Sistemas de software basados en Redes Neuronales
(Inteligencia Artificial Avanzada) y el Estado del Arte de la
Tecnología ...
Inteligencia Artificial Avanzada
(Redes Neuronales con 3 ejes de visión)
1 Arquitectura , Algoritmos y Aplicaciones.
2 H...
“The Gene is by far the most sophisticated program around.”
- Bill Gates, Business Week, June 27, 1994
Genes , Atoms , Cel...
Road Map de Tecnología y la evolución
de la PC hasta la Web 4.0
Semántica de las conexiones sociales
The Semantic Web (AI)
La evolución hacia la Inteligencia Artificial
EVOLUCION HACIA LA INTELIGENCIA ARTIFICIAL
 Aumento exponencial de la capacidad computacional de las PC´s.
 Capacidad de...
“BACK TO THE FUTURE” 2010 -2050
CONVERGENCIA DE LAS GRIN TECHNOLOGIES
G
R
N
N
I
ARQUITECTURAS ALGORITMOS Y
APLICACIONES
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez Motte Ph.D...
Qué es una Red Neuronal ?
Una Red Neuronal Artificial es un sistema de procesamiento de información o
Processing Element (...
Qué es una Red Neuronal ?
Cada conexión (LINK) tiene un peso asociado, el cual, en una red neuronal típica
multiplica la s...
Red Neuronal Biológica
Soma Soma
Synapse
Synapse
Dendrites
Axon
Synapse
Dendrites
Axon
NEUROMORPHIC TECHNOLOGIES in CO BRA...
Estructura de una neurona biológica
1 Nucleo; 2 Nucleoide; 3 Soma; 6 Membrana Celular; 7 Región Sináptica , 8 Axon; 11 Den...
“Building block” de una Neurona Artificial
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez Motte P...
Anología entre Redes Neuronales Biológicas y
Redes Neuronales Artificiales
Biological Neural Network Artificial Neural Net...
Recursos disponibles en
computadoras y cerebro humano
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jime...
• Aplicaciones
- Clasificación (Classification)
- (Clustering)
- Aproximación de funciones (Function approximation)
- Pred...
Perceptron
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
Redes Neuronales en Reconocimiento y Clasificación de de Patrones
(Pattern Recognition/Pattern Recognition )
NEUROMORPHIC ...
Reconocimiento de caracteres
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez Motte Ph.D.EE (c), M....
Reconocimiento de caracteres
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez Motte Ph.D.EE (c), M....
Perceptron
(Pattern Recognition/Pattern Classification)
M
S
D
N
.
MSDN…
…………
…………
………...
MSDN…
…………
…………
………...
REDES NEURONALES BASADAS EN
COMPETICION (Neural Computation)
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernan...
Self Organizing Map (SOM)
Neural Network
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez Motte Ph....
Self organizing maps
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B...
Mapas auto organizantes del cuerpo humano
(Somatotensory cortex)
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fe...
Topología de mapeo SOM
Dos capas: capa de entrada y la capa de salida (mapa o COMPUTATIONAL LAYER).
Capas de entrada y de ...
SELF ORGANIZING NEURAL NETWORK
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez Motte Ph.D.EE (c), ...
4x4 SOM Network
(4 nodes down, 4 nodes across)
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez Mot...
Self-Organizing Maps (Kohonen Maps)
Estructuras de salida típicas: (“Computational layer”)
Uni-dimensional
(completamente ...
Mapas Auto Organizantes
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E...
Self Organizing Neural Network
Kohonen Learning
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez Mo...
“Self Organizing Traffic Lights”
La evolución de la inteligencia artificial la concepción moderna es que la ciudad es la r...
Arreglo lineal de “cluster” units
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez Motte Ph.D.EE (c...
Arquitectura
• Grilla Rectangular
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez Motte Ph.D.EE (c...
Arquitectura
• Grilla hexagonal
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez Motte Ph.D.EE (c),...
