COMSM0152 - Foundations of ProAI

Unit Information

This unit introduces students to a range of methodologies and applications across the field of AI. Each week will focus on a different theme, with three activities:

Required reading -- background reading to introduce the topic of the week
Guest seminars (TB1: Mondays, 1300-1400, Queen's 1.58) -- a guest speaker will present an introduction to their research area, connected to the week's topic
Group discussion (TB1: Fridays, 1000-1100, Queen's 1.68) -- student led discussion on the required reading and guest seminars. This is a chance to go over the important concepts and raise questions. The sessions will be chaired by a student, to facilitate the discussion (they are not required to present the topic). Link to spreadsheet with session chairs
AI Lab Lunch and Learn Seminars: (usually Wednesdays at 1330-1500, Queen's 1.07) -- these sessions are not exclusive to the CDT and we invite the whole AI Lab and beyond, with visiting researchers as well as our local speakers giving talks, and, importantly, free pizza. Many foundational and applied AI topics are covered here, so it is worth attending.

Bear with us while we organise the schedule below -- Topics are likely to change a bit as we arrange with our guest speakers. Please do give us feedback at any time on how we can make the unit work better! In particular, if you find some topics assume prior knowledge you don't have, or would like to go further into certain topics, let us know. Please see below for some great all-round textbooks on AI, which may help you fill in some knowledge gaps... We also have a Team on MS Teams. Please do use it to ask us questions, discuss AI topics you are interested in, post interesting blogs, videos or papers, etc...

Staff

James Cussens (JC) Edwin Simpson (ES)

