(or personality) and external stimulus, get some output. (respond or ignore) for ... Synaptic connections between the two types of neurons. â âThresholdâ of the ...
By Milind Zirpe Project Guides: Dr. Rene Doursat Dr. Phil Goodman University of Nevada, Reno Computer Science Department
Overview • • • • • •
Background The Problem Methodology Implementation Results Next Steps
Background • Traditional Neural Network are good for only what you train them to do. • They are vulnerable to noisy data and parameters. • Using a combination of signal patterns, inherent behavior (or personality) and external stimulus, get some output (respond or ignore) for the external stimulus. • The model is analogous to real life brain cell functioning.
The Problem • Modeling Neocortical Locks: – Modeling of Neocortical Locks from the Lock and Key model. – Lock is the network of neurons, consisting of excitatory and inhibitory cells. – Key is the external stimulus, ideally consisting of random audio or video stimulus from the environment. – Objective is to find a lock which resonates quite perfectly with a given key (i.e. has same phase as key, when the key is applied). – Later on, let the network learn based on reward and punishment scheme. (e.g. Hebbian learning).
Methodology • Model a network of neurons which will have some inherent behavior (Lock). • Introduce a signal as an external stimulus to the postsynaptic cells (Key). • Idea is to adjust the strength of the synaptic conductance (G) from pre-synaptic cells to the post-synaptic cell to get the lock to resonate with the key.
Methodology • Other parameters to model are: – – – – – – – –
Short and long term synaptic dynamics. Number of neurons in the network. The inherent behavior of the neurons. Synaptic connections between the two types of neurons. “Threshold” of the “Compartment”. Spike shape, if needed. “Absolute use” value of the synapse in case of learning. Type of learning, duration, FSV (Frequency of Sampling Value), and various other parameters.
Methodology • Results: – Expected result is to obtain reproducible, unique response from the lock for a particular given key pattern. – Further part of project is to develop a network of neurons which learns as it experiences external stimulus.
Implementation • Implementation modules: – The neural network model (*.in files) generating the behavior of the lock. – Programs in Matlab for analyzing the report files from NCS and for interpreting the results.
Implementation • Software and Languages: – Neo Cortical Simulator developed in GBCL. Accepts the *.in files as an input. – Matlab 7.0, mostly for analyzing the results. – Pre-developed Python scripts for automating the generation of *.in file.
• Environment: – Windows XP and Linux (NCS clusters).
16 Cell Neural Network Internal behavior Sine 00 out of phase Sine 1200 out of phase Sine 2400 out of phase
External Stimulus
Results
Result of the previous slide’s non-spiking network for combinations of Pre-Syn1, 2 and 3 (100%) and single, shifted Poisson signal applied to Pre-Syn 4 cells applied to Post-Syn 1 to 12 cells.
Results Case a:
Single spike at 1 Hz
Case b:
Single spike at various phase differences of - 0, 45, 90, 135, 180 and 240
Case c:
Multiple spike at 1 Hz
Case d:
Multiple spike at various phase differences of - 0, 45, 90, 135, 180 and 240
Conductance (G) vs Power (1 Hz) component 0.040
Power of 1 Hz Component
0.035 0.030 0.025
Power Case a Power Case b
0.020
Power Case c
0.015
Power Case d
0.010 0.005 0.000 0.000
0.002
0.004
0.006
0.008
0.010
0.012
Conductance (G)
Synaptic Conductance (G) vs. Power of 1 Hz component in Avg. Global potential.
Next Steps • Matlab analysis programs development is almost done. • Simulate the network with different parameters to determine the optimum parameter values. • Change the external stimulus to other random type of signals. • Increase the number of cells and use of python for autogeneration of brain model file. • Self-tuning (learning) network (lock) as per the stimulus (key).
Artificial Neural Network. (*.in files ... Lock is the network of neurons, consisting of excitatory and ... Spiking if membrane potential is above threshold (Excitatory.
Apr 11, 2006 - Seminaire ADSTIC, 11 avril 2006. François Grimbert. Gray Matter / White Matter. • Gray matter is a folded sheet of neurons. (thickness: 2-4 mm).
mal joysticks meaning that four players could take part in one ... Using weapons, explosives and their animal friends, these Warriors with their METHODS OF ...
who all have special moves, were only available as are both more graphical ... is, can this product are unleashed on the play not pullanypunches when The ...
Jul 15, 2007 - tudes measured when observers perceived the waves (Table 1). Although it may .... that the curves are periodic because each trial began 9 s ...
medial surface. The method applied on MRI data, is based ... INTRODUCTION. Human cortical ..... Proceedings of Medical Imaging, Image Processing, 1996,.
Feb 17, 1997 - temperature rises above 150 C, the device will go into thermal shutdown. The junction-to-ambient thermal resistance (ThetaJA) is 39 C/W.
Apr 25, 2007 - Goodman, P. H. & Doursat, R. - Cortical Mesocircuit Dynamics. 15. ⢠Theories populating the mesolevel: a zoology of STPs. â synfire chains & ...
