Current cog Lecture 6

Taxonomy II: Sources / Directing of attention

• Stimulus-driven attention: Attention drawn to salient stimulus

  • “Salience-driven”; “Bottom-up”; “Exogenous”; “Involuntary”; “Automatic”; “Feedforward-based”

• Goal-driven attention: Attention initiated by current behavioural requirements

  • “Top-down”; “Endogenous”; “Voluntary”; “Controlled”; “Feedback-based”

• Experience-driven attention: Attention driven by learned context or value

  • “Selection history”; “Reward”; “Priming”; “Automatic”

In the real world

• We often know what we want, just not where it is

• We activate some representation of the goal, or target object

• We scan around until we find a match

• Especially helpful when target object does not “pop out”

In the lab: Goal-driven visual search

• Task: Find the red vertical bar

• You probably try limit search to the red items, looking for a vertical one

• Slower, more serial than single feature search

Goal-driven attention to space: Spatial cueing

• We can deliberately direct our attention on purpose in space (independent of the eyes!)

• Task: Keep eyes on central fixation cross. Attend to location pointed out by arrow cue.

• Respond to target stimulus appearing.

• Conditions: Target appears in cued location (valid cue, most of the times, e.g. 80%), or in other location (invalid cue, rare, e.g. 20%)

• Primary measure: Manual RT

• Results: Cue validity effect: Faster response after validly cues, slower after invalid cues → Covert attention

Goal-driven attention to features: Contingent Capture

• We can deliberately look for certain features and not others, e.g. red.

• Task: Look for a red X or = sign in the final display – ignore anything else

• A cue display appears just before, one of the cues matches the color of what you’re looking for (i.e. red)

• Conditions: The cue indicates the target position (valid cue, 25% of trials), or not (invalid cues, 75%)

• Primary measure: manual RT

• Results: Cue validity effect

• Conclusion: When actively looking for a feature (here red), attention is inadvertently caught by that feature: Contingent capture

• But is this because observers are deliberately looking for red, or because red is salient?

• We need a control condition: Look for a single white onset target instead

• Same task: Determine if X or =

• Same spatial validity manipulation of red cue

• If goal-driven, then no cuing effect

• Results: No cue validity effect

• Conclusion holds!

Goal-driven attention to objects and their meaning

• Task: Visual search for a target object

• Conditions: There can be visually related and semantically (meaning) related distractors

• Primary measure: Eye fixations as a function of time

• Results: people’s attention is drawn to both visual and semantic distractors, but visual distractors seemed to be more sailent

Goal-driven attention is flexible and task-dependent

• Pioneering work by Yarbus (1967)

• People fixated eyes differently on pictures depending on what task they got before → Top-down goals determine exploration

While stimulus-driven attention is fast and short

• Task: make an eye movement to left-tilted item, and ignore the right-tilted item

• Two main conditions: target can be the more salient one, or the less salient one (and we also varied eccentricty)

• Here probability of going to the salient item

Goal-driven attention is low and longer-lasting

• Task: make an eye movement to left-tilted item, and ignore the right-tilted item

• Two main conditions: target can be the more salient one, or the less salient one (and we also varied eccentricty)

• Here probability of going to the target item

Attention versus expectation

• Top-down attention

  • about the relevance of a stimulus

  • selects a stimulus

  • enhances the signal

  • flexible, adaptive

• Top-down expectation (prediction)

  • about the likelihood of a stimulus

  • selects an interpretation

  • suppresses the expected signal (predictive coding)

  • learned, engrained

Comparison

• Stimulus-driven

  • Driven by physical feature contrast

  • “Pops out” in parallel

  • Involuntary, automatic, little control

  • Early, rapid, and short-lived

  • Feedforward

• Goal-driven

  • Driven by goal: space, physical feature, object, or more abstract properties like meaning

  • Helps selection especially when target cannot be detected in parallel

  • Voluntary, controlled

  • Relatively late and slow

  • Feedback (as we will see next)

A Relevance Model

• Guided Search model (Wolfe, 1990 onwards)

• Feature maps (as in saliency models)

• Top-down bias by increasing weights on relevant features

• Bottom-up: feature maps (like in saliency models) that show what stands out in the image.

