Top-down versus bottom-up

Tricky to define.

Here’s an attempt by Wolfe et al (2003) [Changing Your Mind: On the Contributions of Top-Down and Bottom-Up Guidance in Visual Search for Feature Singletons, JEP:HPP, 29, 483-502]:

When you look at Figure 1 [a picture with some shapes and things, in the middle of which is a spiky object], your attention is probably attracted to the spiky diamond. It is a salient item, and, all else being equal, salient items that are different from their neighbors tend to attract attention (Egeth, 1977; Julesz, 1986; Moraglia, 1989). The information that guided your attention to that item can be labeled as bottom-up—meaning that it did not depend on the observer’s knowledge of the stimulus. The stimulus itself provides the guidance.

If you are now asked to find the white vertical lines, you can do this with no particular difficulty. The stimuli did not change when you performed the second task. Instead, you changed your mind in response to the suggestion that you look for white verticals. You can find white verticals by guiding your attention to the intersection of the set of white items and the set of vertical items (Wolfe, Cave, & Franzel, 1989). The information that guided your attention in this case can be labeled top-down—meaning that it depended on the observer’s knowledge.

Not sure I like this. I see what they’re getting at, but it’s as if in one case the outside world is causing something whereas in the other case it’s being caused by the head. In reality, there must be some encoding of knowledge which causes one to view a particular stimulus as more salient than another, even in the absence of a feeling of control.

Another (Hahn et al, 2006 [Neuroanatomical dissociation between bottom–up and top–down processes of visuospatial selective attention, NeuroImage, 32, 842-853]):

Bottom–up or exogenous attentional control is stimulus-driven, i.e., attention is spontaneously oriented towards an oncoming stimulus. Top–down or endogenous attentional control, by contrast, is intentional and cognitively driven, i.e., directed by knowledge, expectation and current goals.

I definitely don’t like this. If you’re using cognitive models then everything is cognitive. Percepts don’t just move directly from world to mind. There must be some goal achieved by orienting to a stimulus, perhaps just not the main goal being processed at a particular time.

Some more [previously] scribbles on Burgess et al’s review of BA10… This stuff builds upon Shallice’s Supervisory Attentional System (SAS), a theory of behavioural organisation in which there are four levels:

  1. Cognitive/action units (e.g. particular actions like reaching).
  2. Schemata (groups of cog/action units, tied together if they often appear together)
  3. Contention scheduling (quick selection of actions in routine behaviours)
  4. Supervisory Attentional System (the bit that does what feels like deliberation)

Types of cognition, according to the authors:

  • Stimulus oriented thought (SOT): provoked by something sensed or oriented towards something sensed
  • Stimulus independent (SIT): examples include daydreaming, rumination, introspection.

Many include a mixture of both and there will be continual competition. Again, an attempt to get at top-down versus bottom-up type distinctions. Though this is nicer, I feel.

Here’s a picture of version 2 of their theory (from Burgess et al, 2007).


Paul W. Burgess, Jon S. Simons, Iroise Dumontheil, and Sam J. Gilbert. (2005). The gateway hypothesis of rostral prefrontal cortex (area 10) function. In J. Duncan, L. Phillips, & P. McLeod (eds.), Measuring the Mind: Speed, Control, and Age. Oxford University Press, Oxford. pp. 217-248.

Burgess, P. W., Dumontheil, I., & Gilbert, S. J. (2007). The gateway hypothesis of rostral prefrontal cortex (area 10) function. Trends in Cognitive Sciences, 11, 290-298.

4 thoughts on “Top-down versus bottom-up”

  1. I think you’re quite right — there appears to be some confusion here: all perception must involve both stimulus and knowledge elements, an interaction of top-down and bottom-up processing. What the authors of the first two papers seem to be talking about is “bottom-up” and “top-down” *goals* for perceptual *attention*. Is the appropriate? Could you reframe it, at least in the first example, as two different kinds of (top-down) goals: 1. Look at whatever appears to stand out 2. Look at the white lines?

  2. That sounds better.

    It’s all very subtle, isn’t it?

    I’ve been looking around at computational models of saliency (this is all a distraction from Real Work, so I’m not doing a very thorough job of it!). They’re quite nice as you can ask punters to pick out the salient object in an image (or do it via eyetracking) and then see if your model would pick the same object—and importantly, you can vary how much information your model gets and see how that affects things. So one of the papers I looked at—and I didn’t grasp the details, it’s all rather involved—described lots of filters, and detectors for change, and smoothers, and such like, systematically breaking the input, so you can work out with such impoverished stimulus what on earth you could do.

    So rather than top-down versus bottom-up, you’ve got parallel processes all running, all competing with each other, some controling others, feeding back, and so on. I want—and someone has probably done it, I don’t know the visual attention lit—a description in terms of the various kinds of processing going on in the wet stuff or in a cognitive-model abstraction. Rather than a “bottom-up” process winning, I’d want a description of a motion detector in foo-pathway grabbing the eyes via bar-pathway.

    Capabilities, connections, and control? With recursion.

    This is all horribly vague. I’ll fire some examples of description I found Very Soon (in the coming months) to give you the flavour of what I mean…

  3. In Ayurveda, there are three physical constitutions- Vata, Pitta, and Kapha. Each one is unique, and most of us are a mixture of them. But we each have a dominant constitution- if I am overweight and usually chilly, I’m kaphic. If I am thin and spacy-minded, I’m more vata. I wonder if the mind is similar. That is, that each person has their own mixture. Their own way of sorting their information, based on genetics & environmental factors.
    Maybe the way we relate to a stimulus may be inherently tied to our own physical, environmental, and genetic circumstances. Thereby leading to some extreme mental states in some of us while not in others, based on all of these things.
    I have been out of the loop of psychology for a little while now…and when I was in it, it was mostly psychoanalytic and contemplative Buddhist-psychology-related things. So I apologize if what I’m saying is just common sense. It helps me to organize the information though 🙂 Hope you all are well!

Comments are closed.