Topological SOM (Matlab)
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez Motte Ph.D.EE (c), M.S.E....
Radius 1
Radius 2
Winning unit, only one to update if radius=0
Units to be updated when radius=1
Units to be updated when ...
Entrenamiento no supervizado en SOMs
Ejemplo: Aprendiendo la representación unidimensional de un espacio de entrada
(“inpu...
Entrenamiento no supervizado en SOMs
Ejemplo : Aprendiendo una representación bi-dimensional de un espacio de entrada
(“in...
Aplicaciones
• WEBSOM: Organización de una colección masiva de
documentos.
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH I...
SOFM
“Phoneme Recognition”
 Phonotopic maps
 Recognition result for “humppila”
NEUROMORPHIC TECHNOLOGIES in CO BRANDING ...
Aplicaciones
• Clasificando la pobreza mundial
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez Mot...
Neural Networks in Agriculture
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez Motte Ph.D.EE (c), ...
Entrenamiento no supervizado en SOMs
Ejemplo : Aprendiendo una representación bi-dimensional de un espacio de entrada
(“in...
Self Organizing Map SOM
“Elastic virtual network”
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez ...
Self Organizing Map SOM
“Elastic virtual network”
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez ...
Entrenamiento no supervizado en SOMs
Ejemplo:
Aprendiendo el mapeo bi-
dimensional de texturas de
imágenes.
NEUROMORPHIC T...
Autonomous Land Vehicle in a Neural Network
ALVINN system
• Input: 30 x 32 grid of
pixel intensities (960
nodes)
• 4 hidde...
Autonomous Land Vehicle in a Neural Network
ALVINN system
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando ...
Wireless Localization Using Self-Organizing Maps
– Utiliza una Red Neuronal Auto Organizante para determinar las
posicione...
Multiple Self-Organizing Maps
for Intruder Detection
Euclidean Norm NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS...
Self-Organizing Neighborhood Wireless Mesh
Networks
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimene...
Millions of Neural Nets covering the world over the Internet
Each TCIP / Node becomes a Neuron so that a SuperNet is built...
Sistema de control
Neuro Difuso Genético
Modelo de Sistema o
Proceso (Red Neuronal)
Test
Driver
Control Difuso
Fuzzy Infer...
Self Organizing Map SOM
“Semantic mapping”
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez Motte P...
SOM “building block” CAD/CAM
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez Motte Ph.D.EE (c), M....
N13(1) N13(2)
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
Self organized semantic map: Word categories
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez Motte...
Cronología de SOM´s
• Kohonen (1984). Speech recognition - a map of
phonemes in the Finish language
• Optical character re...
“Skeletonization of images”
• Esqueletización de imágenes ruidosas.
• Aplicación del MST SOM: robustez con respecto al
rui...
CELULLAR NEURAL NETWORK CNN ARCHITECTURE WITH 3X3
NEIGHBORHOOD CONNECTIVITY
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH ...
RED NEURONAL DE MULTIPLES CAPAS
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez Motte Ph.D.EE (c),...
Backpropagation Neural
Network
(Red Neuronal de Retropropagación)
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
F...
Three-Layer Network
321
SSSR Number of neurons in each layer:
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fe...
Algoritmo de aprendizaje
Backpropagation
Las siguientes diapositivas describe el proceso de enseñanza del algoritmo de apr...
Algoritmo de aprendizaje
Backpropagation
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez Motte Ph....
Algoritmo de aprendizaje
Backpropagation
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez Motte Ph....
Algoritmo de aprendizaje
Backpropagation
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez Motte Ph....
Algoritmo de aprendizaje
Backpropagation
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez Motte Ph....
Algoritmo de aprendizaje
Backpropagation
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez Motte Ph....
Algoritmo de aprendizaje
Backpropagation
Propagación de las señales a través del “output layer”
NEUROMORPHIC TECHNOLOGIES ...
