neural modeling - fall 13861 neural transformation strategy to discover the brain functionality...
TRANSCRIPT
Neural Modeling - Fall 1386 1
NEURAL TRANSFORMATION
Strategy to discover the Brain Functionality
Biomedical engineering GroupSchool of Electrical EngineeringSharif University of Technology
Neural Modeling - Fall 1386 2
Transmission of Action Potential/ Dendrite potential: Corrected
t
vRC
x
v
2
2
),(2),(
2
2
fxfRCVjx
fxV
xfRCjxfRC eeVfxV 0),(
)1()1(22
22
22
02
4/
2/2
2
jfRCjfRCefRCm
fRCefRCjm
fRCjm
j
j
Neural Modeling - Fall 1386 3
NEURAL TRANSFORMATION
Neural representation paves the way for a useful understanding of neural transformation
Can be characterized using linear decoding The representational decoder A
transformational decoder. Transformation on encoded information: To
extract information other than what the population is taken to represent.
Neural Modeling - Fall 1386 4
Transformation vs Decode
x Representation yTransformation
Decodex
F(x)
What transformations a neural population can, in principle, support?
How well a given neural population can support the transformations defined by a particular class of functions?
observed in a neurobiological system.
Neurons with certain response properties support particular transformations better than others
we do not need to rely on training to have interesting,biologically plausible models.
Neural Modeling - Fall 1386 5
Static vs Dynamic Static computation of a function is not, alone, a
good description of the kinds of transformations neurobiological systems typically exhibit.
To characterize the dynamics of the transformations that neural populations support
Engineering Tools: State Vectors Neural representations can play the role of
“State Vectors” To address issues that have proven very difficult
for other approaches
Neural Modeling - Fall 1386 6
About THREE PRINCIPLES OF NEURAL ENGINEERING Guiding assumptions of our
approach. Numerous examples of detailed
models of a wide variety of neurobiological systems
To demonstrate how to use these principles
A methodology for applying these principles.
Neural Modeling - Fall 1386 7
3 Principles P1: Neural representations are defined by the
combination of nonlinear encoding (exemplified by neuron tuning curves) and weighted linear decoding
P2: Transformations of neural representations are functions of variables that are represented by neural populations. Transformations are determined using an alternately weighted linear decoding (i.e., the transformational decoding as opposed to the representational decoding)
P3: Neural dynamics are characterized by considering neural representations as control theoretic state variables. Thus, the dynamics of neurobiological systems can be analyzed using control theory
Addendum : Neural systems are subject to significant amounts of noise. Therefore, any analysis of such systems must account for the effects of noise
Neural Modeling - Fall 1386 8
Methodology Central goal: To provide a general framework
for constructing neurobiological simulations The guiding principles A methodology for applying those principles
A software package in MatLab The models have been implemented with this
package. Three stages:
System description Design specification Implementation
Neural Modeling - Fall 1386 9
System description Identify the relevant neurobiological
properties (e.g., tuning curves, connectivity, etc.).
Specify the representations as variables (e.g., scalars, vectors, functions, etc.).
Provide a functional description including specification of subsystems and overall system architecture.
Provide a mathematical description of system function.
Neural Modeling - Fall 1386 10
Design specification
Specify the range, precision, and signal-to-noise ratio for each variable.
Specify the temporal and dynamic characteristics for each variable.
Neural Modeling - Fall 1386 11
Implementation Determine the decoding rules for
implementing the specified transformations.
Determine which parts of the model are to be simulated to which degrees of detail.
Perform numerical experiments using resulting simulation.
Neural Modeling - Fall 1386 12
A POSSIBLE THEORY OF NEUROBIOLOGICAL SYSTEMS
Presented: A ‘framework’ that consists of a set of three principles and a corresponding methodology.
The possibility that the three principles can be properly called a theory of neurobiological systems.
The practical utility of the framework itself is independent of whether this claim is found convincing
Neural Modeling - Fall 1386 13
The state of theories in neuroscience There aren’t any. There aren’t any good ones
Churchland and Sejnowski 1992 Marder et al. 1997 Stevens 1994 Crick and Koch 1998 Stevens 2000
Neuroscience is, in other words, “data rich, but theory poor” (Churchland and Sejnowski 1992, p. 16).