Unit Materials

TB1

TB2

Week	Reading (to complete before Monday's lecture)	Guest Seminar	Discussion chair	Lecturer responsible
1 (w/c 22/09/25)	Embeddings and Deep Learning for NLP Slides from the seminar Embeddings BERT, Devlin et al., 2018 Optimising Factual Consistency in Summarisation via Preference Learning from Multiple Imperfect Metrics: This is our recent paper to be presented at EMNLP Findings in November, showing how we are using embeddings and transfer learning to improve text generation with small LLMs (please read before next Friday).	Edwin Simpson	Tom	ES
2 (w/c 29/09/25)	Attention and Transformers. Both of the following are really good introductions to attention and transformers - you can pick one to read, or go through both if that helps your understanding. Please cover one of these before Monday: The Illustrated Transformer Jurafsky and Martin Chapter 9, "Transformers" For the Friday reading group, please also read: Learning a Text-Video Embedding from Incomplete and Heterogeneous Data (and OPTIONALLY, for the keen) TOKEN MERGING: YOUR VIT BUT FASTER	Mike Wray	Ed	ES
3 (w/c 06/10/25)	Large language models: Jurafsky and Martin, Large Language Models Jurafsky and Martin, Post-training Assessing the ability of LSTMs to learn syntax-specific dependencies, Linzen et al. (2016), TACL Beyond the limitations of any imaginable mecahnism: large language models and Psycholinguistics, Houghton, Kazanina and Sukumaran, Behavioral and Brain Sciences, (2023).	Conor Houghton -- linguistic perspectives on LLMs	Gonzalo	ES
4 (w/c 13/10/25)	Data and labels for image and Video understanding: ImageNet, the competition that kick-started the deep learning revolution in Computer Vision Kitti, a highly impactful dataset in vision and robotics. Optional extra: The winner of the ImageNet 2015 competition, ResNet.	Dima Damen -- tutorial on data and labels for video understanding	Ike	ES
5 (w/c 20/10/25)	Causal World Models for Practice-oriented AI: From Representation to Decision-Making. This topic will touch on LLMs, Reinforcement learning and practical case studies. Suggestions for reading the papers: firstly, don't worry that the PDFs are very long -- the core content of each paper is only in the first part. Don't feel compelled to read every section fully, e.g., you can probably skip over some methodological details. The papers are listed in a suggested 'order of priority' -- it's up to you to decide whether to read them all, e.g., if there are many concepts that are new to you, you might focus on the first one or two. World Modelling (introductory slides from Ha and Schmidhuber). Curious Causality-Seeking Agents Learn Meta Causal World (NeurIPS 2025 by Mengyue's team) Attaining Humans Desirable Outcomes in Human-AI Interaction via Structural Causal Games Learning World Models with Identifiable Factorization (an earlier NeurIPS 2023 paper on learning world models) James' Cussen's talk on Causality has been moved to TB2 to fit speakers' scheduling constraints.	Mengyue Yang	Isobel	ES
6 (w/c 27/10/25)	CONSOLIDATION WEEK	(no seminar/reading group)
7 (w/c 03/11/25)	Philosophy of AI	James Ladyman	Noah	ES
8 (w/c 10/11/25)	Bias, fairness and transparency	Miranda Mowbray	Russell	JC
9 (w/c 17/11/25)	Shaping Healthcare Data with AI: Imaging and Language Models Imaging: CellSeg3D, Self-supervised 3D cell segmentation for fluorescence microscopy Generative Diffusion Models for Medical Imaging Language: Detecting the clinical features of difficult-to-treat depression using synthetic data from large language models Training LLMs with Synthetic Data (How NVIDIA Trained Nemotron)	Qiang Liu	Chakaya	JC
10 (w/c 24/11/25)	Robust AI and Generalisation	Gabriel Oliveira	Abby	JC
11 (w/c 01/12/25)	Ethical and regulatory frameworks of AI	TBC	Priya	JC
12 (w/c 08/12/25)	Assessment period	TB1 essay due this week
13 (w/c 19/01/26)	Knowledge representation and reasoning R. Davis, H. Shrobe, and P. Szolovits. What is a Knowledge Representation? AI Magazine, 14(1):17-33, 1993 (16pp. + references) Delgrande, J. P., Glimm, B., Meyer, T., Truszczynski, M., and Wolter, F. (2023). Current and future challenges in knowledge representation and reasoning. arXiv preprint arXiv:2308.04161 (30pp. + appendices and an extensive bibliography) Additional reading: What are Non-classical Logics and Why Do We Need Them? An Extended Interview with Dov Gabbay and Leon van der Torre	Oliver Ray		JC
14 (w/c 26/01/26)	Causality -- reading list to be updated. Preliminary suggestions (list from last year): Chapter 1 of Causal Inference in Statistics: A Primer by Pearl, Glymour and Jewell. You can skip Section 1.3 if you're already familiar with the material it contains. The tale wagged by the DAG: broadening the scope of causal inference and explanation for epidemiology by Krieger and Davey-Smith Comments on: The tale wagged by the DAG by Pearl Reply to Pearl: Algorithm of the truth vs real-world science by Krieger and Davey-Smith Supplementary reading: Schölkopf et al, Towards Causal Representation Learning	James Cussens		JC
15 (w/c 02/02/26)	Reinforcement learning (TBC)	TBC		ES
16 (w/c 09/02/26)	Robotics (TBC)	TBC		ES
17 (w/c 16/02/26)	Multi-agent systems (TBC)	TBC		JC
18 (w/c 23/02/26)	CONSOLIDATION WEEK	No seminar/reading group
19 (w/c 02/03/26)	Learning from Temporal Data (TBC)	TBC		JC
20 (w/c 09/03/26)	Explainable and interpretable AI (TBC)	TBC		ES
21 (w/c 16/03/26)	Privacy (TBC)	TBC		JC
Easter Vacation (w/c 23/03/26)
22 (w/c 13/04/26)	Weak supervision (TBC)	TBC		ES
23 (w/c 20/04/26)	Human-in-the-loop AI -- design and evaluation (TBC)	TBC		JC

Assessment Details

Deadlines: at end of TB1 and end of TB2, dates TBC

After each Teaching Block students submit an essay of about 5,000 words (10 pages) on a research topic jointly chosen by them and their Academic Mentor. The essay should describe the background, state of the art, and open challenges with regard to the chosen topic. Each essay is assessed on a pass/fail basis in terms of scholarly content and academic writing. Narrative feedback is also provided, indicating strong points as well as areas for improvement. Passing the unit requires passing both essays. More guidance will be provided by the unit lecturers during the first term.

Text books

Bishop, C. M., Pattern recognition and machine learning (2006). This is one of the best ML textbooks and will provide a solid foundation across many aspects of ML. The book is freely available here.
Russell, S. and Norvig, P., Artificial Intelligence, A Modern Approach, 4th Edition (2020). The canonical introduction to AI, 3rd edition available at here.
Jurafsky, D. and Martin, J.H., Speech and Language Processing, 3rd edition drafts (2024). A great NLP textbook, but very readable and an interesting way to think about the challenges of designing AI systems in general. Also good for showing a contrast between deep learning and feature engineering approaches. Online only, here.
Murphy, K., Probabilistic Machine Learning: An Introduction (2022) and Murphy, K., Probabilistic Machine Learning: Advanced Topics (2023). A more recent ML textbook with particularly good coverage of probabilistic methods, freely available via here.

Some of you may prefer videos to books. Youtube has many video lectures, including the Machine Learning course from Stanford by Andrew Ng, which is very good.