2. There are two main approaches used
in AI programming
Rules-based Approach Machine Learning Approach
3. Rule Based Approach
Rule-based AI is a type of AI that uses rules to solve a problem.
Human experts typically write rules but are not learned from data.
4. Machine Learning Approach
Machine Learning gives computers the capability to learn from data without being
explicitly programmed for it.
It gives a computer ability to learn, just like a human.
5.
6. Each approach has its applications,
advantages and disadvantages
Next slides will show you some examples on where each approach is used
11. Classifying an object in a photo
Rules-based
Machine Learning
Machine Learning
Which Approach to Use?
12. Each has its benefits
Rule-based Approach Machine Learning
– Rules are defined
– Improvements come from
algorithms and network
– Learns patterns from data
– Improvements may from
additional data
Recap
21. You can see Sequence
Prediction in action by using
Glide type, Handwriting
recognition or Translation task on
the Google’s Gboard app.
22. + =
Style Transfer
Style transfer is a computer vision technique that takes two images—a content image and a
style reference image—and blends them together so that the resulting output image retains
the core elements of the content image, but appears to be “painted” in the style of the style
reference image.
35. DL is machine learning using neural networks.
Glimpse of a basic neural network
36. Unlike former ML algorithms,
Neural Nets can handle image, video, audio and
text.
37. TensorFlow is a deep learning framework developed by Google.
TensorFlow makes it possible to build large deep neural networks
without having to worry about all the mathematics behind it.
38. You can either use existing
models provided TensorFlow
Or build your own neural network
39. TensorFlow can accelerate your models using your GPU.
This is made possible using
Also, TensorFlow is highly scalable and can
utilize the full power of multi GPU systems.
40.
41. ⮚ This is one of most basic
introductory example for
TensorFlow.
⮚ We will use MNIST dataset
included in TensorFlow by default.
⮚ We are going to train a small
neural network for this task, just
like you have seen before.
Let’s try to recognize handwritten
digits using TensorFlow
Click to open Google Colab at
start training the network
7
Input Output
NN
MNIST Dataset
45. Real Life
Applications
• Social media apps use facial recognition to
detect and tag users
• Using computer vision, AR gear detect
objects in real world in order to determine
the locations on a device’s display to place a
virtual object
• Consumer devices use facial recognition to
authenticate the identities of their owners
• Law enforcement agencies also rely on facial
recognition technology to identify criminals
in video feeds
46. How It Works?
Data Labelling
Train and
Test the
Model
Output And
Improve Steps to Train the Model
Convolutional
Neural
Network
(MODEL)
48. Advancements
⮚Few Shots Learning
⮚Image Captioning
⮚Super Resolution
⮚Action Recognition
⮚3D object Detection
IMAGE CAPTIONING MODEL
A Dog Sitting
on the Beach
3D Object
Detection
49. Getting
Started
• Basic:
- OpenCV (you can use either
python or C++)
- Image Processing
• Intermediate:
- TensorFlow
- Basic CNN Model for
Classification(Cats vs Dogs)
50. Natural Language Processing
Analysis and synthesis of natural
language and speech.
To make machines understand and draw
out meaning and relationship
between natural language
51. • Chatbots
• Language Translator
• Sentiment Analysis
• Email Classification and Filtering
53. •An organization might use sentiment
analysis to classify reviews into
different categories when the amount
of reviews is large making it inefficient
to manually classify them.
54. Recurrent Neural
Networks
Recurrent Neural Networks enable
you to model time-dependent and
sequential data problems, such as
stock market prediction, machine
translation, and text generation
Why use RNN: When the
sequence of the data is important
and depicts a meaning then RNN is
advantageous whereas machine
learning algorithms like SVM fails
to make use of and understand
this sequencial meaning.
56. More NLP Networks
Long short-
term memory
(LSTM)
Gated
Recurrent
Unit
Networks
Bidirectional
LSTM
57. Word2Vec
It is used to create a distributed representation of words into numerical vectors
KING
-
MAN
+
WOMAN
=
QUEEN
58. ChatGPT
Generative Pretrained Transformer
GPT-3 and GPT-4 are state-of-the-art language processing AI models
developed by OpenAI
They are capable of generating human-like text and have a wide
range of applications, including language translation, language
modelling, and generating text
61. ⮚ You can know more about TensorFlow at tensorflow.org
⮚ For beginners, Google has set-up a free crash course named
Machine Learning Crash Course with TensorFlow APIs at
https://developers.google.com/machine-learning/crash-course
⮚ offers a TensorFlow Developer Professional
Certificate for those who want to excel at using this library.
⮚ And you can always go search on YouTube for free
TensorFlow courses. Check out freecodecamp, they provides
good free courses on YouTube.
How to learn TensorFlow?
62. Free Courses
Course Platform
Machine Learning for Everybody – Full Course YouTube
Machine Learning Specialization by Andrew Ng
(Audited)
Coursera
Learn the Basics of Machine Learning
Simplilearn
63. Paid
Certificate
Courses
Course Platform
Machine Learning Specialization by Andrew Ng
(with Certificate)
Coursera
Machine Learning with Python by IBM Coursera
Deep Learning Specialization by Andrew Ng
Coursera
Machine Learning A-Z : AI, Python & R + ChatGPT
Bonus [2023] Udemy
64. Kaggle is a extremely popular site among Machine Learning
Enthusiasts and Data Scientists all over the world.
It hosts many live ML and Data Science competitions
each year. Many prominent programmers take part in
these competitions.
It has a huge repository of datasets which can be used
by anyone for their projects.