Adaptation
A definition:
Adaptation tunes the response properties of a system to the
statistics of the incoming signal and the needs of the organism.
Principles:
-
Adaptation increases the dynamic range for signal processing.
-
Adaptation exists at many levels of organization, from molecules to
behavior.
-
Adaptation involves trade-offs, with some signal features being
emphasized at the expense of others.
Adaptation
Our approach to the topic:
- Emphasis on vision as a model system
- The stimulus is precise.
- There is a rich history of linking psychophysics,
biophysics, and molecular biology.
- There are parallels with other systems, sensory and
otherwise.
- Principles and mechanisms of adaptation
Adaptation
A definition:
Adaptation tunes the response properties of a system to the
statistics of the incoming signal and the needs of the organism.
Principles:
-
Adaptation increases the dynamic range for signal processing.
-
Adaptation exists at many levels of organization, from molecules to
behavior.
-
Adaptation involves trade-offs, with some signal features being
emphasized at the expense of others.
Encoding a Natural Distribution
Change in Mean
Change in Variance
Modified from Dunn and Rieke (2008)
Efficient Coding
1. The highest resolution is reserved for the most common signals.
2. All neural responses are equally likely.
(“Response” or “Histogram” Normalization)
Laughlin (1981)
redrawn in Wark & Fairhall (2007 )
Changing Stimulus Distribution
Laughlin (1981)
redrawn in Wark & Fairhall (2007 )
Changing Stimulus Distribution – Examples
Bullfrog hair cell
Turtle cones
Mouse hair cell
Cat retinal ganglion cell
Shapley and Enroth-Cugell (1984), Assad et al. (1989), Stauffer and Holt (2007)
Changing Stimulus Distribution
Laughlin (1981)
redrawn in Wark & Fairhall (2007 )
Changing Stimulus Distribution – Examples
Newt olfactory sensory neuron
Human somatosensory afferent
Kurahashi and Menini (1997), Knibestol and Vallbo (197
Efficient Coding
Laughlin (1981)
redrawn in Wark & Fairhall (2007 )
Adaptation
A definition:
Adaptation tunes the response properties of a system to the
statistics of the incoming signal and the needs of the organism.
Principles:
-
Adaptation increases the dynamic range for signal processing.
-
Adaptation exists at many levels of organization, from molecules to
behavior.
-
Adaptation involves trade-offs, with some signal features being
emphasized at the expense of others.
Increasing Dynamic Range through Adaptation
Neurons must encode a broad range of stimulus levels using a limited
dynamic range.
Stimulus levels can span >10 log units but most neurons do not operate
over more than ~2 log units.
Solutions:
-
Divide labor among cells of different sensitivities.
-
Adaptation
- Reduce the incoming signal
- Reduce signal capture
- Reduce response gain
- Reduce response summation
Division of Labor
Half-saturating intensity
Rods, 10 photons m-2
Cones, 103 photons m-2
IpRGCs, 107 photons m-2
~2 log units
400-700 nm, 10 log units of intensity
Peter Munro, Berson et al (2010), Baylor et al (1979)
Reducing the Signal Level
~10 log units
High photon capture
and optical aberrations
Low photon capture,
few aberrations
Mod. from Shapley and Enroth-Cugell (1984)
Reducing Signal Capture:
Fewer Receptor Molecules
Retinal Pigment Epithelium
Rods,
Cones
Mod. from Wang and Kefalov (2011)
R ti l G
li C ll
Reducing Gain
Macroscopic Response
Unitary Response
Response Compression
Arshavsky and Burns (2012)
Reducing Summation
Baylor et al. (1979), Matthews et al. (1990)
Mechanisms of Photoreceptor Adaptation
Weber-Fechner Law: S = 1/[1+(IS/I0)]
-
Rods
Reduction in signal capture
Some reduction in gain
Some reduction in summation
Response compression
-
Cones
Reduction in signal capture
Reduction in gain
Reduction in summation
No compression
-
IpRGCs
Reduction in gain.
Reduction in summation
No compression
Adaptation - Psychophysics
Light Adaptation
Weber-Fechner
Law
Weber-Fechner
Law
Barlow (1965), Shapley and Enroth-Cugell (1984)
Audition: Division of Labor
Intensity range: 10-12 – 102 W/m2 (104 perforates the eardrum)
Frequency range: 20 - 20 kHz
Audition: Reducing the Signal Level
Acoustic reflex - Muscle contraction
limits motion of the malleus and
stapes to decrease energy
transmission to the cochlea.
