Personal profile

Research interests

My research is in natural language processing (NLP) and interactive machine learning. I focus on how to learn an NLP model by interacting with a user or aggregating unreliable annotations from multiple sources. Current NLP models usually require many training examples to adapt them to a new task or domain. However, if we could instruct a model what to do differently in a new task, users could quickly take control of its behaviour. When we need to collect annotations, the labels often come from multiple annotators, e.g., from crowdsourcing. I am interested in how to combine these labels to estimate a gold standard, especially for ambiguous labelling tasks with high disagreement between annotators. My work employs Bayesian methods in combination with deep learning to handle uncertainty due to small or unreliable datasets.

Among various applications of my work, I am particularly interested in disaster response, disaster risk reduction and sustainability, where large amounts of valuable information are often stored in unstructured data sources, such as social media, analysts’ reports or satellite imagery. These are prime use cases for interactive methods, as models must adapt to new locations and situations in limited time.

Biography

I joined the University of Bristol in early 2020 as a lecturer (assistant professor), working at the join between natural language processing (NLP) and machine learning. Previously, I was a postdoc at the Ubiquitous Knowledge Processing (UKP) lab at TU Darmstadt, Germany, from 2016 to 2020, where I developed experience in NLP. Prior to that I was a Phd student and then postdoc at the Machine Learning Research group at the University of Oxford, working on Bayesian machine learning methods. Before my PhD, I was a research engineer at HP Labs. I have extensive experience collaborating with industry partners, including Man-AHL and Aleph Insights, as well as research organisations such as The Zooniverse.   

I am responsible for teaching new courses on Text Analytics (Data Science MSc), Dialogue and Narrative (CDT in interactive AI). I am a member of the Intelligent Systems research group at Bristol.

Keywords

Natural language processing, NLP, machine learning, Bayesian methods, interactive AI, interactive learning, crowdsourcing, annotation, preference learning.

Education/Academic qualification

Combined Decision Making with Multiple Agents, University of Oxford

Feb 20101 Aug 2014

Award Date: 1 Aug 2014

External positions

Postdoctoral researcher, TU Darmstadt

Apr 2016Feb 2020

Postdoctoral researcher, University of Oxford

Nov 2013Mar 2016

PhD Student, University of Oxford

Feb 2010Oct 2013

Research Engineer, Hewlett Packard

Apr 2006Oct 2009

Research Groups and Themes

  • Intelligent Systems Laboratory

Keywords

  • natural language processing
  • NLP
  • machine learning
  • Bayesian methods
  • interactive AI
  • interactive learning
  • crowdsourcing
  • annotation
  • preference learning

Fingerprint

Dive into the research topics where Edwin D. Simpson is active. These topic labels come from the works of this person. Together they form a unique fingerprint.
  • 1 Similar Profiles

Collaborations and top research areas from the last five years

Recent external collaboration on country/territory level. Dive into details by clicking on the dots or
  • Low Resource Sequence Tagging with Weak Labels

    Simpson, E., Pfeiffer, J. & Gurevych, I., 12 Feb 2020. 8 p.

    Research output: Contribution to conferenceConference Paperpeer-review

    Open Access
    File
    291 Downloads (Pure)
  • Scalable Bayesian Preference Learning for Crowds

    Simpson, E. & Gurevych, I., 6 Feb 2020, In: Machine Learning. 109, 4, p. 689-718 30 p.

    Research output: Contribution to journalArticle (Academic Journal)peer-review

    Open Access
    File
    22 Citations (Scopus)
    127 Downloads (Pure)
  • A Bayesian Approach for Sequence Tagging with Crowds

    Simpson, E. & Gurevych, I., 7 Nov 2019, Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). ACL, p. 1093-1104 12 p. D19-1101

    Research output: Chapter in Book/Report/Conference proceedingConference Contribution (Conference Proceeding)

    Open Access
    File
    31 Citations (Scopus)
    144 Downloads (Pure)