Algoritmo de aprendizaje
Backpropagation
En el paso siguiente del algoritmo se obtiene la señal de salida de la red y se
c...
Algoritmo de aprendizaje
Backpropagation
La idea es la de propagar la señal de error d (calculado en la etapa de
enseñanza...
Algoritmo de aprendizaje
Backpropagation
La idea es la de propagar la señal de error d (calculado en la etapa de
enseñanza...
Algoritmo de aprendizaje
Backpropagation
Los pesos wmn «coeficientes utilizados para propagar errores
hacia atrás son igua...
Algoritmo de aprendizaje
Backpropagation
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez Motte Ph....
Algoritmo de aprendizaje
Backpropagation
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez Motte Ph....
Algoritmo de aprendizaje
Backpropagation
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez Motte Ph....
REDES NEURONALES
HARDWARE - SOFTWARE
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez Motte Ph.D.EE...
Redes Neuronales en Procesamiento Digital de Señales
DSP
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando J...
Redes Neuronales en Ingeniería de Control
(Neural Network Adaptive Control Systems)
NEUROMORPHIC TECHNOLOGIES in CO BRANDI...
Redes Neuronales en Ingeniería de Control (Modelamiento Inverso)
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fe...
Redes Neuronales en Ingeniería de Control (Navigation & Collition
Avoidance)
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH...
NEURO COPTER 1.0 DESIGN
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E...
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
Redes Neuronales en Ingeniería de Control (Pathfinder + Sojourner
en Marte)
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH ...
Redes Neuronales en Ingeniería de Control
(Control No Lineal de Manipuladores de Robot)
NEUROMORPHIC TECHNOLOGIES in CO BR...
Redes Neuronales en Ingeniería de Control Inteligente
(“Upper backer Track Neural Network”)
NEUROMORPHIC TECHNOLOGIES in C...
Redes Neuronales en Reconocimiento de lenguage y voz
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimen...
Redes Neuronales en Medicina y Bio Ingeniería
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez Mott...
Redes Neuronales en Interfase Hombre Máquina
“Human Machine Interfase/Bionics”
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WI...
Neural Network-Based Face Detection. (Face Pattern Recognition).
A Convolutional Neural Network Hand Tracker. (Gesture Rec...
Microsoft Neural Network Algorithm
SQL Server
In SQL Server Analysis Services, the Microsoft Neural Network
algorithm comb...
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
NEURO SOLUTION...
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
CNAPS Chip
(Paralell processor for Pattern Recognition, Image Processing and Neural networks)
64 x 64 x 1 in 8 µs
(8 bit i...
……..
……..
64
128
4
Execution time : ~500 ns
Weights coded in 16 bits
States coded in 8 bits
with data arriving every BC=25...
Very fast architecture
• Matrix of n*m matrix
elements
• Control unit
• I/O module
• TanH are stored in
LUTs
• 1 matrix ro...
Three-Layer Network
321
SSSR Number of neurons in each layer:
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fe...
CELULAR NEURAL NETWORK CNN ARCHITECTURE WITH 3X3
NEIGHBORHOOD CONNECTIVITY
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH I...
Matlab + Simulink Neural Network
Toolbox
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez Motte Ph....
Celullar Neural Network
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E...
Boltzman Machine Neural Network
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez Motte Ph.D.EE (c),...
Radial Basis Function RBF
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez Motte Ph.D.EE (c), M.S.E...
Neocognitron
GLOBAL NAVIGATIONAL SATELLITE SYSTEM GNSS
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez Motte Ph...
INTELIGENCIA ARTIFICIAL /NEURO
CIBERNETICA
La emergencia de la Biónica
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/US...
“Dinámica Estocástica del Complejo Neuronal”
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez Motte...
Representación de Red de la actividad Neurocelular
Aspectos estocásticos de un complejo neuronal infectado por ruido.
NEUR...
Teoría de Campo Neuronal
Control del complejo organizativo (minimización de entropía)
NEUROMORPHIC TECHNOLOGIES in CO BRAN...
