Machine Learning & AI

This learning stream packages together two modules, taking you through the theory and application of machine learning tools. The course begins with classical machine learning (supervised and unsupervised) and progresses through to deep learning neural networks and Large Language Models (LLMs).

Machine Learning

Supervised:

Key areas taught in L2D's exploration of Supervised Machine Learning include:

• Classification: preparing data for classification, training classifier models.
• State space plot of model predictions
• Prediction probabilities and feature importance
• Complex training and testing of data
• Comparison of different model classes
• Stratified shuffle split
• Evaluation of classification using AUC and ROC curves
• Metrics for model evaluation
• Permutation scoring and confusion matrices
• Normalisation and hyperparameter tuning
• Refinement and progressive adjustment

Unsupervised:

Key areas taught in L2D's exploration of Unsupervised Machine Learning include:

• Gaussian Mixture Models (GMMs) and sci-kit learn
• Clustering and automated data labelling
• Quantitative scoring using ground truth
• Introductions to the concept and pitfalls of clustering techniques
• GMMs in medical image segmentation and object detection
• Dimensionality reduction (reducing computational workloads of large high-dimensionality datasets)
• PCA (Principal Component Analysis)

 

Artificial Intelligence:

The Artificial Intelligence module builds upon the concepts and techniques taught in the machine learning module. This broad topic covers three primary AI approaches:

• Multi-Layer Perceptrons (MLPs)
• Convolutional Neural Networks (CNNs)
• Generative AI (Large Language Models).

Neural networks are introduced in the MLP lesson, with CNNs demonstrated as powerful tools for image recognition and analysis. Generative AI and its integration into the handling of scientific data are explored in the final lesson of this module, with a view to upscaling the power of biological workflows and analyses.

The course examines the concepts underpinning these tools, teaching students how to implement them using PyTorch – an industry-standard Python framework for building deep learning models. Additionally, we explore how to better interpret the output of deep learning models, despite their “black box” nature, with the goal of gaining insight into the patterns these models identify in data.

  Note: L2D's Basic Python and Data Handling modules (or their equivalent) are compulsory requirements to taking our Machine Learning course.