00_pytorch_and_deep_learning_fundamentals.pdf

Machine learning is turning things (data)
into numbers and finding patterns in those
numbers.
The computer does this part.
How?
Code & math.
We’re going to be writing the code.

Arti
fi
cial
Intelligence
Machine
Learning
Machine Learning vs. Deep Learning
Deep
Learning

Traditional
programming
Machine
learning
algorithm
Starts with
Inputs Output
1. Cut vegetables
2. Season chicken
3. Preheat oven
4. Cook chicken for 30-minutes
5. Add vegetables
Starts with
Inputs Rules
Makes
Output
Figures out
1. Cut vegetables
2. Season chicken
3. Preheat oven
4. Cook chicken for 30-minutes
5. Add vegetables
Rules

“Why use machine learning (or
deep learning)?”

Better reason: For a complex
problem, can you think of all the rules?
(probably not)
Good reason: Why not?

Source: 2020 Machine Learning Roadmap video.

— A wise software engineer… (actually rule 1 of Google’s Machine Learning Handbook)
“If you can build a simple rule-based system
that doesn’t require machine learning, do
that.”
(maybe not very simple…)

What deep learning is good for
• Problems with long lists of rules—when the traditional
approach fails, machine learning/deep learning may help.
• Continually changing environments—deep learning can
adapt (‘learn’) to new scenarios.
• Discovering insights within large collections of data—can
you imagine trying to hand-craft rules for what 101 di
ff
erent
kinds of food look like?
🤖✅

What deep learning is not good for
• When you need explainability—the patterns learned by a deep
learning model are typically uninterpretable by a human.
• When the traditional approach is a better option — if you can
accomplish what you need with a simple rule-based system.
• When errors are unacceptable — since the outputs of deep
learning model aren’t always predictable.
• When you don’t have much data — deep learning models
usually require a fairly large amount of data to produce great
results.
🤖🚫
(typically)
(though we’ll see how to get great results without huge amounts of data)

Unstructured data
Deep
Learning
Structured data
Machine
Learning
Algorithm: neural
network
Algorithm: gradient
boosted machine

Structured data Unstructured data
• Random forest
• Gradient boosted models
• Naive Bayes
• Nearest neighbour
• Support vector machine
• …many more
• Neural networks
• Fully connected neural network
• Convolutional neural network
• Recurrent neural network
• Transformer
• …many more
What we’re focused on building
(with PyTorch)
(since the advent of deep learning these are
often referred to as “shallow algorithms”)
(depending how you represent your problem,
many algorithms can be used for both)
(common algorithms)

“What are neural networks?”

Neural Networks
[[116, 78, 15],
[117, 43, 96],
[125, 87, 23],
…,
[[0.983, 0.004, 0.013],
[0.110, 0.889, 0.001],
[0.023, 0.027, 0.985],
…,
Inputs
Numerical
encoding
Learns
representation
(patterns/features/weights)
Representation
outputs
Outputs
(a human can
understand these)
Ramen,
Spaghetti
Not a diaster
“Hey Siri, what’s
the weather
today?”
(choose the appropriate
neural network for your
problem)
(before data gets used
with a neural network,
it needs to be turned
into numbers)
Each of these nodes is
called a “hidden unit”
or “neuron”.

Anatomy of Neural Networks
Input layer
(data goes in here)
Output layer
(outputs learned representation or
prediction probabilities)
Hidden layer(s)
(learns patterns in data)
Note: “patterns” is an arbitrary term, you’ll often hear “embedding”, “weights”, “feature representation”,
“feature vectors” all referring to similar things.
# units/neurons = 2
# units/neurons = 3
# units/neurons = 1
Each layer is usually combination of
linear (straight line) and/or non-
linear (not-straight line) functions
Overall
architecture

Supervised
Learning
Unsupervised &
Self-supervised
Learning
Transfer
Learning
Types of Learning
We’ll be writing code to do these,
but the style of code can be adopted across learning paradigms.

“What is deep learning actually
used for?”

Deep Learning Use Cases
Recommendation Translation
“Hey Siri, who’s the
biggest big dog of
them all?”
Speech recognition
Computer Vision Natural Language Processing (NLP)
To: daniel@mrdbourke.com
Hey Daniel,
This deep learning course is incredible!
I can’t wait to use what I’ve learned!
To: daniel@mrdbourke.com
Hay daniel…
C0ongratu1ations! U win $1139239230
Spam
Not spam
Sequence to sequence
(seq2seq)
Classi
fi
cation/regression
(some)

What is PyTorch?
• Most popular research deep learning framework*
• Write fast deep learning code in Python (able to run on a GPU/many
GPUs)
• Able to access many pre-built deep learning models (Torch Hub/
torchvision.models)
• Whole stack: preprocess data, model data, deploy model in your
application/cloud
• Originally designed and used in-house by Facebook/Meta (now open-
source and used by companies such as Tesla, Microsoft, OpenAI)
*Source: paperswithcode.com/trends February 2022