NEUROMORPHIC ENGINEERING
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez Motte Ph.D.EE (c), M.S.E....
Analog Neural Network Core (ANNCORE)
containing 128k synapses and 512 membrane circuits which can be grouped
together to f...
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
Próximos Chips...
Bionic Eye
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
Bionic ear
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
THE MESSENGER
(Neural Network Somatosensory maps)
NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS
Fernando Jimenez ...
Año 2030-2050
El cerebro Artificial: Positronic Brain
Red Neuronal Artificial para emular un Cerebro Humanoide
El Humanoid...
Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
Director/Technology Product Manager/Consultor
Technology Director CTO...
Próxima SlideShare
Cargando en…5
×

Advanced Artificial Intelligence Microsoft Technet MSDN summit 2013

505 visualizaciones

Publicado el

My conference in Advanced Artificial Intelligence presented at Microsoft Technet MSDN summit 2013.

Publicado en: Tecnología
0 comentarios
0 recomendaciones
Estadísticas
Notas
  • Sé el primero en comentar

  • Sé el primero en recomendar esto

Sin descargas
Visualizaciones
Visualizaciones totales
505
En SlideShare
0
De insertados
0
Número de insertados
2
Acciones
Compartido
0
Descargas
9
Comentarios
0
Recomendaciones
0
Insertados 0
No insertados

No hay notas en la diapositiva.

Advanced Artificial Intelligence Microsoft Technet MSDN summit 2013

  1. 1. Sistemas de software basados en Redes Neuronales (Inteligencia Artificial Avanzada) y el Estado del Arte de la Tecnología Global NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  2. 2. Inteligencia Artificial Avanzada (Redes Neuronales con 3 ejes de visión) 1 Arquitectura , Algoritmos y Aplicaciones. 2 Hardware + Software. 3 Neurocibernética y el advenimiento de la Biónica. NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  3. 3. “The Gene is by far the most sophisticated program around.” - Bill Gates, Business Week, June 27, 1994 Genes , Atoms , Cells and Neurons are the Building Blocks of the XXI Century - Fernando Jiménez M, México 2011
  4. 4. Road Map de Tecnología y la evolución de la PC hasta la Web 4.0 Semántica de las conexiones sociales
  5. 5. The Semantic Web (AI)
  6. 6. La evolución hacia la Inteligencia Artificial
  7. 7. EVOLUCION HACIA LA INTELIGENCIA ARTIFICIAL  Aumento exponencial de la capacidad computacional de las PC´s.  Capacidad de manejo computacional masivo, paralelo y distribuído.  Escalamiento de la ciencia a la nube. (Scaling Science to the Cloud).  Advenimiento de Sistemas Neuro y biomiméticos.  Emergencia de la Neurocomputación o Neurocomputing.  Very large Scale Integrated Systems VLSI.  Convergencia de las GRIN Technologies. NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  8. 8. “BACK TO THE FUTURE” 2010 -2050 CONVERGENCIA DE LAS GRIN TECHNOLOGIES G R N N I
  9. 9. ARQUITECTURAS ALGORITMOS Y APLICACIONES NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  10. 10. Qué es una Red Neuronal ? Una Red Neuronal Artificial es un sistema de procesamiento de información o Processing Element (PE) en el caso de una neurona, el cual tiene características de performance en común con redes neuronales biológicas. Han sido desarrolladas a través de generalizaciones de modelos matemáticos (Matemática Neurobiológica) basados en cognición humana o biología neuronal y basadas en asunciones como: El procesamiento de información ocurre en muchos elementos simples llamados neuronas. Las señales son enviadas entre neuronas a través de conexiones. (LINKS) NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  11. 11. Qué es una Red Neuronal ? Cada conexión (LINK) tiene un peso asociado, el cual, en una red neuronal típica multiplica la señal transmitida. Cada neurona aplica una función de activación (usualmente no lineal) a su entrada neta (net_input) para determinar la señal de salida. NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  12. 12. Red Neuronal Biológica Soma Soma Synapse Synapse Dendrites Axon Synapse Dendrites Axon NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  13. 13. Estructura de una neurona biológica 1 Nucleo; 2 Nucleoide; 3 Soma; 6 Membrana Celular; 7 Región Sináptica , 8 Axon; 11 Dendritas NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  14. 14. “Building block” de una Neurona Artificial NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  15. 15. Anología entre Redes Neuronales Biológicas y Redes Neuronales Artificiales Biological Neural Network Artificial Neural Network Soma Dendrite Axon Synapse Neuron Input Output Weight NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  16. 