Tutorial 1
Machine Learning Methods: the essentials
Dr Huiqi Lu – Fulford Junior Research Fellow; Daphne Jackson Research Fellow
This tutorial aims to introduce the basics of machine learning (ML) methods, especially in dealing with time-series data. In this module, you will learn the “essential” ML methods, the important pre-processing steps that are appropriate for time-series data, and how to frame the problem into a ML task. We will firstly briefly introduce fundamental ML methods for classification and regression tasks, then focuses on time-series models, including Autogresssive process, Gaussian Process, and survival analysis.
Tutorial 2
Advanced Recurrent Neural Networks for Time-series Data
Dr Anshul Thakur – Postdoctoral Research Assistant, Institute of Biomedical Engineering, Oxford
This talk will discuss the deep learning architectures to process the time-series data. The recurrent unit (along with its variants such as GRUs and LSTMS) and the neural architectures developed using these recurrent units for different tasks such as denoising, classification and prediction will be described. We will extend these basic neural architectures to develop autoencoders and variational autoencoders for time-series data. The nuances of different implementations of RNN autoencoders will also be presented. Finally, attention-based transformer models will be introduced as an alternative to RNNs. A brief overview of methods to process the irregularly sampled time-series will also be presented if time permits.
Tutorial 3
Challenges of Implementing Clinical ML Research
Dr Davide Morelli – Biomedical Engineer, AI researcher, CTO, Huma, Livourne, Italie
The challenges of implementing clinical ML research into industrial practice are complex and involve:
– understanding of regulatory pathways with careful planning and execution of necessary activities;
– gaining the trust of the clinical community, traditionally sceptical of black-box deep learning models;
– usability issues, like communicating difficult concepts to users (e.g. absolute risk, hazard ratios, correlation vs causation, etc).
In this talk, I will touch on each of these points and explain how we approached them in our projects.