PhD Studentship in Episodic memory for control
University of Cambridge - Department of Engineering
|Funding for:||UK Students, EU Students|
|Funding amount:||Not specified|
|Placed on:||4th November 2016|
|Closes:||7th December 2016|
|★ View Employer Profile|
Applications invited for a PhD studentship to work on the use of episodic memory for control.
One of the next major challenges for Artificial Intelligence is to develop agents that can successfully learn from interactions with their environments and use the knowledge thus acquired for flexible decision making. All leading high tech companies (Google, Facebook, etc) are heavily investing into such technologies with particular strength in the UK (e.g. Google DeepMind). Although current AI algorithms can now beat humans in select specific domains (e.g. chess, GO, face recognition, etc), they fall far behind humans in domains that require ongoing interaction with the environment and flexible decision making. Thus, understanding how the human brain achieves its unparalleled performance in this more general domain will not only yield important contributions to basic neuroscience but can also provide crucial insights into developing better AI algorithms. The proposed work will build on our earlier success in using machine learning-based approaches to understand the functional advantages of particular aspects of the organisation of the nervous system, which have then been indeed fed back to the development of new record-beating AI algorithms.
The aim of the project is to explore and quantify, by both analytical and numerical techniques, the advantage of episodic memories in AI, and then use this analysis to understand the organisation of human memory into different systems, including episodic, semantic, and procedural memories. We expect two major outcomes for the project. For machine learning, we will develop new algorithms for episodic memory-based control, which will be calibrated against other control algorithms. Note that our earlier work along these lines (Lengyel & Dayan, NIPS 2008) has been recently used by Google DeepMind (Blundell et al, arXiv 2016) to beat state-of-the-art state-of-the-art deep reinforcement learning algorithms algorithms. For neuroscience / cognitive science, we will provide a new understanding of the use and organisation of episodic memories in the context of other memory systems.
Applicants should have, or expected to gain, a high 2:1, preferably a 1st class honours degree in Engineering, Computer Science, Statistics, Mathematics, Physics, Cognitive Science, Psychology, Neuroscience or similar subject. A good knowledge or experience in machine learning is an advantage.
This EPSRC funded studentship is available for Home and EU students. Home students and certain EU students will receive a full studentship including fees and Maintenance. EU students will receive a fees only award. Details on eligibility can be found of EPSRC Web site: https://www.epsrc.ac.uk/skills/students/help/eligibility/ Overseas students are not eligible and should not apply
Applications should be made on-line via the Cambridge Graduate Admissions Office before the deadline: http://www.admin.cam.ac.uk/students/gradadmissions/prospec/apply/ with Máté Lengyel identified as the potential supervisor
The University values diversity and is committed to equality of opportunity.
Share this PhD
Type / Role:
South East England