Farah Baracat

Day 4 - Information Processing and Plasticity - Role of Dendrites

Morning Session #1

• Why dendrites?
  • they have elaborate dendritic trees which are very diversse across regions and function (e.g motor neuron, )

  • also in terms of structure/morphology

  • eg. Pyradimal cell:
    • soma: that looks like a pyramid
    • basal dendrites on the bottom
    • apical dentras on the top from the apical trunk

  • dendrites are input devices: synapses are located near the dendritic tree

  • e.g L2/3 receives 7000 synapses. There seems to be more bias towards the distal dentdrites (receive more input)

  • Persistant spines (the ones that stay for months) ?
• Dentrites express ion channels across the tree
  • Distribution of the channels vary along the dendritic axes, also regions and cell tyoes.
  • Spatial control of ionic channels across the tree
  • ion channels can generate spikes: dendritic spikes why? to efficiently propagate the spikes to the soma
  • back propagating AP are generated from the soma and propaate back to the apical dendrites
  • NMDA spikes (100 ms): cannot propagate very far, its more localized (they won’t reach the soma) as opposed to Ca+ spikes (10s ms) or Na+ spikes (<5ms)
    • Why are these localized needed? to control the hot zone of the Ca+spike at the apical. They serve as control but not
• Should we care about modelling the dendrites
  • dendritic spikes are another non-linearity
• input specificity:
  • segregation of the input: where it is applied is an important characteristics
• What can these dendrites do? how does the neuron use it to perform operation?

  1. attenuation: when putting synapses far away, they are less in amplitude –> neuron’s solution is having a local zone (NMDA) but this can be utilized as a feature to have a coincidence detector: only is th einput is arriving in a specific direction, it will have an effect (direction selectivity). This has also been shown in neuromorphic hardware: cf paper Kwabena, dendrocentric learning

  2. Integration

     • Scatter-sensitive vs cluster sensitive:
       • In scatter-sensitive, scatter the inout across the dendritic tree –> we get an I_syn that reaches a plateau
       • Cluster-sensitive: more localized

Bartlett

• Interested in object reconition: how can we detect boubdaries (this is what allows us to recognize objects, to buid up shape representation)
• “they were trying to understand how to achieve this task or what is needed to achieve this task”
  • how do we then detect boundaries: maybe from NMDA non-linearity: each edge is coding in a single neuron firing. When two enuron (edges) fire together they would present that shape (i.e 2 edges connecting)

  • cells are really goos at foing binding and pooling

  • Now looking at the variety of cells we have in the visual system: e.g Parvo and magno cells. How to combine those variables? Typically the brain does that all the time:
    • Combining a lot of (bunch of) non-linear basis functions and add weights on top of that.
    • cd. paper universal computation: if you have many dendritic non-linearities, we can perform very complex pattern recognition
    • The way those basis functions are created is by the connectivity patterns of the inhibition on the differnet dendritic columns??
      • flexibility to hit the dendrite at different spatial locations
      • • nmda non-linearity
  • Dendritic branch that has spatial non-linearity in it.s