Thanks to visit codestin.com
Credit goes to www.scribd.com

Scribd
Upload
23 views13 pages

Deep Learning Demo

The document discusses various aspects of machine learning (ML) and deep learning (DL), highlighting the differences in feature engineering and model building between the two. It covers artificial neural networks, activation functions, best practices for model building, and steps to create a DL model, including network architecture and training execution. Additionally, it touches on image processing, convolutional neural networks, and approaches like transfer learning and data augmentation.

Uploaded by

vikas
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as XLSX, PDF, TXT or read online on Scribd
23 views13 pages

Deep Learning Demo

The document discusses various aspects of machine learning (ML) and deep learning (DL), highlighting the differences in feature engineering and model building between the two. It covers artificial neural networks, activation functions, best practices for model building, and steps to create a DL model, including network architecture and training execution. Additionally, it touches on image processing, convolutional neural networks, and approaches like transfer learning and data augmentation.

Uploaded by

vikas
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as XLSX, PDF, TXT or read online on Scribd
You are on page 1/ 13

ML Feature engineering Manual

Model building Automated

DL Feature engineering Automated


Model building Automated

Artificial Neural Networks

Neural networks are non-linear machine learning models that

Mimic the human decisions

BNN - Biological neural nets


Feed Forward Neural Network / Sequential Neural Network:

x1

x2
N1
x3

x4
N2
x4

x6
N3
x7

x8
N4
x9

x10

Activation functions:

An Activation Function decides whether a neuron should be activated or not. This m


it adds non-linearity to the model

Best Practices:
Regression Model
Hidden layers = RELU

Binary Classification
Hidden layers = RELU

Multinomial classificaiton
Hidden layers = RELU

Steps to build DL model:


Network Architecturer
Number of hidden layers
Number of nuerons in each layer
Activation for each layer

Compilation
Loss Function
Optimizer

Execution- Training
Number of iterations = epochs
Batch size
linear machine learning models that can be used for both supervised and unsupervised tasks.
N5

O1

N6

on should be activated or not. This means that it will decide whether the neuron’s input to the network is important or not
Output layer = No activation/Linear

Output layer = Sigmoid

Output layer = softmax


erons in each layer

ations = epochs
mportant or not
Computer vision

Input output
image Image classficiation
video Object detection
Object localization
Object segmentation

What Is Image Processing?

Image processing is the process of transforming an image into a digital form

Colored grey colored image


RGB
225*225*3 225*225

Deep Learning Approaches


1. Build the model from scratch
High accuracy Lots of data, lots computational res

2. Transfer learning (pre-defined models on some data) - you will fine t


Less accuracy, Less of amount of data, less compu
Alex Net
ZF-Net
VGG Net
Goolge net
Resnet

Lack of data?

Data Augumentation:
getting more data

Convolution Neural Networks


Convolution Image * FeatureMap (matrix)

Input 3*3
10 5 6 3 0.3
68 230 98 176 0.5
13 16 107 177 0.2
28 124 78 151
34 91 32 244
0.3
0.5
0.2

Feature map
Feature detector
Kernal
Filter
mask
neuron
convolution matrxi
Image classficiation
Object detection
Object localization
Object segmentation

mage into a digital form and performing certain operations to get some useful information from it.

ta, lots computational resources

ome data) - you will fine tune the model parameters using existing data - build prediction
mount of data, less computational resources
FeatureMap (matrix) features

sharpen Hidden
0.1 0.2
0.2 -0.1 N1 N2
0.1 0.1 N3 N4

edge detector
0.1 0.2
0.2 -0.1
0.1 0.1

Feature map
Feature detector

convolution matrxi

You might also like