Outer hair cells – Somata contract in
counterphase with the basilar membrane to
dampen vibration. (They can also amplify
vibration by contracting in phase.)
Audition: Reducing Gain
Adaptation
A definition:
Adaptation tunes the response properties of a system to the
statistics of the incoming signal and the needs of the organism.
Principles:
-
Adaptation increases the dynamic range for signal processing.
-
Adaptation exists at many levels of organization, from molecules to
behavior.
-
Adaptation involves trade-offs, with some signal features being
emphasized at the expense of others.
Adaptation at Multiple Levels of Organization
Molecule
(Na+ channel)
Cell
Synapse
Rudy (1978), Lancaster and Nicoll (1987), Tsodyks & Markram (1997), Raman and Bean (2001)
Where to Adapt?
Van Essen et al (1992) Science
Adaptation
A definition:
Adaptation tunes the response properties of a system to the
statistics of the incoming signal and the needs of the organism.
Principles:
-
Adaptation increases the dynamic range for signal processing.
-
Adaptation exists at many levels of organization, from molecules to
behavior.
-
Adaptation involves trade-offs, with some signal features being
emphasized at the expense of others.
A Cost of Adaptation:
Lack of Information about Absolute Signal Values
Brainard et al (2006); Ed Adelson (MIT)
A Cost of Adaptation:
Loss of Information about Absolute Signal Values
Modified from Walraven et al (1990); gurneyjourney.blogspot.com/2010/01/color-constancy.html
A Cost of Adaptation:
Loss of Information about Absolute Signal Values
30
Patrick Cavanagh
A Cost of Adaptation: Noise
Signal
Response
Adaptation
Signal
Adaptation
A definition:
Adaptation tunes the response properties of a system to the
statistics of the incoming signal and the needs of the organism.
Principles:
-
Adaptation increases the dynamic range for signal processing.
-
Adaptation exists at many levels of organization, from molecules to
behavior.
-
Adaptation involves trade-offs, with some signal features being
emphasized at the expense of others.
Where to adapt?
Van Essen et al. 1992
Consequences of pooling inputs
1. Acuity‐sensitivity tradeoff
2. Improved signal‐to‐noise ratio
3. Risk of saturation in the target neuron
List of topics
1. Coupling and uncoupling of AII amacrine cells: adaptation controls the amount of pooling in the output signal
2. How much pooling is desirable in the signal controlling adaptation,
i.e. how late in the circuit should adaptation take place?
3. Adaptation to stimulus variance: contrast adaptation
4. Adaptation to stimulus variance: reward prediction error neurons (this week’s paper)
Convergence onto an AII amacrine cell
200‐450 rods
20‐30 rod bipolars
1 AII amacrine
gap‐junctions with ~20 other AII amacrines
gap‐junction to ON cone bipolar
inhibit OFF cone bipolar
AII‐AII coupling changes with light level
dark‐adapted
constant dim
adapting light
Bloomfield and Volgyi 2004
Bloomfield and Volgyi 2009
Gap junctions dissipate noise while preserving signal
No gap junctions
“single‐photon” stimulus
applied to 5 coupled AIIs
presynaptic response
AII response
With gap junctions
presynaptic response
AII response
simulations
Smith and Vardi 1995
Acuity‐sensitivity tradeoff
Starlight: Rod responses are rare; limit coupling to avoid dissipating these signals.
Twilight: Trade spatial acuity for sensitivity, i.e. the ability to detect correlated activity against a noisy background. Daylight: Signal is well above the noise. Uncoupling the network preserves acuity. Consequences of pooling inputs
1. Acuity‐sensitivity tradeoff
2. Improved signal‐to‐noise ratio
3. Risk of saturation in the target neuron
List of topics
1. Coupling and uncoupling of AII amacrine cells: adaptation controls the amount of pooling in the output signal
2. How much pooling is desirable in the signal controlling adaptation,
i.e. how late in the circuit should adaptation take place?
3. Adaptation to stimulus variance: contrast adaptation
4. Adaptation to stimulus variance: reward prediction error neurons (this week’s paper)
Problem:
Noise in the signal controlling gain can produce noisy fluctuations in gain.
