How do expectations shape perception
- Author
- Floris P. de Lange, Micha Heilbron, Peter Kok
- Published
Abstract
Perception and perceptual decision-making are strongly facilitated by prior knowledge about the probabilistic structure of the world. While the computational benefits of using prior expectation in perception are clear, there are myriad ways in which this computation can be realized. Here, we review recent advances in our understanding of the neural sources and targets of expectations in perception. Furthermore, we discuss Bayesian theories of perception that prescribe how an agent should integrate prior knowledge and sensory information, and investigate how current and future empirical data can inform and constrain computational frameworks that implement such probabilistic integration in perception.
Perception and perceptual decision-making are strongly influenced by prior knowledge about the probabilistic structure of the world.
This paper a) introduces recent studies of the neural basis, and b) introduces theories of perception based on Bayesian models.
Expectation in perception
According to Andy Clark and others, the brain is a prediction machine. a) Create a predictive model of the world and interpret sensory input based on it, b) Update the predictive model through sensory input. In this review we will look at a).
It introduces approaches in computational psychiatry that interpret schizophrenia and autism on the basis of this theory.
Perceptual consequences of expectation
The world is highly predictable. For example:
- Objects usually remain still or change slowly with the passage of time.
- Light sources are usually above us.
The brain uses this predictability to construct ‘expectations’ for sensory information and interprets sensory information based on these. Interaction between sensory information and expectations:
- When sensory information is inaccurate, expectations have a significant impact on interpretation. Example: After learning that a red dot moves to the left and a blue dot to the right, if a dot with ambiguous movement is shown, the judgment about the movement varies depending on the color of the dot.
- When sensory information is sufficiently clear or expectations are significantly misaligned, expectations do not have a significant effect on interpretation.
In other words, similar to Bayesian inference (prior + information —> posterior —> new prior). When current expectations are reliable and new information is weak, the prior changes little.
Where do expectations come from?
Forming expectations requires being able to recognize statistical regularities that exist in the world. How does the brain perform this task?
Depending on the type of regularity, it is encoded in different layers of the cortex:
- Regularities based on physical properties. Example: Shadows are usually formed underneath objects (because the light source is usually above them), and objects in peripheral vision usually disappear soon (because people usually walk forward).
- Regularity based on conditional probability. Example: While driving, other moving objects on the road are more likely to be other cars than sharks.
- Regularities formed by integrating multiple sensory information. Example: If you hear a dog barking, you are likely to see a dog.
Other cases:
- Highly variable regularities that must be learned quickly: Example: The sound an ice cream truck makes(?). Hippocampus and Prefrontal cortex are probably involved.
- Expectations created by one’s own movements: Example: Having moved your hand to catch a ball, you expect the tactile sensation of touching it. It is said that there are signals flowing from the motor system to the sensory areas.
Pop-out effect:
- Preattentive processing
- Tunable processing (process that can be tuned (by attention))
How do expectations modulate sensory processing?
Not only well-known cognitive factors such as Attention, but also expectations have a strong influence on the operation of early sensory areas. Stimuli that match expectations generally result in reduced reactivity.
- There is still debate about how to interpret this action.
- Some researchers speculate that it acts as a kind of ‘data compression’ because if it matches expectations, the information value is low.
Expectation in computational models of perception
It is clear that the responsiveness of low-level neurons varies depending on expectations, but the computational functions of this modulation require further study.
- Hypothesis 1 under discussion: Bayesian inference. The normative approach is that an ideal prediction machine would follow Bayesian methods. Supported by some experiments.
- Hypothesis 2 under discussion: Predictive coding. A method of creating a virtual model of the world and continuously comparing the movements of this model with real sensory information. Supported by some experiments.
Alterations of expectation in psychopathology
Research on expectations may also be useful in psychopathology. It makes it possible to explain some neurodevelopmental disorders within a unified framework.
- Autism Spectrum Disorder: Expectations are given too little weight. They process the details of sensory information very well, but are unable to integrate or interpret it. Excessive sensitivity to unexpected stimuli.
- Schizophrenia: Expectations are given too much weight. Sensory information is perceived as present even when it does not actually exist.
Concluding Remarks and Future Perspectives
We looked at recent discussions about how the brain uses prior information about the world to better interpret sensory information.
Interesting research topics to pursue in the future:
- How to integrate predictive information from different sensory channels to create a single model. Is there such a thing as a “blackboard” shared between channels?
- Find out what we can learn from deep learning research.
- Technological advances are making it possible to observe larger areas of the brain more precisely.