these different modalities which could be
probed in other tasks such as WM tasks.
The parameters for the SR fields were all at
the same values
Mapping to Neural Function and Generating Hemodynamics
The inferior frontal junction (IFJ) is sensitive to dual-task conditions in a way that varies with practice: there
is larger IFJ activation early in learning; this activity decreases to single-task levels by the end of learning3.
Frontal neurons in the DNF model are potential candidates to correspond to IFJ activity: they are sensitive
to the number of tasks being executed at any given time while also mediating the coordination and execution
of these tasks by modulating the activity of SR fields
The BOLD signal is most strongly
correlated with the Local Field Potential
(LFP) which reflects the synaptic activity
over a large population of neurons4
LFPs can be estimated from DNF
models by computing the sum of the
absolute value of excitatory and inhibitory
interactions at each timestep4.
To generate a BOLD signal, we convolve
this timecourse of synaptic activity with a
general impulse response function
We ran the model in the same paradigm
as Dux et al. (2009): 8 sessions
composed of dual-task and single task
trials for each task
ACT-R: Generated both behavior and
hemodynamics; however, this is not a
real-time neural model. Rather, the
mapping from model behavior to mean
hemodynamics was indirect and
computed over separate steps5. Our
model, on the other hand, behaved in realtime, selecting responses on every trial via
emergent neural population dynamics.
Leabra: Simulated both behavioral and
neural dynamics; however, fits to both
data sets were qualitative in nature6
DFT provides a rich theoretical framework
that we are currently extending to the full set
of modality pairings as well as other
response selection paradigms (e.g., PRP).
Further, the model is generative by making
both behavioral and neural predictions (e.g.,
the metric details of stimuli and responses
should interact with the dynamics of RS; the
hemodynamic response associated with
dual-task conditions with Vis-Voc/Aud-Man
pairing should be more persistent over
learning)
These behavioral and neural effects
emerged from cascading neural
dynamics and not from a discrete
processing stage or from an a-modal RS
stage.
1. Anderson, J. R, Taatgen, N. A., & Byrne, M. D. (2005). Learning to achieve perfect timesharing:
architectural implications of Hazeltine, Teague, & Ivry (2002). Journal of Experimental
Psychology: HPP, 31(4), 749-761.
2. Meyer, D. E. & Kieras, D. E. (1997). A computational theory of executive cognitive processes and
multiple-task performance: Part I. Basic mechanisms. Psychological Review, 104(1), 3-65.
3. Dux, P.E., Tombu, M.N., Harrison, S., Rogers, B.P., Tong, F., & Marois, R. (2009). Training
improves multitasking performance by increasing the speed of information processing in human
prefrontal cortex. Neuron, 63, 127-138.
4. Deco, G., Rolls, E.T., & Horowitz, B. (2004). “What” and “where” in visual working memory: a
computational neurodynamical perspective for integrating fMRI and single-neuron data. Journal of
Cognitive Neuroscience, 16(4), 683-701.
5. Anderson, J. A., Qin, Y., Jung, K.-J., & Carter, C. S. (2007). Information-processing modules and
their relative modality specificity. Cognitive Psychology, 54, 185-217.
6. Herd, S.A., Banich, M.T., & O’Reilly, R.C. (2006). Neural mechanisms of cognitive control: an
integrative model of Stroop task performance and fMRI data. Journal of Cognitive Neuroscience,
18(1), 22-32.
Research supported by NSF BCS-1029082 awarded to JPS and EH .