• Top-down: our goals or expectations, which boost the weight of relevant features (e.g. if you’re looking for something red, red features get more weight).

• Combines into top-down relevance map (here “activation map”)

Evidence for weighting: fMRI multivoxel pattern analysis

• Task: Attend to humans or vehicles

• Some voxels are more active for some categories than for others, while other voxels are less active

• Hence different categories of objects come with different voxel patterns

• Conversely: Once we know the patterns for a particular person, we can reconstruct what they are seeing, or attending!

• Results:

The same for space: fMRI

• First, map out space for each of four locations with retinotopic mapping

• Task: Monitor each of the locations in turn for a brief flip of the line

• Result: Enhancement of location-related brain activity

The same for space: EEG

• Attend to the left or attend to the right

• Early components in the event-related potential (P1, N1, P2, N2) are enhanced for attended stimuli

• So attention somehow enhances the neural representation of the attended information. How?

  • → Attention results in increased firing rate of relevant neurons

  • → Attention changes receptive field tuning

Receptive field tuning

• Look these three objects, and imagine three V4 neurons, each sensitive to one of them

• Receptive field sizes will overlap

• Plus neurons are mutually inhibiting: lateral inhibition

• As a consequence, neurons compete heavily for representing “their” object!

• Solution: Attention biases this competition, and the receptive field effectively shrinks

Reasons for attention

• “Capacity limits” or “Resource limits”

• But these have limited explanatory value

• We need to make explicit what the limiting factors are

• Competition for receptive field representation is one such limit

Areas driving the feedback

• Posterior parietal areas: top-down switching attention whether to locations or features

• Task: Attend Left/Right, or Attend Red/Green, and report the motion direction

• Manipulation: Change task between location and color tasks, or both

• → general purpose fronto-parietal attention network

• (A ventral network driving stimulus-driven selection)

• A dorsal network driving goal-driven selection

• With considerable overlap → priority map which combines bottom-up and top-down information

Repetition effects: Contextual cueing

• Task: Find the T and indicate its rotation

• Conditions: some displays are repeated in the next block

• Result: Repetition leads to faster detection

• Even though people are unaware of the repetition! → Implicit context effect

• Does not occur for amnesics with hippocampal damage

Other context effects: Reading

High frequent and predictable words are not fixed

Context effect: Scene knowledge

• Scene context strongly guides where we look

• When given a task: E.g. look for a pan

• But even under free viewing there are strong biases

  • “Places of interest”

  • Center bias

• Faster recognition with age

• Semantic and syntactic violations demand extra attention

• So experience influences selection

Repetition effects: Intertrial priming

• Task: Find the uniquely colored diamond , determine which side cut off

• Conditions: color scheme repeats, or switches from one trial to next

• Faster with repetition, but become slower if countinuesly flips around

• Is this automatic or under voluntary control?

• Conclusion: Observers are strongly driven by what they did previously, with little control

Repetition effects: Distractor suppression

• Can you prevent capture?

• Same design, stimuli & task as Theeuwes (1992)

• Two conditions: Distractor at any position (random, low probability) vs. distractor most frequently at one particular position (high probability)

• Results: Frequent distractor location suppressed or avoided

• Conclusion: Some control possible with extensive repetition

Value-driven attention: Effects of rewards

• Two phases: Training phase and test phase

• Training phase task: Find red or green target, respond to line inside

• Conditions: One target associated with high monetary reward, other with low reward

• Test phase task: Find diamond, respond to line inside. Ignore anything else

• Conditions: Distractor previously associated with high versus low reward, or no distractor

Value-driven attention: Effects of punishment

• Task: Find deviant shape, ignore distractors

• One distractor associated with electric shock (CS+), the other not (CS-)

The complete priority map

• Stimulus-driven, goal-driven, and experience-driven influences interact to create an overall priority map.

• Frontal Eye Fields may be one candidate to represent this overall priority map