16. Recursos disponibles en computadoras y cerebro humano NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  17. 17. • Aplicaciones - Clasificación (Classification) - (Clustering) - Aproximación de funciones (Function approximation) - Predicción (Prediction) - Toma de decisiones en situaciones con restricción (constrained optimization) • Tipo de conexión - Estática / Static (feedforward) - Dinámica / Dynamic (feedback) • Topología - Una sóla capa (Single layer) - Multicapa (Multilayer) - Recurrente (Recurrent) - Auto Organizacional (Self-organizing Neural Net) • Método de aprendizaje - Supervizada (Supervised) - No Supervizada (Unsupervised) Tipos de Redes Neuronales NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  18. 18. Perceptron NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  19. 19. Redes Neuronales en Reconocimiento y Clasificación de de Patrones (Pattern Recognition/Pattern Recognition ) NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  20. 20. Reconocimiento de caracteres NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  21. 21. Reconocimiento de caracteres NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  22. 22. Perceptron (Pattern Recognition/Pattern Classification) M S D N . MSDN… ………… ………… ………... MSDN… ………… ………… ………...
  23. 23. REDES NEURONALES BASADAS EN COMPETICION (Neural Computation) NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  24. 24. Self Organizing Map (SOM) Neural Network NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  25. 25. Self organizing maps NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  26. 26. Mapas auto organizantes del cuerpo humano (Somatotensory cortex) NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  27. 27. Topología de mapeo SOM Dos capas: capa de entrada y la capa de salida (mapa o COMPUTATIONAL LAYER). Capas de entrada y de salida están completamente conectadas. Las neuronas de salida están interconectadas dentro de un vecindario definido. Una topología (relación de vecindad) se define en la capa de salida. NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  28. 28. SELF ORGANIZING NEURAL NETWORK NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  29. 29. 4x4 SOM Network (4 nodes down, 4 nodes across) NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  30. 30. Self-Organizing Maps (Kohonen Maps) Estructuras de salida típicas: (“Computational layer”) Uni-dimensional (completamente interconectada para determinar la unidad ganadora. (“winner”) Bi-dimensional i i Vecindario de la neurona i NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  31. 31. Mapas Auto Organizantes NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  32. 32. Self Organizing Neural Network Kohonen Learning NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  33. 33. “Self Organizing Traffic Lights” La evolución de la inteligencia artificial la concepción moderna es que la ciudad es la red neuronal. Las diferentes vías, avenidas y autopistas de una ciudad emulan a las dendritas y axons de la estructura intracelular y extracelular de una neurona o red neuronal biológica. En la red víal de una ciudad, las intersecciones definen entonces neuronas y la red neuronal víal aprende del flujo de tráfico de vehículos. NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  34. 34. Arreglo lineal de “cluster” units NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  35. 35. Arquitectura • Grilla Rectangular NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  36. 36. Arquitectura • Grilla hexagonal NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  37. 37. Topological SOM (Matlab) NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  38. 38. Radius 1 Radius 2 Winning unit, only one to update if radius=0 Units to be updated when radius=1 Units to be updated when radius=2 La topología determina que unidad cae dentro del radio R de una unidad o neurona ganadora. La unidad ganadora adaptará sus pesos en la cual acercará a ese “cluster” hacia el vector de entrenamiento. Actualización de pesos dentro de un radio de acción NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  39. 39. Entrenamiento no supervizado en SOMs Ejemplo: Aprendiendo la representación unidimensional de un espacio de entrada (“input space“) bi dimensional. 0 25000 20 100 1000 10000 NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  40. 40. Entrenamiento no supervizado en SOMs Ejemplo : Aprendiendo una representación bi-dimensional de un espacio de entrada (“input space“) bi dimensional. NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  41. 41. Aplicaciones • WEBSOM: Organización de una colección masiva de documentos. NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  42. 