|
Publications
N Venkateswaran, R Rajesh, N Sudarshan, R Rajasimhan, C Chandramouli, R
Chidambareswaran, B Harish, Kolluru Arvind, M Muhilan y(PSI)NAM For Massive Neuronal Assembly Modeling: Part-I, Processing Elements,
The
6th International Conference on Computational Intelligence and Natural
Computing, 2003
Abstract
The title y(PSI)NAM stands for Parallel SImulation (y-PSI) for Neuronal Assembly
Model(NAM). An instruction driven NAM processor architecture had been proposed
to model a complex stochastic neuronal assembly, capable of modeling a wide
class of neuronal models including the stochastic model. However, modeling the
neuronal assembly using this processor will be computationally inefficient as it
does not include special functional units to model the signal processing
characteristics of complex dendritic tree structure. This paper proposes two
processor architectures: one based on mixed signal approach - the Mixed signal
NAM (MNAM) processor,and the other based on digital approach - the Digital
Dendritic NAM (DDNAM) processor. In reality, modeling the complex brain
functions and its fault simulation demands enormous computational resources
beyond the number crunching capability of either a single MNAM or a single DDNAM
processor. The companion paper, Part-II proposes a novel array architecture to
meet this end. Insert contents.
DOCUMENT IN
PDF
N Venkateswaran, R
Chidambareswaran, B Harish, Kolluru Arvind, C Chandramouli, R Rajesh, N Sudarshan, R Rajasimhan,,
y(PSI)NAM For Massive Neuronal Assembly Modeling: Part-II, The
Array Architectures, The
6th International Conference on Computational Intelligence and Natural
Computing, 2003
Abstract
The proposed processors in "y(PSI)NAM For Massive Neuronal Assembly Modeling:
Part-I,Processing Elements" DDNAM and MNAM, are effective for small-scale
neuronal assembly modeling. Realistic modeling of massive neuronal assemblies
demands enormous computation resources beyond the number crunching capability of
either a MNAM or a DDNAM processor. This has led to the design of two array
processors, based on mixed signal and digital approaches for brain function
modeling. Extensive Simulation of these array architectures for modeling
assemblies and a comparative analysis with published experimental results were
carried out. The comparison shows the fulfillment of the objective of the array
processor with regards to realistic modeling of massive neuronal assemblies. For
neuronal modeling, Array processors with specialized processing elements, like
NAMs, have ever been proposed. Insert contents. DOCUMENT
IN PDF
Unpublished Work (Tech Reports)
N Venkateswaran, R
Chidambareswaran, B Harish, Kolluru Arvind,Integrated
Process for Simulating the Retinal Pathway on the Special
Purpose Neuronal Assembly Model Array Processor , WTR-CH-03.
Abstract
This
paper presents the integration of existing mathematical
models of the photoreceptors, the horizontal, the bipolar
and the ganglion cells, for simulating the shape recognition
processes occurring along the retinal pathway. The simulation
of the retinal pathway further involves integration
of Biologically realistic models of synapses, Green’s
function based model of dendritic trees & Hodgkin
Huxley(HH) models of neuronal cells. The entire simulation
process has been carried out on the special purpose
Neuronal Assembly Model Array Processor(NAMAP) model.
Integrated simulation of the retina pathway involving
large number of neurons has not been carried out on
a special purpose array processor such as NAMAP.Two
case studies pertaining to shape recognition process
are presented to demonstrate the capability of NAMAP
to simulate the retinal pathway.
DOCUMENT
IN PDF
N Venkateswaran, R
Chidambareswaran, B Harish, Kolluru Arvind, DNA
Based Evolvable Instruction Set Architecture & Arithmetic
Unit, WTR-CH-04.
Abstract
The
paper proposes a novel DNA based concept towards the
natural evolution of instruction set architecture and
arithmetic unit. Silicon based conventional systems
have insignificant storage and huge hardware size when
compared with DNA based systems. The silicon systems
can never have natural evolution. Current intelligent
systems based on Artificial Intelligence concepts falsify
the basic law of evolution. Huge silicon systems are
designed and fabricated and treating them as idiots,
a learning process is evolved to train them. It is just
like allowing a baby to grow into an adult and then
trying to teach them ”ABCD”. On the other hand DNAAP
(DNA based Arithmetic Processing unit), naturally evolves
with time into a larger intelligent system as and when
it learns.
DOCUMENT
IN PDF
|