Prediction and planning
Networks and people are good at prediction
Temporal difference learning
Word segmentation
Sentence comprehension
Forward modeling
Learn a model of the world
Use model to predict the consequences of actions
Select actions based on the best predicted outcome
Can we use our ability to predict to aid in selecting the
best response?
1 3
What is cognitive control?
The models we’ve looked at are largely “recognition-based”
(map input to corresponding output)
These models ignore the human ability to control which
response we produce (if we produce one at all)
Networks can’t learn to produce different responses to the same input
Animals have trouble doing this, too
Can PDP models account for
the flexibility of human behavior?
2 4
What is cognitive control? The Stroop task
We can override prepotent responses: No effect of ink color on word
You’re hungry. There’s a sandwich on your roommate’s desk, but you reading
don’t eat it.
When the color name conflicts
We can ignore things in the environment that aren’t with the word, reaction times
are the slowest
relevant for the task at hand:
Color naming is slower than
You’re at a busy train station looking for a friend wearing a red coat, and
word reading
you only look at the faces of people who are wearing red. (Miller &
Cohen, 2001)
The networks we’ve looked at process all input equally.
We can perform multiple tasks at the same time:
You’re writing an e-mail while listening to someone on the telephone.
Networks typically perform one tasks at a time.
5 15
What is cognitive control? Automatic vs. controlled processes
Automatic: fast, don’t require attention for execution, can
What must cognitive control entail? occur involuntarily
Select appropriate perceptual information for processing (e.g. Controlled: slower, voluntary, require attention
only people wearing red)
Inhibit inappropriate responses (e.g. don’t eat that sandwich)
Maintain relevant contextual information (e.g. this friend likes Word reading automatic
cream in his tea) Color naming controlled
Most of the networks we’ve seen can’t do this. When outputs conflict,
Is cognitive control qualitatively different from other kinds of controlled process will be
knowledge or processes? slowed
Example: Stroop task
6 16
MacLeod & Dunbar (1988) PDP model of Stroop task (Cohen et al., 1990)
Is there really a dichotomy between automatic and
controlled processes? At rest (R), a change in the net
input has little effect on
Taught subjects to use color names as names for neutral-colored
activation
shapes
After modulation by task units
(C), a change in the net input
“green” has a larger impact on activation
Task information sensitizes
Initially, color naming interfered with shape naming these units to external input
With extended training on shape naming, effects reversed All units in pathway activated
equally
Speed of processing and interference depend on the degree of
No specific information about
automatization (due largely to practice) the correct response
Graded nature of effects suitable for connectionist modelling
17 19
PDP model of Stroop task (Cohen et al., 1990) PDP model of Stroop task (Cohen et al., 1990)
Separate pathways for
word reading and color
Task demand units bias
naming; Word reading
processing in favor of the
pathway is stronger
weaker pathway
More practice, more
systematic task; doesn’t These units “guide” (or
need top-down support implement) attention to
overcome the dominant
Presence of a conflicting response
color produces no
interference
Color naming requires
top-down support (control)
to override “prepotent”
response from word
pathway
18 20
PDP model of Stroop task (Cohen et al., 1990) Attention & cognitive control
In the Stroop case, the task demand units are guiding
attention to enable cognitive control
Attention is:
“… the modulatory influence that representations of one type
have on selecting which (or to what degree) representations of
other types are processed…” (Cohen et al., 2004)
Attention biases competition between representations
competing to generate response
Bias can be “bottom-up” or “top-down”
In the Stroop case, it’s top-down (instructions given by the
experimenter to color name)
21 23
PDP model of Stroop task (Cohen et al., 1990) Cognitive control and the PFC
(Cohen et al., 2004)
Prefrontal cortex (PFC)
Automaticity is a continuum of strength of processing subserves the function of the
No qualitative distinction between “controlled” vs. “automatic” “task demand” units
processes Can sustain the activation of
representations that “bias the
Same kinds of processing and representation used for flow of activity along task
word reading and color naming participate in “cognitive relevant pathways”
control” Models of PFC use recurrent
connections
“Attractor dynamics”
Units with mutually excitatory
connections can actively
maintain themselves without
external input
22 24
Cognitive control and the PFC Cognitive control and the ACC
(Cohen et al., 2004) (Cohen et al., 2004)
Lesions to parts of the PFC Attention reduces conflicts in
can produce deficits in processing
“working memory” Occurrence of conflict signals
Patients are unable to maintain need for more attentional control
task relevant information
Anterior cingulate cortex (ACC)
Patients are also easily appears to respond to conflict in
distracted during a task processing pathways and/or
response representations
Lesions to the ACC result in an
inability to detect errors and
severe difficulty with Stroop task
25 27
Cognitive control and the PFC Cognitive control and the ACC
(Cohen et al., 2004) (Cohen et al., 2004)
PFC must also be able to update task representations
New input may signal a change in task or need to be ignored
The current degree of control may be insufficient to do the task
PFC needs to increase amount of “biasing”
Patients with lesions to the PFC may perseverate
Continuing to produce a response even when inappropriate or not
relevant to the task
How does the PFC know when to alter the current
amount of control?
26 28
Cognitive control & the VTA Cognitive control & expertise
(Cohen et al., 2004)
How is a task representation Strong cognitive control (e.g. color naming in the Stroop
selected from many possible task) is effortful and errorful
representations? We can do it, but it’s not easy
Maybe via temporal difference
learning How can we reduce the cognitive demands of a difficult
Ventral Tegmental Area (VTA) task?
responds to errors in predicted Novice chess players must study the board at length before
reward (Shulz et al., 1997) and selecting a move, but experts “see” the move immediately
communicates with PFC
If a response is associated with Learn to place more of the burden on the “recognition”
reward, ensure that relevant side of things
external cues signal a task change Humans and networks are inherently good at pattern recognition
to the PFC in the future
29 31
Cognitive control: Summary Cognitive control & expertise
(Cohen et al., 2004)
Experts encode perceptual input differently than novices
Chase & Simon (1975): participants viewed a chessboard with
pieces on it for 5 seconds
Later, they were asked to recall the positions of the pieces
Experts were better at recall when the pieces were in legal
configurations
No difference between experts and novices with the pieces were
placed randomly on the board
30 32
Cognitive control & expertise Summary
“Control processes” need not be different than any other
The experts had become better at recognizing the kinds of processing and representation
relevant aspects of the input Amount of control depends on strength of the relevant task
They could attend to the important details rather than all
information available Networks (and humans) can be more flexible than simply
“input response”
Another example: face processing Can learn to produce the appropriate response given a particular
task instruction, a context, a reward, etc.
Experience guides attention, reducing the required
Can either maintain that response or switch to produce a new
degree of top-down cognitive control
one
e.g. word reading in the Stroop task required less activation from
Can use prediction of reward or outcomes to learn when to use
the task units
appropriate task representations
33 34