Transcript Dynamic Decision Making in Complex Task Environments

Dynamic Decision Making in Complex Task
Environments:
Principles and Neural Mechanisms
Annual Workshop Introduction
August, 2008
FY07 MURI BAA06-028 Topic 15
Building Bridges between Neuroscience, Cognition,
and Human Decision Making
Objective: The general goal is to form a complete and thorough
understanding of basic human decision processes … by building a
lattice of theoretical models with bridges that span across fields ….
The main effort of this work is intended to be in the direction of new
integrative theoretical developments … using mathematical and/or
computation modeling … accompanied and supported by rigorous
empirical models tests and empirical model comparisons …. .
From BAA 06-028, Topic 15
Our MURI Grant
• Builds on past neurophysiological and theoretical
investigations of the dynamics of decision making in
humans and non-human primates.
• Extends the empirical effort by employing fMRI, EEG,
and MEG convergently to understand the distributed
brain systems involved in decision making.
• Extends both the theory and experimental investigations
to successively more complex decision making
environments as the project continues.
• Bridges to investigations concerned with decision
making processes in real-life situations (e.g. those faced
by air-traffic controllers and pilots).
Aims of the Grant
• Aim 1: Investigate dynamics of decision making in classical tasks via
– Theory and Modeling
– Primate Neurophysiology
– Human Cognitive Neuroscience
• Fundamental tenets of the research:
– Decision making occurs through a real-time dynamic process that
depends upon neural activity distributed across a wide range of
participating brain areas, each shaping the decision making process in its
own way.
– An effort to understand decision making as an optimization problem is
useful because
• They allow us to understand how closely behavioral and neural processes can
approximate optimality
• They allow us to understand how simple neural mechanisms can lead to
optimal performance.
Aim 2: Extending the theory to decision
making in continuous time and space
• Detection of targets in noisy backgrounds when
time of onset and possible location of targets is
uncertain.
– Optimality analysis, role of leaky integration, threshold
tuning, and adjustment of integration rate in achieving
or approximating optimality.
• Locating targets in a continuous space.
– How is optimization achieved and regulated in
response to different demands for speed and
precision?
– How does the neural representation of a continuous
value (e.g. location in space) evolve over time during
processing?
Aim 3: Extensions to Real-World
Situations
• Distraction, vigilance, and divided
attention.
– Extension of neurocognitive models to
address such phenomena.
– Examination of the neural basis of the Central
Bottleneck:
• Competition among neural populations
representing stimuli/responses associated with
different tasks?
Goals for this workshop
• Review progress on Aim 1
– Primate behavior and neurophysiology
• Experiment
• Optimality analysis
• Relationship between neural activity and behavior
– Human experiments and model tests
– Further cognitive neuroscience investigations
• Brainstorm on wrapping up Aim 1, and
moving forward to Aims 2 and 3.
Wald (1947) “Sequential Probability Ratio Test (SPRT)”
Lt  L( E (t )) 
Pr(H1 | E (t ))

Pr(H 0 | E (t ))
l1 (e1 )l1 (e2 )  l1 (et )
l0 (e1 )l0 (e2 )  l0 (et )
T  inf{t  1 : Lt  ( A, B) where0  A  1  B}
Multiple hypotheses setting
•
Armitage (1950): N(N-1)/2 pair-wise
likelihood ratio processes
•
Baum and Veeravalli (1994):
Bayesian analysis on posterior
probability of N hypotheses;
•
Dragalin et al, (1999, 2000):
asymptotic optimality of MSPRT
Change-point detection setting
•
Page (1954): CUSUM procedure
•
Shiryayev (1963): Bayesian scheme
with geometric prior
•
Roberts (1966): modifying Shiryayev to
non-Bayesian version
“Sequential Methods” in Statistics
The Drift Diffusion Model
• Continuous version of the
SPRT
• At each time step a small
random step is taken.
• Mean direction of steps is
+m for one direction, –m
for the other.
• When criterion is reached,
respond.
• Alternatively, in ‘time
controlled’ tasks, respond
when signal is given.
Two Problems with the DDM
• The model predicts correct
and incorrect RT’s will
have the same
distribution, but incorrect
RT’s are generally slower
than correct RT’s.
Hard
Errors
RT
• Accuracy should gradually
improve toward ceiling
levels, even for very hard
discriminations, but this is
not what is observed in
human data.
Prob. Correct
Easy
Correct
Responses
Hard -> Easy
Usher and McClelland (2001)
Leaky Competing Accumulator Model
• Addresses the process of deciding
between two alternatives based
on external input (r1 + r2 = 1)
with leakage, self-excitation,
mutual inhibition, and noise:
dy1/dt = r1-l(y1)+af(y1)–bf(y2)+x1
dy2/dt = r2-l(y2)+af(y2)–bf(y1)+x2
Wong & Wang (2006)
~Usher & McClelland (2001)
Contributions from Princeton
• Holmes et al:
– Mathematical analysis of dynamical models of decision making.
– Investigations of optimality and deviations from optimality.
– Relations between models and levels of description
• Cohen et al:
– Neural basis of executive function and cognitive control.
– Functional brain imaging and neurally grounded models in many
areas of cognitive neuroscience.
Comparative Model Analysis
(Bogacz et al, 2006)
Physiology of Decision and Value
•
Neural basis of decision making based on
uncertain sensory information, recording
from individual neurons in primates.
•
How do neurons represent (and update
our representation of) the value of a
choice alternative?
Other Participants
• Urban lab:
– Biophysical processes that allow neurons to oscillate and
synchronize their activity
– Roles of oscillation and synchrony in information processing in
neural circuits
– Urban-McClelland collaboration:
• Use of MEG to investigate functional synchronization of neural
populations across brain areas.
– Will extend to decision making in concert with ongoing EEG
investigations.
• Johnston / Lachter:
– Processing limitations affecting throughput, accuracy, and timely
responding in human operators.
– Attentional limitations and the central bottleneck revealed in dual
task situations.
– MURI work: investigating decision dynamics using continuous
response measures.