1
Machine Learning for Design
Lecture 3
Machine Learning for Images. Part 1
2
A bit more on regression and classification
And your very first contact with (deep) neural networks
3
Linear Regression
4
5
6
7
8
9
10
11
12
Finding the best parameter values
- Training the model
- Gradient Descent: an algorithm to find the minimum point of a function
- Hyperparameters: parameters of the Gradient Descent
- Learning Rate: speed of descent
- Epochs: max number of steps
13
14
15
Classification
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
Fully connected Neural Network
- Hyperparameters
- Learning rate
- Number of epochs
- Architecture
- #layers, #nodes, activation functions
- Batch vs. mini-batch vs. stochastic gradient descent
- Regularization parameters:
- Dropout probability p
40
Classifying into multiple classes - Softmax function
- Return a probability for each class
- example C1= ADMITTED, C2 = NOT ADMITTED, C3 = NEW TEST
- p(C1) = 0.37, p(C2) = 0,21, p(C3) = 0,42
- We use the Softmax activation function for the output layer
41
42
Machine Learning and Images
43
44
45
Images
- Each pixel in an image is a feature
- numerical
- 0 or 1 for Black and White
- Between 0 and 255 for greyscale
- 16M values for RGB
- numerical
- Dimensionality -> n x m
46
Computer Vision
Building algorithms that can “understand” the content of images and use it for other applications
- It is a “Strong AI” problem
- signal-to-symbol conversion
- The semantic gap
A general-purpose vision system requires
- Flexible, robust visual representation
- Updated and maintained
- Reasoning
- Interfacing with attention goals, and plans
47
What specific tasks can we train a CV system to perform?
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
Machine Learning for Design
Lecture 3
Machine Learning for Images. Part 1
69
Credits
CMU Computer Vision course - Matthew O’Toole.
Grokking Machine Learning. Luis G. Serrano. Manning, 2021