Fully funded PhD positions on Knowledge Representation for Learning and Uncertainty (University of Edinburgh)
Short link: http://bit.ly/2f4tYS5

Application deadline: 9 December 2016 (**see below for more information**)

The overall goal of these projects is to develop new methods and formal
languages that can effectively bridge the areas of knowledge representation,
probabilistic reasoning and machine learning. Formal languages and symbolic
techniques have a long and distinguished history in AI, and have widely
impacted many scientific and commercial endeavors in diverse areas such as
verification, robotics, planning, logistics and human-level commonsense
reasoning. However, many of the applications in these areas often need to
handle inherent uncertainty, complemented by an increased prominence of
data-oriented algorithms and statistical techniques. From a foundational
perspective, the question of how knowledge representation languages need to
be augmented to handle these complex notions of uncertainty is an open and
challenging one. From a practical perspective, enriching existing machine
learning algorithms by human-readable representations and background
knowledge can be very useful.

Sample PhD projects include:
- First-order logic has enormous expressive power to represent things like
objects, relations, dependencies, hierarchies and temporal assertions.
Advances in robotics and machine learning, in contrast, learn features of
the world using probabilistic graphical models. An exciting new trend in AI
is to investigate languages (e.g., relational graphical models) and methods
(e.g., model counting) that combine the best of both worlds. In this
context, motivated by vision and language models, where a system encounters
new unknown objects on the fly and is plagued by numerical and qualitative
uncertainty, the research project aims to push the representational
expressiveness and algorithms of existing models to handle unbounded
domains, identity uncertainty, and so on.

- Research in first-order logic for dynamical systems has to led to a large
body of work on high-level programming languages. Building on an ontology of
physical and sensing actions, these languages are interpreted with respect
to a first-order logical database. To be able to apply these languages to
robots, where sensors are typically noisy, the natural question then is: how
can the characterization of the actions and background knowledge incorporate
stochastic uncertainty? What is the relation between such enriched languages
and probabilistic programming, and probabilistic computation, more
generally? This line of work can be seen to contribute to verifiable
behaviors for robotics.

- Automated planning is a major endeavor in AI, where we seek to synthesize
a sequence of actions to enable goal conditions. A recent effort in
automated planning considers the synthesis of plans with rich control
structures such as loops and branches. To be able to apply these languages
to robots, as above, what are the algorithms needed to reason about
stochastic uncertainty while synthesizing such plans?

- A number of more specialized topics on the investigation of symbolic
techniques for machine learning and numerical optimization, and the
application of state-of-the-art constraint solving technology for stochastic
uncertainty, are also possible.

These positions are an opportunity to combine cutting edge research at the
intersection of knowledge representation and machine learning.
We envision the application of these methods to challenging problems arising
in logistics, planning, robotics and commonsense reasoning.

Background Required
The project is suitable for a student with a top MSc or first-class
bachelor's degree in computer science, mathematical logic, statistics,
physics, or a related numerate discipline.

Previous coursework or experience in machine learning and mathematical
logic/knowledge representation is desirable, although we do not expect
students to have both of these.

We envision the development of new software tools that demonstrate the
languages and methods involved, and the application of these methods to
challenging problems arising in logistics, planning, robotics and/or
commonsense reasoning. Therefore, a programming background is desirable.

Why Edinburgh
The School of Informatics at the University of Edinburgh has one of the
largest concentrations of computer science research in Europe, with over 100
faculty members and 275 PhD students. The school is particularly strong in
the research area of artificial intelligence. Our strength in these areas
have been recognized by award of EPSRC Centre for Doctoral Training in Data
Science. The University of Edinburgh is one of the founding partners of the
Alan Turing Institute, the UK's national research institute for data
science.

Funding Information
The scholarship consists of an annual bursary up to a maximum of three
years. Overseas applicants are advised to apply before the standard
informatics deadlines and apply for other scholarships. See
http://www.ed.ac.uk/schools-departments/informatics/postgraduate/fees and
http://www.ed.ac.uk/informatics/postgraduate/apply/key-dates.

Applicants can also consider applying for a combined MSc + PhD programme in
our centre for doctoral training in Data Science and/or Robotics and
Autonomous Systems; see http://datascience.inf.ed.ac.uk and
http://www.edinburgh-robotics.org

Application Information
For informal enquiries about the positions, please contact Vaishak Belle
. Formal application must be through the School's normal
PhD application process:
http://www.ed.ac.uk/studying/postgraduate/degrees/index.php?r=site/view&id=4
94

For more information on CISA, see http://web.inf.ed.ac.uk/cisa/study-with-us

For full consideration, please apply by Dec 9, 2016.