Two postdoc positions on plausible reasoning with vector space embeddin
From email@example.com@21:1/5 to All on Thu Apr 20 00:46:43 2017
Keywords: vector space embeddings, statistical relational learning, knowledge representation, neural networks, explainable AI
Duration: 24 months
Start date: 1 June 2017 (or later this year)
Closing date: 20 May 2017
Applications are invited for two Postdoctoral Research Associate posts at Cardiff University’s School of Computer Science & Informatics:
• The focus of the first position will be on developing methods for exploiting entity embeddings in statistical relational learning, to enable robust plausible reasoning from sparse relational data. Entity embeddings can be used to identify plausible
formulas that are missing from a given knowledge base, intuitively by applying a kind of similarity or analogy based reasoning. Statistical relational learning can also be used to infer plausible formulas, but instead relies on modelling statistical
dependencies among relational facts at the symbolic level. Unifying both methodologies will allow us to develop powerful inference methods that combine their complementary strengths. The resulting method will be applied to zero and one shot learning
tasks, with a focus on automated knowledge base completion.
• The focus of the second position will be on learning vector space embeddings of events and the causal relations between them. In contrast to existing approaches, the learned embeddings will explicitly model which entities participate in the events,
how they are related, and how their relationships are affected by different events. This will require combining ideas from neural network models for event embedding (e.g. based on LSTMs) with ideas from knowledge graph embedding models. Among others, the
resulting model will allow us to uncover more intricate causal relationships, to generate supporting explanations for causal predictions, to incorporate prior knowledge, and to transfer learned knowledge between domains. Intended applications include
recognising textual entailment, stock market prediction, and event-focused information retrieval.
Successful candidates are expected to have a strong background in natural language processing, machine learning, or knowledge representation. This research will be part of Steven Schockaert's FLEXILOG project, which is funded by the European Research
Cardiff University is a member of the Russell Group of research universities, and was ranked 5th in the UK based on the quality of research in the 2014 Research Evaluation Framework. The university has a successful School of Computer Science &
Informatics with an international reputation for its teaching and research activities. Cardiff is a strong and vibrant capital city with good transportation links and an excellent range of housing available.
For more details about the positions, please contact Steven Schockaert (SchockaertS1@cardiff.ac.uk). For instructions on how to apply, please go to www.cardiff.ac.uk/jobs and search for job 5878BR. Please note the requirement to evidence all essential
criteria in the supporting statement.