Remark: This Autumn School will be organized using online sessions. More details about the practical information of these online sessions will be available soon.
Rather than taking a blackbox approach to statistical modelling, as has been the trend in Machine Learning application via off-the-shelf methods, in this workshop we will take a step back and explore some basic and core statistical methods. The topics selected are chosen since they are integral to many Machine Learning methods. Therefore, obtaining a sound appreciation of such methods can enhance the understanding and complement the application of Machine Learning toolboxes and packages.
In this Autumn School we will focus on explainable and statistically interpretable rigorous methods that can be applied with sound statistical assumptions and verification methods to assess adequacy of assumptions underpinning the models and methods.
The topics included in the workshop cover largely methodological and conceptual approaches to feature extraction and model building. In this manner, all participants can then develop specific detailed applications to explore and build upon these concepts for their particular use cases.
Topics discussed include:
- Part 1: Introduction to feature extraction and kernel families
- Part 2: Gaussian processes, multiple output Gaussian processes, convolution Gaussian processes and applications in causal analysis and spatial modelling
- Part 3: Natural Language Processing: data wrangling and preparation of text data, multiple output convolutional Gaussian processes models for NLP settings.
Part 1 and 2 of the course will focus on R via RStudio and the tidyverse framework. Part 3 of the course will be demonstrated in Python. All scripts and examples as well as course notes will be provided in advance of the course.
Generally, the level of the course will tend to focus on methodology and providing an introduction for participants to concepts and statistical modelling approaches in the topics mentioned above. Proofs of concepts discussed will be referenced in the slides for participants to follow up afterwards should they wish to explore more technical details.
Speaker 1: Prof. Dr. Gareth W. Peters.(CStat-RSS, YAS-RSE, FIOR)
Chair Professor for Statistics in Risk and Insurance,Director of the Scottish Financial Risk Academy,Department of Actuarial Mathematics and Statistics, Heriot-Watt University, Edinburgh, UK.
Speaker 2: Mr. Ioannis Chalkiadakis (Ph.D. candidate HW)
Department of Mathematics and Computer Science, Heriot-Watt University, Edinburgh, UK.