You are on page 1of 18

Hands-On

Deep Learning
Nurulaqilla Khamis
Skymind Holdings Berhad
(XpressAI Sdn Bhd)
Doctor of Philosophy (PhD)
Master of Philosophy (MPhil)

Nurulaqilla Khamis
Deep Learning Engineer

Bachelor of Engineering
Electrical Electronics (B.Eng)
CONTENTS
1. Recap on Machine Learning (ML) & Deep Learning (DL)
2. Deep Learning Pipeline
3. Integrated Development Environment (IDE) & Dependencies
4. Deep Learning Frameworks
5. Deep Learning Model (Hands-On using Google Colab)
Image Classification Problem

Sentiment Analysis Problem


6. Take Home Message
RECAP ON ML & DL

Time Series Forecasting:

Computer Vision: Stock Price, Fraud Detection

Face Recognition, Natural Language


Image Classification, Processing:
Object Detection
Text Classification, Topic
Modelling, Sentiment
RECAP ON ML & DL
(Cont.) Classification Problem
SUPERVISED
LEARNING Regression Problem
• Data must have label
• Learn from label data

UNSUPERVISED
Clustering Problem
LEARNING
• Data don’t have label
• Learn without label data
• Capture the data
characteristics, pattern

REINFORCEMENT
LEARNING
• Learn make decision
through reward by trial &
RECAP ON ML & DL (Cont.)
NEURAL NETWORK
RECAP ON ML & DL (Cont.)
What Makes Neural Network Not Preferable
To Process Images?

Image Size = 28 x 28 x 1
(width, height, channel)
Drawbacks:
1. Complex nodes and network
2. Huge # of Parameters (Weights),
Channel – 1 (Grayscale Image) tend to overfitting
Channel – 3 (RGB Image)

Imagine Image Size: (720 x 720 x 3) = 1,555,200 neurons!


RECAP ON ML & DL (Cont.)
CONVOLUTIONAL NEURAL
NETWORK (CNN)

7 x 7 x 40 = 1960 neurons

Input Layer Feature Extraction Layer Output Layer

39 x 39 x 3 = 4563 neurons (image down sampling process)


RECAP ON ML & DL (Cont.)
DEEP LEARNING PIPELINE
80%
from HERE!

• Dataset • Augmentatio • Supervised • Classification: • User


Structured
distributions n Confusion Acceptance
• Unsupervised
Matrix, Test (UAT)
• Imbalanced • Image:
Accuracy
dataset OpenCV, PIL • Go Live!
• Regression:
Unstructured • Text:
Mean Squared
NLTK, Spacy
Error (MSE)
• Audio:
Librosa

Data Deep Learning Engineer AI System


INTEGRATED DEVELOPMENT
ENVIRONMENT (IDE) & DEPENDENCIES
Programming Language Integrated Development Environment (IDE) Dependencies/Libraries
DEEP LEARNING FRAMEWORKS
HANDS-ON PLATFORM

HOW TO USE IT?


 Search “Google Colaboratory”
 Sign-in using your

 Pre-installed dependencies/libraries
 Saved on the cloud
 Link with GitHub
 Free GPUs and TPUs
DEEP LEARNING MODEL (IMAGE
CLASSIFICATION)
Let’s Get Started!
1. Data Acquisition
MNIST Dataset (Handwritten Digit Classification
Problem)
10 classes image

2. Exploratory Data Analysis


3. Data Pre-Processing
4. Model Development
5. Model Training MNIST Dataset and Number Classification

6. Model Evaluation
DEEP LEARNING MODEL
(SENTIMENT ANALYSIS)
Let’s Get Started!
1. Data Acquisition
IMDB Dataset (Movie Review Dataset)
2 classes sentiment polarity

2. Exploratory Data Analysis


3. Data Pre-Processing
4. Model Development
5. Model Training Sentiment Analysis of IMDB Movie Review

6. Model Evaluation
TAKE HOME MESSAGE

Practice Makes PERFECT!


Polish Your Programming
Skills!
Build Your Portfolio!
Challenge
Yourself!
CERTIFIED DEEP LEARNING EXAMINATION (CDLE)

https://skymind.global/
https://skymind.global/deep-learning-certification/
aqillakhamis@gmail.com

You might also like