Department of Computer Science and Technology

Course pages 2019–20

Advanced topics in machine learning and natural language processing

Principal lecturers: Prof Simone Teufel, Prof Mateja Jamnik, Dr Ryan Cotterell, Dr Andreas Vlachos
Taken by: MPhil ACS, Part III
Code: R250
Hours: 16 (8 2-hours sessions)
Class limit: max. 32 students
Prerequisites: L90, L95 and L101 or similar for some topics. Students may find attending the not-for-credit M20 Data Science: Principles and Practice helpful in Michaelmas term.

Aims

This course explores current research topics in machine learning and/or their application to natural language processing in sufficient depth that, at the end of the course, participants will be in a position to contribute to research on their chosen topics. Each topic will be introduced with a lecture which, building on the material covered in the prerequisite courses, will make the current research literature accessible. Each lecture will be followed by up to three seminar sessions which will typically be run as a reading group with student presentations on recent papers from the literature followed by a discussion.

Structure

Each student will attend 3 topics and each topic's sessions will be spread over 5 contact hours. Students will be expected to undertake readings for their selected topics. There will be some group work.

There will be a briefing session in Michaelmas term on Friday 8 November at 14:00.

Syllabus

Students choose exactly three topics in preferential order from a list to be published in Michaelmas term. The following topics have been confirmed.

  • Imitation Learning Dr Andreas Vlachos
  • NLP Graph-based text summarisation Prof. Simone Teufel
  • SVM Prof. Mateja Jamnik
  • Graph neural networks Prof. Pietro Lio'
  • Autoencoders Dr Damon Wischik
  • Graph clustering Dr Luca Zanetti
  • NLP Non-standard NLP (speech, social media and languages other than English) Dr Andrew Caines
  • Bias in datasets Dr Marcus Tomalin
  • Reinforcement learning Dr Amanda Prorok
  • Deep Gaussian processes Prof. Neil Lawrence

Objectives

On completion of this module, students should:

  • be in a strong position to contribute to the research topics covered;
  • understand the fundamental methods (algorithms, data analysis, specific tasks) underlying each topic;
  • and be familiar with recent research papers and advances in the field.

Coursework

Students will work in groups to give a presentation on assigned papers. Each topic will typically consist of one preliminary lecture followed by 3 reading and discussion sessions. A typical topic can accommodate up to 9 students presenting papers. There will be at least 10 minutes general discussion per session.

Full coursework details will be published by October.

Assessment

Coursework will be marked by the topic leaders and second marked by the module conveners.

  • Presentation, Participation/attendance, 30%
  • Topic coursework, 70%

Individual topic coursework will be published on the Assessment tab during late Michaelmas term.

Recommended reading

To be confirmed.