42. SOFM “Phoneme Recognition”  Phonotopic maps  Recognition result for “humppila” NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  43. 43. Aplicaciones • Clasificando la pobreza mundial NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  44. 44. Neural Networks in Agriculture NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  45. 45. Entrenamiento no supervizado en SOMs Ejemplo : Aprendiendo una representación bi-dimensional de un espacio de entrada (“input space“) bi dimensional. NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  46. 46. Self Organizing Map SOM “Elastic virtual network” NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  47. 47. Self Organizing Map SOM “Elastic virtual network” NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  48. 48. Entrenamiento no supervizado en SOMs Ejemplo: Aprendiendo el mapeo bi- dimensional de texturas de imágenes. NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  49. 49. Autonomous Land Vehicle in a Neural Network ALVINN system • Input: 30 x 32 grid of pixel intensities (960 nodes) • 4 hidden units • Output: direction of steering (30 units) • Training: 5 min. of human driving • Test: up to 70 miles for distances of 90 miles on public highway. (driving in the left lane with other vehicles present) NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  50. 50. Autonomous Land Vehicle in a Neural Network ALVINN system NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  51. 51. Wireless Localization Using Self-Organizing Maps – Utiliza una Red Neuronal Auto Organizante para determinar las posiciones de los nodos de una red inalámbrica. NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  52. 52. Multiple Self-Organizing Maps for Intruder Detection Euclidean Norm NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  53. 53. Self-Organizing Neighborhood Wireless Mesh Networks NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  54. 54. Millions of Neural Nets covering the world over the Internet Each TCIP / Node becomes a Neuron so that a SuperNet is built over the Internet. The whole network will behave as a Synthetic Intelect, NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  55. 55. Sistema de control Neuro Difuso Genético Modelo de Sistema o Proceso (Red Neuronal) Test Driver Control Difuso Fuzzy Inference Engine Pesos Rules Relaciones difusas Decodificación cromosómica Algoritmos genéticos Control de temperatura Modelo del Sistema Test function Función evaluación GRIN NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  56. 56. Self Organizing Map SOM “Semantic mapping” NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  57. 57. SOM “building block” CAD/CAM NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  58. 58. N13(1) N13(2)
  59. 59. NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  60. 60. Self organized semantic map: Word categories NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  61. 61. Cronología de SOM´s • Kohonen (1984). Speech recognition - a map of phonemes in the Finish language • Optical character recognition - clustering of letters of different fonts • Angeliol etal (1988) – travelling salesman problem (an optimization problem) • Kohonen (1990) – learning vector quantization (pattern classification problem) • Ritter & Kohonen (1989) – semantic maps NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  62. 62. “Skeletonization of images” • Esqueletización de imágenes ruidosas. • Aplicación del MST SOM: robustez con respecto al ruido. NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  63. 63. CELULLAR NEURAL NETWORK CNN ARCHITECTURE WITH 3X3 NEIGHBORHOOD CONNECTIVITY NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  64. 64. RED NEURONAL DE MULTIPLES CAPAS NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  65. 65. Backpropagation Neural Network (Red Neuronal de Retropropagación) NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  66. 66. Three-Layer Network 321 SSSR Number of neurons in each layer: NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  67. 67. Algoritmo de aprendizaje Backpropagation Las siguientes diapositivas describe el proceso de enseñanza del algoritmo de aprendizaje para redes neuronales de retropropagación de múltiples capas. Para ilustrar este proceso, la red neural que utilizaremos consiste de tres capas con dos entradas y una salida como se muestra en la siguiente arquitectura: NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  68. 68. Algoritmo de aprendizaje Backpropagation NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  69. 69. Algoritmo de aprendizaje Backpropagation NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  70. 