INTELLIGENCE ARTIFICIAL · Superficie parabólica Guía de ondas Superficie parabólica Eje apuntar. hacia el satélite Señal reflejada Subreflector Haces de señales F oco focus

AR

TIFI

CIA

L IN

TELL

IGE

NC

E

DEEP LEA

RN

ING

ARE GREAT!

“AI IS THE NEW ELECTRICITY” - Andrew Ng

BINARY CAT CLASIFICATION

1 / 0 CAT

X

1 / 0 NON-CAT

X

?X

PERCEPTRON

NEURONAL NETWORKS

MATHEMATICS, LINEAL ALGEBRA, STATISTICS && GRAPHS

VectorizationThe art to get rid of the for loops in the code

SIMDSingle InstructionMultiple Data

It is a technique used to achieve

parallelism at the data level.

BROADCASTINGExample 1

Example 2

NEURONAL NETWORK REPRESENTATION

Vectorizationexample

BROADCASTING

CPU VS GPU

Per

form

ance

Big neuronal network

Medium neuronal network

Small neuronal network

Traditional learning algorithms (SVM, logistic regression)

GRADIENT DESCENT

J(W,b)

f’(x) = 0

GRADIENT DESCENT

Resource: Dot CSV

GRADIENT DESCENT

Resource: Dot CSV

BINARY CAT CLASIFICATION

1 / 0 CAT

X

1 / 0 NON-CAT

X

?X

COMPUTER VISION

“Unroll”

“Reshape”

VECTORIZATION

CONVOLUTIONAL NEURONAL NETWORKS

COMPUTER VISION

RESOURCE: Science - Daniel Clery

INTERNET INDUSTRIAL REVOLUTION

THINGS INDUSTRIAL REVOLUTION