Job PhD Explainable AI for Enabling Repair of Talk with Conversational Agents

From EMCA wiki
Jump to: navigation, search
PhD AI repair & talk
Type Job
Categoryies (tags) Uncategorized
Dates 2018/10/01 - 2018/10/01
Link https://www.nottingham.ac.uk/jobs/currentvacancies/ref/SCI1818
Address
Geolocation 52° 56' 19.0896", -1° 11' 42.5695"
Abstract due 2018/06/30
Submission deadline 2018/06/30
Final version due
Notification date
Tweet #job #phd Microsoft Research PhD Scholarship - Explainable AI for Enabling Repair of Talk with Conversational Agents, DL 30 june
Export for iCalendar

Job PhD Explainable AI for Enabling Repair of Talk with Conversational Agents:


Details:

Microsoft Research PhD Scholarship - Explainable AI for Enabling Repair of Talk with Conversational Agents

We are recruiting a PhD student to carry out research in partnership with Microsoft Research, Cambridge, UK. Preferred candidates are expected to have a background in either Machine Learning (e.g., natural language processing) or Human-Computer Interaction (e.g., ethnographic user research), with an interest and ability to learn the other. The candidate should demonstrate an enthusiasm for interdisciplinary research bridging Artificial Intelligence (AI), Human-Computer Interaction (HCI), and Conversation Analysis. The candidate should be interested in both the development of novel Machine Learning models, as well as their implementation and evaluation in user-centred interactive systems.

This PhD will aim to create goal-directed natural language agents which can interact more effectively with humans by making the agents reasoning more transparent, or explainable, to the user. This can include many different aspects of reasoning, such as "why did you say that?", "what else were you considering saying?", "how confident are you of that?", "how can I correct only one aspect of your understanding without starting from scratch?", etc. The PhD would explore ways to provide this transparency in ways that are natural to the user, and do not interfere with the interaction.To this end the PhD would build on the fact that, as competent human conversationalists, users can adapt a range of repair practices routinely deployed in everyday talk (as extensively documented by Conversation Analysis) and in a limited sense can correct agents by noticing trouble and attempting to repair them through subsequent repetition and rephrasing. The PhD would develop ways for agents to both detect, and make use of such natural language ‘repair’ practices in order to facilitate more natural and more effective communication with human users.

This fully-funded PhD programme offers the unique opportunity to work with a supervision team of researchers at Microsoft (Nate Kushman, Kenton O’Hara and Sean Rintel) and academics at the University of Nottingham (Joel Fischer and Stuart Reeves) for up to four years, funded by EPSRC and Microsoft. The candidate will benefit from a travel and equipment budget and the possibility to do an internship at Microsoft Research in Cambridge as part of the PhD.

The students will benefit from:

  • A fully-funded four-year PhD programme in a leading-edge research project.
  • A potential internship with Microsoft Research.
  • An enhanced stipend of £19,000 per annum as well as a personal laptop and additional resources for travel and equipment to support your study.
  • A world class research environment with a proven track record of successful graduates.

We have funding for UK/Home and EU students, and the minimum entry requirement is a 2.1 undergraduate degree and English language IELTS average score of 6.5 with no less than 6.0 in any element. Application should be made online at https://www.nottingham.ac.uk/pgstudy/how-to-apply/apply-online.aspx with a detailed CV, transcript, references and a statement of research interests (personal statement). Please quote ref SCI1818.

Closing date: 30 June 2018. Start date: between 1 October 2018 and 1 February 2019 (flexible).

Informal enquiries may be addressed to joel.fischer@nottingham.ac.uk (applications by email will not be accepted).