The course will provide an overview of fundamental concepts and algorithms in machine learning. Mathematical and coding exercises will be provided to deepen understanding of these concepts and algorithms.

Material

• Introduction to machine learning and its applications
• Supervised learning: linear and nonlinear regression
• Supervised learning: logistic regression and multi-class classification
• Introduction to neural networks
• Multi layer perceptrons (MLPs)
• Activation functions
• Feedforward neural networks and backpropagation
• Introduction to convolutional neural networks for computer vision
• Image representations
• Convolutional operators and kernels, max pooling, dense layers, depth vs width
• Dimension calculations and parameter count
• Popular datasets and model zoos
• Transfer learning
• Adversarial attacks and Safety
• Large Languauge Models
• Tokenization
• Pretraining, supervised finetuning, and RLHF
• Decoder only models
• Conclusions and a project

Instructor

• Dr. Adel Bibi (Email: bibiadel93@gmail.com (preferred), adel.bibi@eng.ox.ac.uk)

Tutorial

• Tutor: James Oldfield (Email: jamesalexanderoldfield@gmail.com)

Notes

• NOTE 1: We will go through the following content in 4 days. We will use [google colab] as a platform for exercises.

• NOTE 2: for each colab file below, with a preappended title Exercise:, you should make a copy in your own google drive to edit & run.

• NOTE 3: The coffee break over the four days will always be from 10:30AM to 11:00AM.

Background Review

• Colab/Markdown/latex intro [env intro]

• Coding exercise: Python Basics [Exercise: Numbers] [Exercise: Strings] [Exercise: Lists] [Exercise: Dictionaries] [Exercise: Tuples] [Exercise: IF,FOR, and While] [Exercise: Numpy]

• For background review on linear algebra and probability, read the following course

Lectures

Day 1: August 4th, 2025

• Lecture #1 (part 1): AI and Machine Learning Applications
• Lecture #1 (part 2): Regression
• Excercise #1: Regression, Solutions

Day 2: August 5th, 2025

• Lecture #2: Logistic regression
• Exercise #2: Logistic regression, Solutions

Day 3: August 6th, 2025

• Lecture #3: Neural Networks
• Exercise #3: Neural Networks, [Solutions]

Day 4: August 7th, 2025

• Lecture #4: Convolutional Neural Networks (1)
• Exercise #4: Training a ConvNetwork, [Solutions]

Day 5: August 12th, 2025

• Lecture #5: Convolutional Neural Networks (2)
• Exercise #5: Transfer learning
• Exercise #6: Adversarial attacks

Day 6: August 13th, 2025

• Lecture #6: Large Language Models (part 1)
• Exercise #7: Tokenization, [Solutions]

Day 7: August 14th, 2025

• Lecture #7: Large Language Models (part 2)
• Exercise #8: Two Head Transformer, [Solutions]

Day 8: August 15th, 2025

• Lecture #8: AI Safety

Project (Submission Deadline is August 24th anywhere on earth)

• Class Project Training LLMs

• Notebook requirements:
  • Ensure the entire notebook is run, and all results are displayed before submission.
• Rename the file:
  • Change the notebook file name to include your first and last names.
  • Example: If your name is John Smith, rename the file to Project_John_Smith.ipynb.
• Evaluation:
  • If you implement only the TODOs, you will definitely pass the class. However, your final grade will be based on the final loss function in the last cell of your notebook.
  • Note: Solutions to the multi-head attention (exercise 8) are available. You can use those as a baseline. However, to achieve a lower loss function, you will likely need to implement more heads and additional layers.
• Download your submission:
  • Go to File > Download > Download .ipynb.
• Submit via email:
  • Send the downloaded .ipynb file to jamesalexanderoldfield@gmail.com.
  • Use the following email subject format: OxfordAI_Project_FirstName_LastName.
  • Example: If your name is John Smith, the email subject should be OxfordAI_Project_John_Smith.