All posts – Pedro Leitão

Getting Things Right: How Engineering Teams Scale and Achieve Great Things

Software Engineering

larger doesn’t always mean faster or better

Moving to Quarto

why i think quarto is better

I’ve been using JupyterBook to create this site for a while, and I’ve been pretty happy with it. But I’ve recently started using Quarto, and I think…

France, AI and Back to Nuclear for Germany ?

were the french right all along ?

President Macron has just signalled that France is about to announce €109 billion private sector investments in AI. This is a significant amount of…

Reinforcement Learning - a Primer using Connect Four

the basics of reinforcement learning using connect four as an example, with the gymnasium and baselines libraries

Machine Learning

Reinforcement Learning

One of the mainstay algorithms in machine learning is reinforcement learning (or RL for short). RL is an approach to machine learning that is used…

How GPU’s work, an explainer using the Mandelbrot set

mandelbrot sets and parallelism

Every day pretty much all of us either uses or hears about the mythical GPU, the Graphics Processing Unit. It’s the thing that makes your games…

Regularisation in Machine Learning

a practical guide in overfitting prevention

Machine Learning

Regularisation is a technique designed to prevent models from overfitting. In other words, it helps your model generalise better to unseen data by…

Reasoning Models for Fun and Profit

deepseek r1 and the chinese room

Language Models

Since the advent of GPT-3, foundation models have rapidly progressed from single pass transformer models, to multi-step models that can reason over…

Model Fine-tuning with the Hugging Face `transformers` Library

the basics of traditional fine-tuning

Language Models

Previously, we learned how to use Apple’s MLX framework to fine-tune a language model. This is an Apple specific framework and is not available to…

Fine-tuning an LLM with Apple’s MLX Framework

fine-tuning pre-trained language models in apple silicon

Language Models

Modern GPU’s come with inbuilt memory, which is separate from the CPU’s memory. This means that when training large models, the data has to be…

Model Management with MLflow

how to log, compare, and deploy machine learning models consistently with mlflow

Machine Learning

Model Management

As you develop machine learning models, you will find that you need to manage many different versions and variations as you move towards the desired…

Will Artificial Intelligence Ever be More than Fancy Curve Fitting ?

ai is just hyperdimensional regression

Machine Learning

One of my favourite books from my late 20’s was Roger Penrose’s The Emperor’s New Mind, where he argues that human consciousness is non-algorithmic…

A Classical Machine Learning Problem: Predicting Customer Churn

using machine learning to solve a common business problem

Machine Learning

Customer churn, where customers stop using a company’s services, is a major concern for businesses as it directly impacts revenue. Traditionally…

Text Tasks without Neural Networks

random forests and embeddings for sentiment analysis

Machine Learning

Natural language processing (NLP) is often associated with deep learning and neural networks. However, there are efficient methods for text…

The Basics of Unsupervised Learning: Segmenting an Image

a quick primer on what unsupervised learning means and how it can be used for image segmentation

Machine Learning

Unsupervised Learning

Unsupervised learning is a type of machine learning that looks for previously undetected patterns in a dataset without pre-existing labels and with…

Exploring the Impact of Kolmogorov-Arnold Networks in Machine Learning

a new dawn for machine learning, or just another fad?

Machine Learning

Machine learning never sleeps, and its latest wake-up call is the cutting-edge Kolmogorov-Arnold Networks (KANs), as detailed in this NSF paper.…

The Connundrum of European Tech and Artificial Intelligence

why Europe is lagging behind in ai investment and innovation

The European tech scene is a curious one. It is a region that is home to some of the oldest and most representative companies in the world, yet it…

On Scaling AI: Are we Hitting the Limits of our Current Approaches?

the data problem and the chinchilla scaling law

There’s been a lot of speculation on whether we’re hitting the limits of our current approaches to scaling AI. Will OpenAI and others be able to…

Machine Learning and Predictive Maintenance

using machine learning for a common industrial and engineering application

Machine Learning

Predictive maintenance leverages machine learning to analyze operational data, anticipate potential failures, and schedule timely maintenance. This…

Caching long running jobs

how to cache the results of long running computations in python

In any data science or machine learning pipeline, one often has to re-try and experiment with long running computations. For example, maybe you will…

Time Series Forecasting with Prophet

predicting the future with machine learning

Time Series Analysis

Machine Learning

Forecasting future trends is a common application in time series analysis. In this experiment, we will use Meta’s Prophet library to predict trends…

Understanding Random Forest Classification and Its Effectiveness

why random forests and ensemble methods are the underrated heroes of machine learning

Machine Learning

A Random Forest is a versatile and robust machine learning algorithm used for both classification and regression tasks. It builds upon the concept…

Instance vs Model Learning

a comparison of two machine learning approaches

Machine Learning

Instance-based machine learning and model-based machine learning are two broad categories of machine learning algorithms that differ in their…

Which Car is Best ? Analysing and Predicting MOT Test Results

dive deeper into data analysis using a real-world dataset

Machine Learning

In this experiment, we will be analysing the MOT test results of cars in the UK. The MOT test is an annual test of vehicle safety, roadworthiness…

A Wine Quality Prediction Experiment with SKLearn Pipelines

pipelining in machine learning, for fun and profit

Machine Learning

In this experiment, let us use the Wine Quality Dataset from Kaggle to predict the quality of wine based on its features. We will investigate the…

Evaluating Dimensionality Reduction - PCA vs t-SNE

reducing dimensions for better insights

Machine Learning

Evaluating the effectiveness of dimensionality reduction techniques, such as Principal Component Analysis (PCA) and t-Distributed Stochastic…

Basics of Word Vectors

understanding what word vectors are, and what they mean in modern natural language processing

Machine Learning

Word vectors are a mainstay of NLP, and are used in a variety of tasks, from sentiment analysis to machine translation. In this experiment, we will…