· 10:00 Alex Dimitrov Intoduction
·
10:10 William Levy Energy
Efficiency as a Constraint on Neural Computation
·
10:50 David Goldberg An
Efficiency Approach to Spike Coding
·
11:20 – 11:35 Discussions
and break
· 11:35 Sam Wang Speed
limits in mammalian brains: scaling constraints from biophysics
· 12:15 Mark
Nelson Design
constraints for an active sensing system
· 12:55 –
13:10 Discussions
and break
· 13:10 Raul Rodriguez-Esteban Adjacency as a
measure of cost
·
13:40 Mitya Chklovskii Wiring
cost minimization explains many features of brain architecture
· 14:20 Alex
Dimitrov Unit
failure and neural circuit complexity
· 14:45 Closing
comments and discussions
· 15:00 Session
ends
Title: An Efficiency Approach to Spike Coding
Abstract:
Many
have argued that efficient representation of information in the
nervous system has an evolutionary advantage. This efficiency principle
has been used to predict and explain neural processing on the functional
level, and has provided us with concepts such as redundancy reduction.
In this work, we apply this optimization approach to neural processing
on the physical level, specifically to the question of how neurons
represent information with spikes. We explore how the physical
properties of the channel and latency requirements influence whether a
rate code or a temporal code is more efficient.
Title: Energy Efficiency as a Constraint on Neural Computation
Abstract: Many parametric observations
in Nature can be made sensible
when interpreted from the perspective of energy efficiency. Some of us
have incorporated this energy-efficient perspective into an
information-theoretic perspective of brain function. Specifically,
neuronal coding and neuronal computation, interpreted from the perspective
of energy efficiency, makes sense of the low firing rates and of the high
synaptic failure rates observed in forebrain cortical systems.
Title: Wiring cost minimization explains many features of brain
architecture
Abstract: Just as computer engineers have to solve layout and routing
problems
in the process of chip design, nature had to optimize the placement
of neuronal complexes and the morphology of neuronal arbors in the
course of evolution. We attempt to reverse engineer the brain by
formulating wiring optimization problems and systematically comparing
their solutions with the anatomical data. We began to assemble a solid
theoretical framework that unifies many previously disjoint anatomical
facts. Such framework helps explain existing experimental observations
and guide future experiments.
Title: Design constraints for
an active sensing system
Abstract: The electrosensory system of weakly electric fish provides an
interesting case study for exploring a variety of influences on
the design of an active sensing system.
Topics to be considered
include metabolic constraints associated with the generation of
an electric field to actively probe the envrionment, biophysical
constraints associated with reliable transduction of weak
(microvolt-level) voltage perturbations, signal processing
constraints associated with low signal-to-noise conditions, and
functional constraints that may help explain the presence of
multiple topographic maps in the brainstem electrosensory
nucleus. Neuroethological analysis suggests that sensory system
design is strongly influenced by the dynamic coupling between brain,
body and environment that occurs in the context of natural
behavioral tasks.
Title: Unit failure and neural circuit complexity
Abstract: Are larger and more
complex brains just a collection of more of the same
basic building elements, or do they differ in some more fundamental aspects? I
argue
that one important aspect in which large brains differ from small brains is in
the
different needs to protect from failures of basic computational elements. Large
brains
tend to exist in longer-living animals, and long life places more stringent
requirements
on error correction and unit failure. Using tools from the theory of reliable
computation
(akin to information theory), I derive some trends in relative circuit
complexity and
demonstrate those trends in insects (ultra-short to short-living species) and
mammals
(medium- to long-living species).
Title: Speed limits in
mammalian brains: scaling constraints from biophysics
Abstract: The mammalian neocortex scales according to orderly power
laws. This
allometric scaling suggests that variation in neocortical form, previously
known only as empirical phenomena, reflects shared design principles. We
find that several scaling laws can be explained by a requirement to keep
fast cross-brain conduction times at 1-2 milliseconds, independent of
brain size. Because faster
conduction requires thicker axons, the fastest
axons occupy disproportionately more space in big brains. The consequent
increases in volume can account quantitatively for increases in the ratio
of white matter to gray matter volume.
Thus, macroscopic neocortical
scaling is explainable in terms of biophysical constraints.
Title: Adjacency as a measure
of cost
Abstract: Recent results in the field of neuroanatomy have pointed out
that a simple
metric rule such as accounting the connections between adjacent pairs of
neuronal entities seems to be a good indicator of overall connection cost.
The preeminence of such an elementary adjacency rule over more sophisticated
metric measures is one of an intriguing nature. The possibility that the
design of the nervous system is somehow driven by a save-wire paradigm
compels us to bring this adjacency rule under tighter scrutiny. For this
matter, we have developed a number of different mathematical and
analytical tools to confront this task. We have also proposed a number of
hypotheses on what are the mechanisms involved and what are the
implications that this phenomenon suggests. The use of the adjacency rule
as a primitive chip design technique raises questions about the
relationship between this type of design and the design of the nervous
system architecture.