Вы находитесь на странице: 1из 25

INTELLIGENT TRAFFIC CONTROL IN

HETEROGENEOUS NETWORKS USING


DEEP LEARNING TECHNIQUES

Under the guidance of by


DR.L.RAJESH E.SARANYA
(2018610018)
M.E-COMMUNICATION
& NETWORKING
11/13/2019 1
CONTENT

• Objective
• Introduction
• Literature Survey
• Proposed work
• References

11/13/2019 2
OBJECTIVE

• To improve the better performance in the large scale


network in terms of packets loss rate and throughput

• To reduce the computational complexity

11/13/2019 3
INTRODUCTION

• Deep learning is a class of machine


learning algorithms that use multiple layers to
progressively extract higher level features from raw
input.

• It is a specific approach used for building and training


neural networks with multiple Hidden layers.

11/13/2019 4
BLOCK DIAGRAM

11/13/2019 5
HETEROGENEOUS NETWORK

• Heterogeneous network is a network for which an


area is covered simultaneously by cells of different
sizes like macro-cell, micro-cell, pico-cell, femto-cell

• These networks are multiservice, providing more


than one application or services

11/13/2019 6
LITERATURE SURVEY
Title Author Publication Inference

On Intelligent Zubair Md. IEEE Communications • Deep convolutional


Traffic Control For Fadlullah,Fengxiao Letters,vol.22,issue - neural network(CNN),
Large Scale Tang et.al 12,Journal Article, deep belief network
Heterogeneous Publisher: IEEE, 2018 •Avoid congestion and
Networks: A Value maximize the network
Matrix Based throughput
Deep Learning
Approach

The deep learning Nei kato , Kimihiro IEEE WIRELESS •Greedy layer-wise
vision for Mizutani et.al COMMUNICATION training method and
heterogeneous S, VOL. 24, NO. 3, Back propagation
network traffic PP. 146–153, 2017 algorithm
control: Proposal, •Signaling overhead is
challenges, and low
future perspective •Low average per hop
delay
11/13/2019 7
LITERATURE SURVEY
Title Author Publication Inference

On removing Takeru Inoue, IEEE Wireless •The method used


routing protocol Osamanu Akashi Communications, here is deep
from future et.al vol. 25, no. 1, pp. Convolutional
wireless networks : 154–160, February Neural Network
A real-time deep 2018 (CNN)
learning approach
for intelligent •Avoid congestion
traffic control maximize the
network throughput,
lower average delay
and packet loss rate.

•Action space to
become extremely
large and incurs a
high computational
cost.
11/13/2019 8
LITERATURE SURVEY
Title Author Publication Inference

Routing or Bomin Mao,Zubair IEEE Transactions on •Supervised deep


computing? the Md.Fadlullah et.al Computers, vol. PP, learning algorithm
paradigm shift no. 99, pp. 1–1,
towards intelligent 2017. •lower signaling
computer network Overhead
packet
transmission
based on deep
learning

11/13/2019 9
PROPOSED WORK

The proposed work mainly includes


 Deep CNN provide connection between temporal and spatial
features of input data
 Those features are extracted by convolutional layers
 Train and validate the modeled data
 Accuracy can be obtained and network will be predicted.
 It is implemented using python language and Tensorflow in
Colab Notebook

11/13/2019 10
WORK FLOW
DATA INPUT

TRAINING THE DATA

TESTING THE DATA

TRANSFORM THE DATA

BUILD A CONVOLUTIONAL
MODEL

TRAIN AND EVALUATE THE


MODEL

11/13/2019 11
FEATURE EXTRACTION

• To produce high level features,network goes deeper


with more number of layers.
• In deep CNN there are 4 different layers are
present,they are
1. Convolutional layer
2. Max pooling layer
3. ReLu(Rectified linear units) layer
4. Fully Connected layer

11/13/2019 12
The first The Nth convolutional Fully
convolutional layer layer and pooling connection
Input layer layer output
and pooling layer

.
.
.
.

a) Deep cnn architecture

11/13/2019 13
Build a Sequential 1d-Convolutional Model

• Conv1D : filters=2; kernelsize=5;input_shape=(25,1)


• MaxPooling1D : pool_size=2
• Dropout : 0.2
• Flatten
• Dense=128 : activation=relu
• Dense=8 : activation=relu
• activation=softmax

11/13/2019 14
Epoch values for deep CNN

11/13/2019 15
Epoch values for deep CNN

11/13/2019 16
Epoch values for deep CNN

11/13/2019 17
Epoch values for deep CNN

11/13/2019 18
EVALUATE THE MODEL
OPTIMIZER Loss Accuracy
SGD-Momentum 0.3885813629 0.7780217584

11/13/2019 19
CONFUSION MATRIX

predicted

negative positive

negative TN FP
True False Positive
actual negative

positive FN TP
False True positive
negative

11/13/2019 20
Measure the Performance of the Model on
the Dataset

11/13/2019 21
FUTURE WORK

• The value matrices as the input to activate the


trained neural network
• Decision can be takes place at the supervised learning
part using deep DBN.

11/13/2019 22
REFERENCES

[1] N. K. et al., “The deep learning vision for heterogeneous network


traffic control: Proposal, challenges, and future perspective,” IEEE
Wireless Communications, vol. 24, no. 3, pp. 146–153, 2017.
[2] Z. F. et al., “State-of-the-art deep learning: Evolving machine
intelligence toward tomorrow’s intelligent network traffic control
systems,”IEEE Communications Surveys Tutorials, vol. PP, no. 99,
pp. 1–1, 2017.
[3] B. M. et al., “Routing or computing? the paradigm shift towards
intelligent computer network packet transmission based on deep
learning,”IEEE Transactions on Computers, vol. PP, no. 99, pp. 1–1,
2017.

11/13/2019 23
[4] F. T. et al., “On removing routing protocol from future wireless
networks:A real-time deep learning approach for intelligent traffic
control,”IEEE Wireless Communications, vol. 25, no. 1, pp. 154–
160, February 2018.
[5] F. T. et al., “On Intelligent Traffic Control For Large Scale
Heterogeneous Networks:A Value Matrix Based Deep Learning
Approach,” IEEE Communications Letters, , vol. 22, no. 1, pp. 154–
160, october 2018.

11/13/2019 24
THANK YOU

11/13/2019 25

Вам также может понравиться