CM20315 01 Intro 01

CM20315 - Machine Learning
Prof. Simon Prince and Dr. Harish Tayyar Madabushi

1. Introduction
Semester 1
Logistics
Semester 1
• 2 lectures per week
• 1 lab session per week
• 5 weeks / consolidation week / 5 weeks
• 1 coursework
• Set Monday 21st November
• Due Monday 5th December
• 1 exam (Jan/Feb) – closed book
• Lectures will be recorded
• Slides and notes online
• Ask questions on Moodle logistics message board
Feedback
• Please ask questions in the lectures – put your hand up
• There will be tutors to help in the labs – sometimes Python
programming, sometimes problem sheets
• The solutions to the labs will be posted one week after the lab so you
can see how you are doing and clear up any misconceptions.
• You can ask questions on Moodle
• Please answer other people’s questions if you can!
• There is no such thing as a stupid question.
• Anonymous feedback to me via:
• https://forms.office.com/r/UjVD6Yzz01
Lab sessions
• Python notebooks in CoLab
• You will need a Google account (make one before tomorrow)
• Alternatively, set up Jupyter Notebooks yourself
• Numpy
• Matplotlib
• PyTorch
• Problem sheets
How to pass this course
• Ideas are simple, but build on one another
• Come to all the lectures (take notes)
• Come to all the lab sessions
• Complete the Python notebooks
• Complete the problems
• Work together on non-assessed work
• Read the notes (and take your own) after the class
• Ask questions on the Moodle forums (and answer them)
Book
Book
• Examinable unless specified
• Chapters 1-11,13
Book
• Examinable unless specified
• Chapters 1-11,13
• Not examinable unless specified
• Notes at end of chapters
Landmarks in AI
• 2012 AlexNet (Image classification)
• 2014 Generative adversarial networks (Image generation)
• 2016 AlphaGo
• 2017 Machine translation
• 2019- Language models (Bert, GPT3)
• 2022 Dall-E2 (Image synthesis from text prompts)
2018 Turing award winners
Deep Learning
• Supervised learning
• Tasks
• Models
• Deep learning models
• Unsupervised learning
• Generative models
• Probabilistic generative models
• Latent variable models
• Reinforcement learning
Deep Learning
• Supervised learning
• Tasks
• Models
• Deep learning models
• Unsupervised learning
• Generative models
• Probabilistic generative models
• Latent variable models
• Reinforcement learning
Supervised learning
• Define a mapping from input to output
• Learn this mapping from paired input/output data examples
Five simple examples
Regression
• Univariate regression problem (one output, real value)

• Fully connected network
Graph regression
• Multivariate regression problem (>1 output, real value)

• Graph neural network
Text classification
• Binary classification problem (two discrete classes)

• Transformer network
Music genre classification
• Multiclass classification problem (discrete classes, >2 possible values)

• Convolutional network
Image classification
• Multiclass classification problem (discrete classes, >2 possible classes)

• Convolutional network
What is a supervised learning model?
• An equation relating input (age) to output (height)

• Search through family of possible equations to find one that fits training data well
What is a supervised learning model?
• Deep neural networks are just a very flexible family of equations

• Fitting deep neural networks = “Deep Learning”
Five more examples
• More complex outputs

• But constrained by input (labelling)
Image segmentation
• Multivariate binary classification problem (many outputs, two discrete classes)

• Convolutional encoder-decoder network
Graph node classification
• Multivariate binary classification problem (many outputs, two discrete classes)

• Graph neural network
Entity classification
• Multivariate multiclass classification

• Transformer network
Depth estimation
• Multivariate regression problem (many outputs, continuous)

Pose estimation
• Multivariate regression problem (many outputs, continuous)

Terms
• Classification = discrete classes as output
• Regression = continuous numbers as output
• Two class and multiclass classification treated differently
• Univariate = one output
• Multivariate = more than one output
Three more examples
Translation
Image captioning
Image generation from text
What do these examples have in common?
• Very complex relationship between input and output
• Sometimes may be many possible valid answers
• But outputs (and sometimes inputs) obey rules
Language obeys Natural images also

grammatical rules have “rules”
Idea
• Learn the “grammar” of the data from unlabeled examples
• Can use a gargantuan amount of data to do this (as unlabeled)
• Make the supervised learning task earlier by having a lot of
knowledge of possible outputs

CM20315 01 Intro 01

Uploaded by

Copyright:

Available Formats

CM20315 01 Intro 01

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

CM20315 01 Intro 01

Uploaded by

Copyright:

Available Formats

CM20315 - Machine Learning

Prof. Simon Prince and Dr. Harish Tayyar Madabushi

• Univariate regression problem (one output, real value)

• Multivariate regression problem (>1 output, real value)

• Binary classification problem (two discrete classes)

• Multiclass classification problem (discrete classes, >2 possible values)

• Multiclass classification problem (discrete classes, >2 possible classes)

• An equation relating input (age) to output (height)

• Deep neural networks are just a very flexible family of equations

• More complex outputs

• Multivariate binary classification problem (many outputs, two discrete classes)

• Multivariate binary classification problem (many outputs, two discrete classes)

• Multivariate multiclass classification

• Multivariate regression problem (many outputs, continuous)

• Multivariate regression problem (many outputs, continuous)

Language obeys Natural images also

You might also like