VL8 prediction in vision and beyond

—> expectations shape perception

—> brain is proactive —> coarse to fine processing

• feedback loops allow us to make hypotheses —> anticipate the consequences of actions and formulate expectations

—> brain is not reactive

—> low and high complexity of prediction

• processed information gets increasingly complex

1. retina

2. optic nerve

3. optic chiasma

4. lateral geniculate nucleus

   1. magnocellular layers —> black

   2. koniocellular layers —> blue /yellow

   3. parvocellular layers —> red/green

5. primary visual cortex

1. Dorsal stream —> parietal lobe -> where/how

2. ventral stream —> temportal lobe -> what

processed information gets more and more complex

• 40-60 ms: v1

• -80 ms: v2/v4

• -100 inferotemporal cortex —> object recognition

• -120 ms to PFC and V1

• -140 ms inferotemporal cortex to v2/v4

• 150+ ms pfc to inferotemporal cortext

• information is often

  • incomplete

  • noisy

  • ambiguous

• to predict better/earlier/faster means evolutionary advantage

• to build internal models to plan actions

• Life priors

• contextual priors

• expectations

assumptions and expectations based on an individual's accumulated experiences over their lifetime.

examples

• light usually shines from above

• how reflections look

• textures -> familiarity with textures —> we expect them to look a certain way

Contextual priors are more narrowly focused on specific contexts or environments

• shading

• relative sizes in scenes

• bistability

  • how we see the obect depends on what we expect/have seen before

• cueing! When being told letter E —> recognizing E

• prior experience

incomplete information

taking statistical regulatrities of the world and letting them fill the missing information

two visual pathways

• magnocellular (black) —> dorsal stream —> where?

• parvocellular (red and green) —> ventral stream —> what?

• low spatial frequency

• motion

• fast signal

• where?

—> from v1 to pfc

—> coarse information that is faster —> proactive guessing

• what?

• high spatial frequency

• colors

• slow signal

1. Input

2. two separate pathways simultaneously

   1. lsf object —> possible objects

   2. lsf scene image —> possible context

3. brain region

   1. MPFC / OFC

   2. RetroSplenial Complex; ParaHippocampal Cortex

4. combination in inferotemporal cortex

1. Associations based on experiences are foundation for

2. Analogies which trigger specific

3. Predictions relevant to the current context

• based on past experiences

• brain regions

  • medial temporal lobe

  • hippocampus

  • parahippocampal cortex

• map new inputs to existing representations to look for similarities

• association-based prediction

• not recognition!

• analogies trigger predictions relevant to current context

• spatial frequency

• context information

—> simultaneous bottom up and top down processes due to

• incomplete data

• contextual information

• prior knowledge