Why PyTorch?
Research favourite
Source: paperswithcode.com/trends February 2022

Why PyTorch?
Source: @fchollet Twitter
and PyTorch

What is a GPU/TPU?
GPU (Graphics Processing Unit)
TPU (Tensor Processing Unit)

Neural Networks
[[116, 78, 15],
[117, 43, 96],
[125, 87, 23],
…,
[[0.983, 0.004, 0.013],
[0.110, 0.889, 0.001],
[0.023, 0.027, 0.985],
…,
Inputs
Numerical
encoding
Learns
representation
Representation
outputs
Outputs
(a human can
understand these)
Ramen,
Spaghetti
Not spam
“Hey Siri, what’s
the weather
today?”
(choose the appropriate
neural network for your
problem)
(before data gets used
with an algorithm, it
needs to be turned into
numbers)
These are tensors!

[[116, 78, 15],
[117, 43, 96],
[125, 87, 23],
…,
[[0.983, 0.004, 0.013],
[0.110, 0.889, 0.001],
[0.023, 0.027, 0.985],
…,
Inputs
Numerical
encoding
Learns
representation
Representation
outputs
Outputs
Ramen,
Spaghetti
These are tensors!

“What are we going to
cover?”

• Now:
• PyTorch basics & fundamentals (dealing with tensors and tensor operations)
• Later:
• Preprocessing data (getting it into tensors)
• Building and using pretrained deep learning models
• Fitting a model to the data (learning patterns)
• Making predictions with a model (using patterns)
• Evaluating model predictions
• Saving and loading models
• Using a trained model to make predictions on custom data
How: 👩🔬
👩🍳
What we’re going to cover
(broadly)
(we’ll be cooking up lots of code!)

What we’re going to cover
A PyTorch workflow
(one of many)

“How should I approach
this course?”

2. Explore and
experiment
How to approach this course
1. Code along
Motto #1: if in doubt, run the code!
Motto #2:
Experiment, experiment,
experiment!
3. Visualize what you
don’t understand
Motto #3:
Visualize, visualize, visualize!
4. Ask questions
🛠
5. Do the exercises
🤗
6. Share your work
(including the
“dumb” ones)

How not to approach this course
Avoid: 🧠
🔥
🔥
🔥 “I can’t learn
______”

Resources
https://www.github.com/mrdbourke/pytorch-deep-learning
Course materials
https://www.github.com/mrdbourke/pytorch-deep-learning/
discussions
Course Q&A
https://learnpytorch.io
Course online book
PyTorch website &
forums
This course
All things PyTorch

tensor([[[1, 2, 3],
[3, 6, 9],
[2, 4, 5]]])
0
1
2
0 1 2
tensor([[[1, 2, 3],
[3, 6, 9],
[2, 4, 5]]])
tensor([[[1, 2, 3],
[3, 6, 9],
[2, 4, 5]]])
torch.Size([1, 3, 3])
0 1 2
Dimension (dim)
dim=0
dim=1
dim=2
Tensor dimensions

20 0 24 44
A*J + B*L + C*N A*K + B*M + C*O
D*J + E*L + F*N D*K + E*M + F*O
G*J + H*L + I*N G*K + H*M + I*O
44 38
126 86
58 63
4 6 8
5 0 3
3x2
3x3 3x2
4 7
6 8
8 1
J K
L M
N O
5 0 3
3 7 9
3 5 2
A B C
D E F
G H I
Dot product
3x3 3x2
torch.matmul( )
, )
* * *
= = =
+ +
3x2
For a live demo, checkout www.matrixmultiplication.xyz
Numbers on the inside must match New size is same as outside numbers
=
torch.matmul(
,

[[116, 78, 15],
[117, 43, 96],
[125, 87, 23],
…,
[[0.983, 0.004, 0.013],
[0.110, 0.889, 0.001],
[0.023, 0.027, 0.985],
…,
Inputs
Numerical
encoding
Learns
representation
Representation
outputs
Outputs
Ramen,
Spaghetti
[[0.092, 0.210, 0.415],
[0.778, 0.929, 0.030],
[0.019, 0.182, 0.555],
…,
1. Initialise with random
weights (only at beginning)
2. Show examples
3. Update representation
outputs
4. Repeat with more
examples
Supervised
learning
(overview)

Tensor attributes
Attribute Meaning Code
Shape
The length (number of elements) of
each of the dimensions of a tensor.
tensor.shape
Rank/dimensions
The total number of tensor
dimensions. A scalar has rank 0, a
vector has rank 1, a matrix is rank 2, a
tensor has rank n.
tensor.ndim or tensor.size()
Speci
fi
c axis or dimension (e.g. “1st
axis” or “0th dimension”)
A particular dimension of a tensor. tensor[0], tensor[:, 1]…

00_pytorch_and_deep_learning_fundamentals.pdf

00_pytorch_and_deep_learning_fundamentals.pdf

More Related Content

Similar to 00_pytorch_and_deep_learning_fundamentals.pdf

Recently uploaded

00_pytorch_and_deep_learning_fundamentals.pdf