1 of 15

The neural code�Combining information

Kenneth D Harris, UCL

2 of 15

How do brain circuits combine information?

  • Brain circuits often combine multiple types of information:

​

  • Different sensory modalities
    • E.g. auditory plus visual

​

  • Different stimuli within one modality,
    • E.g. parts of visual space
    • Auditory frequencies

​

  • Sensory responses and cognitive information
    • E.g. choices, representations of movements

​

  • Sensory responses and spontaneous activity
    • Discussed next week!

3 of 15

Additive hypothesis

  •  

 

 

 

4 of 15

Additive coding still leaves an open question

  • How do individual neurons coordinate multisensory responses?

​

  • Hypothesis 1: consistent coding
    • Neurons are selective for related features in the different modalities
    • E.g. stimuli in different parts of the world

​

  • Hypothesis 2: independent coding
    • A neuron may respond to one or both modalities
    • Its response to one modality has no correlation with its response to the other modality
    • “Mixed selectivity”

​

  • Hypothesis 3: segregated coding
    • Any neurons responds to only one of the two modalities

5 of 15

Hypothesis 1: consistent coding

  • Cells in barn own optic tectum can respond to visual and auditory stimuli

​

  • If a cell responds to both, it is driven by visual and auditory stimuli from the same location

​

  • May help owls locate prey combining visual and auditory stimuli

​

​

6 of 15

Hypothesis 2: independent coding

“congruent cell”

“opposite cell”

7 of 15

Hypothesis 2: independent coding

8 of 15

Hypothesis 3: segregated coding

  • Would require separate populations of neurons for each modality
  • And a negative correlation between the two responses
    • E.g. neurons responding to visual are less likely to respond to audition

​

  • Have not found cases like this for intermixed neurons within a brain region
  • When it happens, they seem to be topographically segregated

9 of 15

Nonlinear combination of stimuli

  • Could be very complicated!

​

  • Simple possibilities:
    • Normalization
    • Multiplicative interaction

​

  • More complex possibilities

​

10 of 15

Normalization model

Linear model doesn’t quite fit

 

 

11 of 15

Normalization is seen in many neural systems

12 of 15

Multiplicative interaction

13 of 15

More complex interactions

14 of 15

Nonlinear combinations allow linear readouts

15 of 15

Summary

  •