A1261965618BTWuY

1
Learning About Odors Artificial, Biological
Computational Olfaction
Alan Gelperin Monell Chemical Senses
Center Philadelphia, PA 19104
2

• Outline
• Biological Olfaction Mollusk
• smart, smells well, accessible neurons
• Biological Olfaction Mouse
• record mitrals in vivo during odor
  learning
• Artificial Olfaction Sensors
• enose inputs from OFET sensor array
• Artificial Olfaction Robot
• odor localization with onboard enose

3
Collaborators Mollusk Olfaction David Tank,
David Kleinfeld Kerry Delaney, Jing Wang, Dima
Rinberg John Hopfield, Bard Ermentrout, Chris
Sahley Mouse Olfaction Dima Rinberg, Michale
Fee, Gary Beauchamp, Kunio Yamazaki, Fred
Ollinger, Sensor Arrays Howard Katz, John
Hopfield George Preti, Ananth Dodabalapur, Takao
Someya Robot Olfaction Dan Lee, Dima
Rinberg Boris Shraiman, Eugene Balkovsky,
4
Let me tell you the difference between industry
and academia.
In industry, its dog eat dog.
In academia, its just the opposite.
Edwin Whitehead
5
(No Transcript)
6
Train Test Odor Learning
7
Logic of Limax Learning
1st Order Conditioning A Q- A-, B
2nd Order Conditioning A Q- then B A-
A- , B-, C Compound Conditioning A B
Q- A- , B- , C Block of Conditioning
A Q- then B A- Q- A- , B

• Behavioral experiments probe associative memory
  functions
• Before conditioning, A, B, C are attractive ()
  odors, while
• Q is a repellent (-) taste

8
(No Transcript)
9
(No Transcript)
10
Olfactory lobe, view from the top
11
(No Transcript)
12
(No Transcript)
13
MEMORY BANDS
Train slug to associate mint odor and food
Inject Lucifer Yellow into slug
Dissect brain and examine lobe for LY
Wait 20 min
Wait 60 min
Lucifer Yellow band is parallel to wave front.
14
(No Transcript)
15
(No Transcript)
16
Band of Neurogenesis in the Olfactory Lobe
olfactory lobe labeled with BrdU 2 months prior
to developing BrdU
apex

• Zone of neurogenesis is at the apical tip of the
  olfactory lobe
• As the apex grows, previously labeled neurons are
  left behind

17
Band of Neurogenesis in Olfactory Lobe

• Olfactory lobe is labeled with BrdU 2 days prior
  to developing the BrdU signal

18
Map mollusk to mammal

• Central representations of odors
• spatial pattern?
• temporal pattern?
• Effects of learning on representation
• Redundancy in representation

19
Specificity and generality
Moth nose is narrowly tuned, dog nose is general
purpose
20
Mouse Is Macrosmatic
OLFACTORY BULB

• OB is 0.2 of human brain, 5 of mouse brain

21
(No Transcript)
22
Central Olfactory Pathways
Olfactory bulb targets
Olfactory bulb
Orbitofrontal cortex
Piriform cortex
Olfactory tubercle
Thalamus
Olfactory receptors
Hypothalamus
Amygdala
Entorhinal cortex
Hippocampal formation
from Neuroscience, second edition, edited by D.
Purves, G. J. Agustine, D. Fitzpatrick, L. C.
Katz, A.-S. LaMantia, J. O. McNamara, S. M.
Williams
23

• Enigma of Mouse Olfactory Bulb
• Remove 80 of single remaining bulb,
• then begin to see odor processing deficits
• If odor processing is distributed, what is
• the function of the redundant representation?
• Need to record in vivo before, during and
• after odor meaning is changed by learning
• Need to challenge odor processing system
• with more difficult tasks short, complex odors

24
Mouse Motorized Electrode Chamber
25
(No Transcript)
26
(No Transcript)
27
(No Transcript)
28
(No Transcript)
29
(No Transcript)
30
(No Transcript)
31
(No Transcript)
32
(No Transcript)
33
(No Transcript)
34
(No Transcript)
35
Uses of electronic olfaction
Medical monitoring, quality control and land mine
detection
36
Organic FET Sensors
Odor may interact with surface, bulk, or
interface. Thin active layer Interactions may
depend on gate charge. FET parameters such
as Ion, Ioff, and VT may react differently
to a given odor, increasing selectivity.
37
Series of OFET odor responses
odor
Series of 74 odor responses with source-gate
voltage reversed between successive odor
applications
38
Diversity of Transistor Odor Responses
Blue () response Red (-) response Blackno
data Whiteno response
Responses of 11 sensor materials to 16 odors
39
ROBOSNIFFER AUTONOMOUS ODOR TRAIL FOLLOWING
ROBOT WITH ENOSE
Enose samples substrate odors and sends
sensor data continuously via wireless link to
central computer
Collaboration with Dan Lee Dept. EE
Bioeng. Univ. Pennsylvania
40
ROBOSNIFFER SAMPLES ODOR TRAILS

• Odor samples from floor under robosniffers snout
  are
• continuously sampled by the Cyranose 320 enose

41
Electronic walking moth
Odor wind sensor info via wireless link to
central server for computation of action

42
Moth strategy the vapor chase
Ishida Morrizumi, Handbook of Machine
Olfaction, Pearce et al (eds) 2002
Mothred Plumewhite
Courtesy of J. Hildebrand, University of Arizona
43
Algorithms for odor search
Balkovsky Shraiman, Proc. Natl. Acad. Sci.
USA(2002) 99 12589
44
Slugbot Living off the land, very slowly

• Robot captures slugs and digests them to obtain
  energy

45
Electronic olfaction

• Electronic olfaction is a technology
• in its early evolutionary stages
• E-Noses can provide solutions to
• important odor-sensing problems
• Electronic olfaction contributes to both
• computational and biological olfaction

46

• Summary
• General principles of olfactory information
  processing are emerging
• Plasticity of processing is a universal feature
• in mammals and molluscs
• Early stages of olfactory processing involved
• in experience-dependent modifications
• Artificial, biological and computational
  olfaction
• offer synergistic perspectives