70. Algoritmo de aprendizaje Backpropagation NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  71. 71. Algoritmo de aprendizaje Backpropagation NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  72. 72. Algoritmo de aprendizaje Backpropagation NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  73. 73. Algoritmo de aprendizaje Backpropagation Propagación de las señales a través del “output layer” NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  74. 74. Algoritmo de aprendizaje Backpropagation En el paso siguiente del algoritmo se obtiene la señal de salida de la red y se compara con el valor de salida deseado (valor objetivo o “target La diferencia se llama error de la señal d de la capa de salida de la red neuronal. NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  75. 75. Algoritmo de aprendizaje Backpropagation La idea es la de propagar la señal de error d (calculado en la etapa de enseñanza) de nuevo a todas las neuronas, cuyas señales de salida fueron entrada para dicha neurona. NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  76. 76. Algoritmo de aprendizaje Backpropagation La idea es la de propagar la señal de error d (calculado en la etapa de enseñanza) de nuevo a todas las neuronas, cuyas señales de salida fueron entrada para dicha neurona. NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  77. 77. Algoritmo de aprendizaje Backpropagation Los pesos wmn «coeficientes utilizados para propagar errores hacia atrás son iguales a utilizar, a los que fueron utilizados para calcular el valor de salida. La dirección del flujo de datos cambia (se propagan las señales de salida a entradas de una después de la otra). Esta técnica se utiliza para todas las capas de red. NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  78. 78. Algoritmo de aprendizaje Backpropagation NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  79. 79. Algoritmo de aprendizaje Backpropagation NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  80. 80. Algoritmo de aprendizaje Backpropagation NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  81. 81. REDES NEURONALES HARDWARE - SOFTWARE NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  82. 82. Redes Neuronales en Procesamiento Digital de Señales DSP NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  83. 83. Redes Neuronales en Ingeniería de Control (Neural Network Adaptive Control Systems) NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  84. 84. Redes Neuronales en Ingeniería de Control (Modelamiento Inverso) NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  85. 85. Redes Neuronales en Ingeniería de Control (Navigation & Collition Avoidance) NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  86. 86. NEURO COPTER 1.0 DESIGN NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  87. 87. NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  88. 88. Redes Neuronales en Ingeniería de Control (Pathfinder + Sojourner en Marte) NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  89. 89. Redes Neuronales en Ingeniería de Control (Control No Lineal de Manipuladores de Robot) NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  90. 90. Redes Neuronales en Ingeniería de Control Inteligente (“Upper backer Track Neural Network”) NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  91. 91. Redes Neuronales en Reconocimiento de lenguage y voz NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  92. 92. Redes Neuronales en Medicina y Bio Ingeniería NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  93. 93. Redes Neuronales en Interfase Hombre Máquina “Human Machine Interfase/Bionics” NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  94. 94. Neural Network-Based Face Detection. (Face Pattern Recognition). A Convolutional Neural Network Hand Tracker. (Gesture Recognition). Neural Networks for speech and image processing. Speech recognition with Neural Networks. Conversational Speech Transcription Using Context-Dependent Deep Neural Networks. Speech recognition breakthrough via machine translation that converted spoken English words into computer-generated Chinese language. (Human/natural machine interfase HMI). NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  95. 95. Microsoft Neural Network Algorithm SQL Server In SQL Server Analysis Services, the Microsoft Neural Network algorithm combines each possible state of the input attribute with each possible state of the predictable attribute, and uses the training data to calculate probabilities. You can later use these probabilities for classification or regression, and to predict an outcome of the predicted attribute, based on the input attributes. A mining model that is constructed with the Microsoft Neural Network algorithm can contain multiple networks, depending on the number of columns that are used for both input and prediction, or that are used only for prediction. The number of networks that a single mining model contains depends on the number of states that are contained by the input columns and predictable columns that the mining model uses. NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  96. 96. NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E NEURO SOLUTIONS FOR EXCEL
