Skip to main content Skip to navigation

Foundation of Contemporary AI: Classification and Generation (IM274-30)

Foundations of Contemporary AI: Classification and Generation

Introductory description

This module serves as an introduction to the techniques and technical underpinnings of contemporary artificial intelligence technologies. From early multilayer perceptrons and machine learning methods to the convolutional neural networks of the late 1990s to the rise of deep learning in the mid-2010s, students will gain hands-on experience with constructing and training simple classification models before moving on to generative architectures for text data such as GPT. Such a material understanding of these foundational technologies will be an invaluable prerequisite for future technical, policy, and/or critical work on the topic of AI.

Module aims

In this module, students will be introduced to the foundational concepts of contemporary AI / deep learning. Students will develop an interdisciplinary and critical perspective on AI by building on in-depth understandings of the assumptions and limitations of contemporary AI models. Students will also develop skills to work with contemporary AI models and methods through hands-on practice.

Outline syllabus

Week 01. Introduction: History of Machine Learning and Artificial Neural Networks.

Week 02. Structured Data and Unsupervised Classification

Week 03. Supervised Classification of Tabular Data [e.g. decision trees]

Week 04. Supervised Classification of Image Data [e.g. MNIST/CNNs]

Week 05. Early Generative Image Models [e.g. GANs]

Week 06. Working with Text [e.g. word2vec, RNNs]

Week 07. The Transformer [Part 1]

Week 08. The Transformer [Part 2]

Week 09. Experimenting with Large Language Models

Learning outcomes

By the end of the module, students should be able to:

  • Develop an understanding of the historical trajectory of machine learning and AI
  • Develop hands-on technical skills in generating ML and AI models
  • Apply unsupervised and supervised learning techniques
  • Build and implement simple classificatory artificial neural network architectures
  • Understand the computational principles behind generative text models
  • Develop critical technical literacy which can be applied to domains of policy and critique
  • Understand interdisciplinary connections between contemporary AI and other scientific and social-scientific traditions

Indicative reading list

TBC

Subject specific skills

  • Demonstrate an appreciation of multi-disciplinary approaches to machine learning and deep learning
  • Understand why particular machine learning and deep learning methods are applied to particular forms of data in particular contexts
  • To innovatively extend knowledge and techniques to novel settings as they emerge.

Transferable skills

  • Think critically and creatively
  • Meet regular deadlines
  • Demonstrate time-management skills
  • Demonstrate problem solving skills
  • Demonstrate independent learning skills
  • Participate in class discussions

Delivery over three terms of 10 weeks each

Lectures - 9 sessions of 2 hours

Tutorials - 9 sessions of 1 hour

Assessment

30 CATS

Laboratory Report (Perform a set of training experiments on a simple set of neural network models and write up results.) - 30% weighting

Final Technical Report (Using a larger neural network architecture and training data, propose, run, and report on a set of novel experiments based on the student’s questions/interests about such models.) - 70% weighting

Let us know you agree to cookies