Love Thy Neighbour
Note: we are accepting registrations on a first-come, first-serve basis! This is due to capacity in the lecture theatre. If you attended previous sessions, you will be prioritised.
ELG01 this session (like last session!)
Thank you to those of you who turned up last session, and gave such a warm reception to Mikhail Samin!
🎉 Session 3: Love Thy Neighbour
The Machine Learning (ML)/AI Lecture Series, led by Martin He, continues! 🚀 Last time we looked at how machines find structure in unlabelled data. This week, we give the machine the answers and teach it to predict. Welcome to Supervised Machine Learning, the workhorse behind everything from credit scoring to medical diagnosis, starting with the simplest idea of all: you are who your neighbours are. Then we'll go deeper, into neural networks, and step back to ask the question that matters most: just because we can build it, should we?
Whether you come from STEM, Medicine, Business, Finance, Linguistics, Law or the Humanities, this session will show you how to build models that make predictions, how to tell whether they're any good, and how to think critically about the consequences. Missed an earlier session? No problem, you're still very welcome to join, just catch up through the link below!
We will also introduce the B-side sessions (for industry projects/more applied work/those who definitely can't make Monday sessions).
📚 What to Expect:
Part 1: Supervised ML
🏘️ K-Nearest Neighbours (KNN): Our namesake! Classify a new data point by asking its closest neighbours how they voted. We'll look at how to choose K, why distance metrics matter, and where this beautifully simple algorithm falls short.
📈 Regression (Linear + Logistic): Predicting continuous values with Linear Regression (OLS), and classifying outcomes with Logistic Regression, the two foundations of supervised learning.
🧪 Model Evaluation + Selection: How do you know if a model is actually good? We'll cover the metrics that matter, and how to choose between competing models without fooling yourself.
✂️ Penalised Models (L1, L2, Ridge): Why models overfit, and how regularisation (L1 and L2 penalties, including Ridge Regression) keeps them honest by punishing unnecessary complexity.
🎯 LASSO: How the L1 penalty shrinks coefficients all the way to zero, performing feature selection for free.
🔁 Cross-Validation: The gold standard for estimating how your model will behave on data it has never seen, and for tuning it properly (including picking that all-important K).
🌲 Ensemble Methods (Boosting, Bagging, Random Forest): Why many weak models beat one strong one. We'll explore Bagging, Random Forests, and Boosting, and when to reach for each.
Part 2: Deep Learning
🧬 The Perceptron: Meet the simplest artificial neuron, the building block of every neural network, and see what it can (and can't) do.
❌ The XOR Problem: The famous failure that exposed the limits of the perceptron and demanded a rethink of how we build models.
🕸️ Artificial Neural Networks (ANNs): How stacking layers of neurons lets a network represent far more complex patterns, and how data flows through one.
⚡ Activation Functions: Why non-linearity is the secret ingredient, and how functions like Sigmoid, Tanh, and ReLU shape what a network can learn.
🔙 Backpropagation: The algorithm that lets networks learn from their mistakes, with the chain rule and gradient descent demystified.
Part 3: AI Ethics
⚖️ Normative Ethics (Deontology vs. Consequentialism): Do the rules matter, or only the outcomes? We'll apply two classic ethical frameworks to real AI dilemmas.
🎭 Bias: How models inherit, and amplify, the prejudices hidden in their training data, and what we can do about it.
🌍 Effective Altruism: Using evidence and reason to do the most good, and what that means for those building AI.
🔒 Privacy: What happens to personal data when models are trained on it, and why it matters.
🔍 AI Explainability: Can we trust a model we can't explain? We'll explore why interpretability matters, especially in high-stakes decisions.
Practicalities
💻 Hands-On Coding: Follow along with the Jupyter Notebook (.ipynb) ! You can find at the usual place.
🍕 Pizza, of course. Please give your preferences when you register!
Whether you're brand new to ML, brushing up on the fundamentals, or preparing for data science roles, this session will give you the toolkit to build, evaluate, and improve predictive models with confidence, and the judgement to use them responsibly.
🔥 Bring your laptop, your curiosity, and your questions!
We look forward to seeing you there!
-yours truly, City Optimise x DSAI team
Useful Links
Resources + Earlier Sessions:
https://bit.ly/cityaicollective
WhatsApp:
City Optimise Society Join WhatsApp Group
City DSAI Society Join WhatsApp Group
LinkedIn:
/city-data-science-and-ai-society
Announcement: we are now live on the Student Union! albeit, we need CSGSU to update the description and logo.
https://csgsu.co.uk/explore-and-join/view/428
Also, join DSAI if you can: