Edward Bergman joins our group as a D-REAL PhD student for research on automated algorithm selection in Information Retrieval (AutoIR), Recommender Systems (AutoRecSys) and Machine Learning (AutoML)

Published by Joeran Beel on

We welcome Edward Bergman as a new full-time PhD student here at the School of Computer Science and Statistics in Trinity College Dublin, funded through the new D-REAL SFI Centre for Research Training (CRT) and supported by the ADAPT Centre.

Edward has completed a bachelor’s degree in Computer Science in Trinity College Dublin and undertaken a 3-months research internship with Prof. Douglas Leith in Online Optimization techniques. Eddie’s research interests are around automated machine-learning (AutoML), Neural Architecture Search, Algorithm Selection and Hyper Parameter Optimization, as well as the tooling that surrounds them.

Edward Bergman, PhD Student at Trinity College Dublin, focusing on Automated Information Retrieval (AutoIR), Machine Learning (AutoML) and Recommender Systems (AutoRecSys)
Edward Bergman, PhD Student at Trinity College Dublin, focusing on Automated Information Retrieval (AutoIR), Machine Learning (AutoML) and Recommender Systems (AutoRecSys)

His PhD focus will be on automated algorithm selection, a question of inferring what algorithm is best given the information available. What markers are available? How do you approach this in an unsupervised setting? When is this even viable or worth the extra complexity? With the uncertainties that exist, ad-hoc methods can lead to unsatisfactory and uninformative results. Edward will be primarily based at Trinity College Dublin, supervised by Joeran Beel, and co-supervised by Prof Gareth Jones at DCU.

The goal of Edward’s research will be to quantify, analyse and make rigorous these questions in the hopes of pushing forward our current understanding and developing a framework for answering these questions. If these are to be answered, he hopes to develop tools, integrated into the current ecosystem of automated machine learning such that people can benefit from the increased flexibility provided by algorithm selection. For example, the recommendations for experienced users versus new users benefits from dedicated models per user class but a data-driven approach for choosing a model remains a disparate area of discussion.


Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.