Professional Documents
Culture Documents
Lecture 8
By
Prof. Buddhika Jayasekara
buddhikaj@uom.lk
AI
technique
Natural
Intelligence
Intelligent Solution
Physical
presence
(Hardware/
Software)
11/5/2021 7:07 AM 2
Starting with Supervised Learning
❑ Load in data ( X and t)
❑ Instantiate model
❑ train (fit) model
❑ Evaluate model
Backpropagation of Error
Input space
Target space
X
t
10
8
Tartget values
0
0 1 2 3 4 5 6 7 8 9 10
X values
❑ Deep - ANN
Wide
Deep
4
Convolutional Neural Network (CNN)
5
Convolutional Neural Network (CNN)
6
Convolutional Neural Network (CNN)
7
Stride
❑ Stride refers to the number of pixels the kernel filter will skip
8
Pooling
9
Convolutional Neural Network (CNN)
❑ E.g. datasets
▪ MNIST – 28x28=784 input vector
▪ CIFAR-10 – 32x32x3 =3072 input vector
▪ With modern CNN, 224x224x3 -150000+ features
▪ HD image of 1280*720*3 -2.8 million
10
CNN Architecture
11
CNN Architecture
12
CNN Architecture
13
CNN Architecture
14
CNN -Example
15
Fashion mnist - Dataset
16
Keras: Sequential vs. Functional API
https://www.tensorflow.org/api_docs/python/tf/keras/Sequential
Self study
https://www.tensorflow.org/guide/keras/functional 17
18
19
20
21
22
23
Assignment
24