Noise during a simulated fixation
rods
cones
Rieke and Rudd 2009
Conflicting demands:
Should adapt late (after pooling) to improve the SNR of the signal controlling adaptation.
Should adapt early to avoid saturation of early circuit elements.
Solution: Adapt early when light level is high, late otherwise.
Dunn et al. 2007
Pooling improves the SNR of the signal controlling adaptation
Dunn et al. 2007
List of topics
1. Coupling and uncoupling of AII amacrine cells: adaptation controls the amount of pooling in the output signal
2. How much pooling is desirable in the signal controlling adaptation,
i.e. how late in the circuit should adaptation take place?
3. Adaptation to stimulus variance: contrast adaptation
4. Adaptation to stimulus variance: reward prediction error neurons (this week’s paper)
Adaptation to stimulus variance
Blowfly H1
Stimulus parameter is velocity of horizontal motion of bar pattern
normalized velocity
of stimulus
firing rate
(spikes/sec)
Brenner et al. 2000
Efficient coding of a changing distribution
Laughlin 1981
redrawn in Wark & Fairhall 2007 Efficient coding of a changing distribution: offset and gain
Dunn and Rieke 2006
Variance adaptation as “adaptive rescaling”
firing rate
(spikes/sec)
stimulus velocity (deg/sec)
Brenner et al. 2000
Contrast adaptation in retinal ganglion cells
Baccus and Meister 2002
Contrast adaptation in bipolar cells
(but not in photoreceptors!)
Rieke 2001
Contrast adaptation in bipolar cell dendrites
Rieke 2001
Spiking adapts more than input currents in ganglion cells
Kim and Rieke 2001
Some sites of contrast adaptation in the retina
signal transfer from photoreceptor
to bipolar cell
synaptic depression at bipolar terminal, conjectured by Demb 2008
Na channel inactivation in ganglion cell
Consequences of pooling inputs
1. Acuity‐sensitivity tradeoff
2. Improved signal‐to‐noise ratio
3. Risk of saturation in the target neuron
Proposal:
If a cell has many inputs and a high enough gain to respond to any one input, it risks saturation when all inputs are activated.
To prevent this, contrast adaptation should occur most strongly at sites where many inputs are pooled.
Comparing parallel pathways with different pooling
Baccus and Meister 2004
LGN: slow contrast adaptation in the M pathway but not the P pathway
Solomon et al. 2004
List of topics
1. Coupling and uncoupling of AII amacrine cells: adaptation controls the amount of pooling in the output signal
2. How much pooling is desirable in the signal controlling adaptation,
i.e. how late in the circuit should adaptation take place?
3. Adaptation to stimulus variance: contrast adaptation
4. Adaptation to stimulus variance: reward prediction error neurons (this week’s paper)
stimulus
reward?
Fiorillo et al. 2003
Phasic response components (* and *) code “reward prediction error”:
[updated reward value (current + future)]  [previous reward value]
*
*
In some cells, the sustained response component (*) appears to code “uncertainty”
entropy
*
p
Fiorillo et al. 2003
Questions to keep in mind
Which component of the response adapts?
To which feature of the stimulus does the response adapt, according to the authors? How well have the authors demonstrated that this stimulus feature is actually driving the adaptation?
Guidelines for writing a paper critique: Aim for a length of 800‐1000 words. •
First, begin with a short summary. This should be no longer than 300 words (i.e., much shorter than the summaries you have been writing thus far). It should identify the major question(s) investigated by the paper, the major technique(s), the major result(s), and the paper’s significance. •
Identify the major problem(s) with this paper. The instructor may point out a specific major problem, but in other cases you may be asked to develop a critical assessment from scratch. You should explain the problem(s) in enough detail so that the authors could (in principle) write a targeted rebuttal. If you list any minor problem(s), make a clear distinction between these and the major problem(s). Number the problems you list and avoid repetition. •
For each major problem, state the implications of this problem. Does it make the interpretation of specific experiments difficult? Does it weaken your confidence in a major conclusion? Does it diminish the significance of the paper? •
When possible, indicate what solutions are appropriate. Should the authors soften a specific conclusion? Are new analyses required? If new experiments are needed, be specific.
•
It may also be appropriate to identify particular strengths of the paper. Avoid vague or flowery language: if you praise the paper, be specific and concrete.