  97. 97. NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  98. 98. CNAPS Chip (Paralell processor for Pattern Recognition, Image Processing and Neural networks) 64 x 64 x 1 in 8 µs (8 bit inputs, 16 bit weights, CNAPS 1064 chip Adaptive Solutions, Oregon NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  99. 99. …….. …….. 64 128 4 Execution time : ~500 ns Weights coded in 16 bits States coded in 8 bits with data arriving every BC=25ns Electrons, tau, hadrons, jets Neural Network architecture NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  100. 100. Very fast architecture • Matrix of n*m matrix elements • Control unit • I/O module • TanH are stored in LUTs • 1 matrix row computes a neuron • The results is back- propagated to calculate the output layer TanH PE 256 PEs for a 128x64x4 network PE PEPE PE PE PEPE PE PE PEPE PE PE PEPE TanH TanH TanH ACC ACC ACC ACC I/O module Control unit
  101. 101. Three-Layer Network 321 SSSR Number of neurons in each layer: NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  102. 102. CELULAR NEURAL NETWORK CNN ARCHITECTURE WITH 3X3 NEIGHBORHOOD CONNECTIVITY NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  103. 103. Matlab + Simulink Neural Network Toolbox NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  104. 104. Celullar Neural Network NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  105. 105. Boltzman Machine Neural Network NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  106. 106. Radial Basis Function RBF NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  107. 107. Neocognitron
  108. 108. GLOBAL NAVIGATIONAL SATELLITE SYSTEM GNSS NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  109. 109. INTELIGENCIA ARTIFICIAL /NEURO CIBERNETICA La emergencia de la Biónica NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  110. 110. “Dinámica Estocástica del Complejo Neuronal” NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  111. 111. Representación de Red de la actividad Neurocelular Aspectos estocásticos de un complejo neuronal infectado por ruido. NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  112. 112. Teoría de Campo Neuronal Control del complejo organizativo (minimización de entropía) NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  113. 113. NEUROMORPHIC ENGINEERING NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  114. 114. Analog Neural Network Core (ANNCORE) containing 128k synapses and 512 membrane circuits which can be grouped together to form neurons with up to 16k synapses. NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  115. 115. NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E Próximos Chips Neuromórficos a ser lanzados por IBM e INTEL
  116. 116. Bionic Eye NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  117. 117. Bionic ear NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  118. 118. THE MESSENGER (Neural Network Somatosensory maps) NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  119. 119. Año 2030-2050 El cerebro Artificial: Positronic Brain Red Neuronal Artificial para emular un Cerebro Humanoide El Humanoide Data (“Star Treck”) tiene 100,000 terabytes de memoria (equiv a 100,000,000 de 1-GB hard drives). Tiene una capacidad de almacenamiento de 800 quadrillones de bits (100 quadrillion bytes). Data procesa 60 trillones de operaciones x segundo . Si uno compara su capacidad de almacenamiento de 100,000 terabytes con la los ordenadores actuales, podrìa decirse que la del ser humano es equivalente a unos 3 Terabytes, es decir Data contendrìa 260,000 Cerebros humanos NEUROMORPHIC TECHNOLOGIES in CO BRANDING WITH IT/USERS Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E
  120. 120. Fernando Jimenez Motte Ph.D.EE (c), M.S.E.E, B.S.E.E Director/Technology Product Manager/Consultor Technology Director CTO IT/USERS NEUROMORPHIC TECHNOLOGIES Product Catalog Product Management Consulting and Training. Project Management Consulting and Training. Strategic Product Planing. Workshops and Tutorials on site and via On Line. Scientific Research, Projects and Teaching International Conferences. Nacional Conferences. maurice_2500@hotmail.com

×