Gain control in insect olfaction for efficient odor recognition

1
Gain control in insect olfaction for efficient
odor recognition

• Ramón Huerta
• Institute for Nonlinear Science
• UCSD

2
The goal

• What is time and dynamics buying us for pattern
  recognition purposes?

One way to tackle it
1. Start from the basics of pattern recognition
organization, connectivity, etc.. 2. See when
dynamics (time) is required.
3
How does an engineer address a pattern
recognition problem?

• Feature extraction. For example edges, shapes,
  textures, etc
• Machine learning. For example ANN, RBF, SVM,
  Fisher, etc..

What is easy ? What is difficult?

• Feature extraction very difficult (cooking
  phase)
• Machine learning very easy (automatic phase)

4
How insects appear to do it
Mushroom body (MB)
Machine Learning Stage

• Feature Extraction

High divergence-convergence ratios from layer to
layer.
Mushroom body lobes
5
Bad news
The feature extraction stage is mostly
genetically prewired
Good news
The machine learning section seems to be plastic
6
Spatio-temporal coding occurs here
No evidence of time here
Mushroom body (MB)
Machine Learning Stage

• Feature Extraction

Mushroom body lobes
7
The basic question

• Can we implement a learning machine with
• fan-in, fan-out connectivities,
• the proportion of neurons,
• local synaptic plasticity,
• and inhibition?

Huerta et al, Neural Computation 16(8) 1601-1640
(2004)
8
Marr, D. (1969). A theory of cerebellar
cortex. J. Physiol., 202437-470. Marr,
D. (1970). A theory for cerebral neocortex.
Proceedings of the Royal Society of London B,
176161-234. Marr, D. (1971). Simple
memory a theory for archicortex. Phil. Trans.
Royal Soc. London, 26223-81. Willshaw D,
Buneman O P, Longuet-Higgins, HC (1969)
Non-holographic associative memory, Nature
222960
9
Stage II Learning perception of odors
Stage I Transformation into a large display
CALYX Display Layer Intrinsic Kenyon Cells
MB lobes Decision layer Extrinsic Kenyon
Cells
AL
Learning required
No learning required
PNs (800)
iKC(50000)
eKC(100?)
10
(No Transcript)
11
Odor classification
Odor N
Class 2
Odor 4
Odor 3
Odor 2
Odor 1
Class 1
12
Sparse code
Probability of discrimination
of active KCs
13
Capacity for discriminating
We look for maximum number of odors that can be
discriminated for different activate KCs,
TOTAL OF ODORS
of active KCs
Note we use Drosophila numbers
14
It has been shown both inLocust (Laurent)and
Honeybee (Menzel)the existence of sparse
code1 activity
15
Narrow areas of sparse activity
Without GAIN CONTROL There can be major FAILURE
16
GAIN CONTROL
Mushroom body (MB)
Machine Learning Stage

• Feature Extraction

But nobody knows why
Mushroom body lobes
17
Evidence for gain control in the AL

• These neurons can fire up to100 Hz
• The baseline firing rate is 3-4Hz

Data from Mark Stopfer, Vivek Jayaraman and
Gilles Laurent
18
Honeybee Galizias group

• There seems to be local GABA circuits in the MBs.
• Locust and honeybee circuits are different
• Honeybee 10 times more inhibitory neurons than
  locust

19
Lets concentrate on the locust problem How do
we design the AL circuit such that it has gain
control?
20
Mean field of 4 populations of neurons
21
We apply mean field
22
Define new set of variables
To obtain the mean field eq.
Where we use
23
We look for the condition such that
Whose condition is
with
and
The gain control depends only on the inhibitory
connections
This works if and are linear
BUT!
24
SIMULATIONS 400 Neurons
The excitatory neurons are not at high spiking
frequencies or silent, but
but not very high (3-4) Hz. So
25
The gain control condition from the MF can be
estimated as
26

• A few conclusions
• Gain control can be implemented in the AL network
• It can be controlled by the inhibitory
  connectivity. The rest of the parameters are free.

Things to do I do not know whether under
different odor intensities the AL representation
is the same.
27
Thanks to

• Marta Garcia-Sanchez
• Loig Vaugier
• Thomas Nowotny
• Misha Rabinovich

• Vivek Jayaraman
• Ofer Mazor
• Gilles Laurent