TBC
Lead Research Organisation:
University of Oxford
Department Name: Engineering Science
Abstract
This project falls within the EPSRC Information and Communication Technologies (ICT) research area.
With the explosion of data, as gathered by companies and governments, Machine Learning has enjoyed tremendous success as evident in the technological impact it delivered in recent years. This has been driven by the research and development of techniques in modelling and inference that were combined to build highly functional machines that excel in the task of prediction.
It is our belief that, with its reliance on prediction, there remain significant limitations to the potential of machine learning. This is because predictive, a.k.a. associative, relationships between data suffer from a lack of robustness, explainability, and actionability, which are direct consequences of the relatively weak relationship that association/correlation between data features is.
We believe that investigating the intersection between causal inference and machine learning provides opportunities for improvement on these areas, since causal relationships are more robust against dataset shift (where the environment of application changes), are more explainable (human beings relate many of their concepts causally), and are more actionable (with interventional and counterfactual analysis).
There has been significant independent research in the fields of causal inference and machine learning, but the exciting intersection has been relatively under-explored. With our research, we hope to explore the possibilities to develop models of causal inference harnessing the advanced computation techniques available with machine learning.
The methodology of research will be heavily focused on computational experimentation to establish state-of-the-art techniques for modelling and inference. Our research will focus on the area of a probabilistic graphical modelling approach using Bayesian inference. We believe that the specific choices of modelling and inference fits the use-case of causal inference extremely well. We will explore across a wide-range of datasets and environments to ensure generality of our techniques.
The scope of application of our techniques will be large given the richness of reliable information provided by causal relationships and interventional/counterfactual analyses. It would be exciting to see what advanced computation and inference methods can bring to areas where causal inference is already applied, such as healthcare and economic policy, and to examine new use cases where data is abundant, such as in the biological sciences.
With the explosion of data, as gathered by companies and governments, Machine Learning has enjoyed tremendous success as evident in the technological impact it delivered in recent years. This has been driven by the research and development of techniques in modelling and inference that were combined to build highly functional machines that excel in the task of prediction.
It is our belief that, with its reliance on prediction, there remain significant limitations to the potential of machine learning. This is because predictive, a.k.a. associative, relationships between data suffer from a lack of robustness, explainability, and actionability, which are direct consequences of the relatively weak relationship that association/correlation between data features is.
We believe that investigating the intersection between causal inference and machine learning provides opportunities for improvement on these areas, since causal relationships are more robust against dataset shift (where the environment of application changes), are more explainable (human beings relate many of their concepts causally), and are more actionable (with interventional and counterfactual analysis).
There has been significant independent research in the fields of causal inference and machine learning, but the exciting intersection has been relatively under-explored. With our research, we hope to explore the possibilities to develop models of causal inference harnessing the advanced computation techniques available with machine learning.
The methodology of research will be heavily focused on computational experimentation to establish state-of-the-art techniques for modelling and inference. Our research will focus on the area of a probabilistic graphical modelling approach using Bayesian inference. We believe that the specific choices of modelling and inference fits the use-case of causal inference extremely well. We will explore across a wide-range of datasets and environments to ensure generality of our techniques.
The scope of application of our techniques will be large given the richness of reliable information provided by causal relationships and interventional/counterfactual analyses. It would be exciting to see what advanced computation and inference methods can bring to areas where causal inference is already applied, such as healthcare and economic policy, and to examine new use cases where data is abundant, such as in the biological sciences.
Organisations
People |
ORCID iD |
Stephen Roberts (Primary Supervisor) | |
Lawrence Wang (Student) |
Studentship Projects
Project Reference | Relationship | Related To | Start | End | Student Name |
---|---|---|---|---|---|
EP/R513295/1 | 30/09/2018 | 29/09/2023 | |||
2441798 | Studentship | EP/R513295/1 | 30/09/2020 | 25/06/2024 | Lawrence Wang |
EP/T517811/1 | 30/09/2020 | 29/09/2025 | |||
2441798 | Studentship | EP/T517811/1 | 30/09/2020 | 25/06/2024 | Lawrence Wang |
EP/W524311/1 | 30/09/2022 | 29/09/2028 | |||
2441798 | Studentship | EP/W524311/1 | 30/09/2020 | 25/06/2024 | Lawrence Wang |