Postdoc positions on ERC project fun2model in 'strong' AI/verification at Oxford
Marta Kwiatkowska
An exciting opportunity has arisen at the intersection of AI and verification: three postdoctoral positions and two fully funded doctoral (DPhil) studentships are available under the supervision of Professor Marta Kwiatkowska on the ERC Advanced Grant project FUN2MODEL (www.fun2model.org), to commence in October 2019 or as soon as possible thereafter.
The FUN2MODEL "From FUNction-based TO MOdel-based automated probabilistic reasoning for DEep Learning" project (www.fun2model.org) aims to make advances towards provably robust 'strong' Artificial Intelligence. In contrast to 'narrow' AI perception tasks realised by deep learning, which are limited to learning data associations, and sometimes referred to as function-based, 'strong' AI aims to match human intelligence and requires model-based reasoning about causality and 'what if' scenarios, incorporation of cognitive aspects such as beliefs and goals, and probabilistic reasoning frameworks that combine logic with statistical machine learning.
The objectives of FUN2MODEL are to develop novel probabilistic verification and synthesis techniques to guarantee safety, robustness and fairness for complex decisions based on machine learning, formulate a comprehensive, compositional game-based modelling framework for reasoning about systems of autonomous agents and their interactions, and evaluate the techniques on a variety of case studies.
The positions are briefly described below; please follow the
links for information about the selection criteria and how
to apply.
Senior Research Associate on FUN2MODEL, fixed term
for 3 years from 1st October 2019, with the possibility of
extension
Grade 8: Salary £40,792 – £48,677 p.a. (note: post may
be under-filled at grade 7: £32,236 - £39,609 p.a.)
The successful appointee will be expected to provide overall leadership for the development of probabilistic verification and synthesis methods, including software implementation and PRISM extensions, with emphasis on data-centric modelling, coordination and reasoning for autonomous multi-agent systems, capturing cognitive and affective aspects. This includes causal reasoning based on Bayesian networks; game-theoretic methods and algorithmic schemes for coordination and collaboration; formalisation of provably robust and beneficial collaboration; extensions of PRISM modelling language and software; and relevant case studies.
Research Associate post 1 on FUN2MODEL, fixed term
for 3 years from 1st October 2019, with the possibility of
extension
Grade 7: Salary £32,236 - £39,609 p.a.
http://www.cs.ox.ac.uk/news/1683-full.html
The successful appointee will be expected to contribute to the development of probabilistic verification and synthesis methods, with emphasis on developing automated probabilistic verification and synthesis methods for machine learning components. This includes Bayesian interpretation; provable probabilistic robustness guarantees for neural networks; provably correct synthesis for neural networks; complex correctness properties for machine learning decisions; software implementation; and relevant case studies.
Research Associate post 1 on FUN2MODEL, fixed term
for 3 years from 1st October 2019, with the possibility of
extension
Grade 7: Salary £32,236 - £39,609 p.a.
http://www.cs.ox.ac.uk/news/1682-full.html
The successful appointee will be expected to contribute to the development of probabilistic verification and synthesis methods, with emphasis on developing an algebraic theory of probabilistic components amenable to machine learning (ML). This includes study of interfaces and algebraic operations for ML components; contract-based probabilistic reasoning for ML components; reasoning about complex ML decisions; integration with autonomous multi-agent system models and reasoning tools; and relevant case studies.
The division of responsibilities between the three research posts may be adapted following interview depending on the qualifications and experience of the candidates.
2 x Doctoral (DPhil) Studentships on FUN2MODEL, 3.5
years from 1st October 2019, with the possibility of
extension
Stipend of at least £15600 per annum p.a, including fees at
EU/home level, laptop and conference travel
For more information about the studentships, selection criteria and how to apply see http://www.cs.ox.ac.uk/news/1681-full.html
Studenship 1: Fairness and bias in multi-agent interactions
Fairness and bias of algorithmic decisions is critical to ensure their acceptance in society, but has been lacking in recently deployed AI software, for example Microsoft’s bot Tay. As a result, a variety of definitions of algorithmic fairness and corresponding verification approaches have been developed. However, these do not capture the influence of the cognitive and affective aspects of complex decisions made by autonomous agents, such as preferences and emotional state, which are essential to achieve effective collaboration of human and artificial agents. This project aims to develop a probabilistic, Bayesian framework based on causal inference for reasoning about fairness and bias in multi-agent collaborations, together with demonstrator case studies and associated software tools.
Studentship 2: Causal reasoning about accountability and blame
While deep learning is able to discern data associations, Bayesian networks are capable of reasoning about counterfactual and interventional scenarios, for example “What if the car had swerved when the child stepped on to the road?”. However, in order to model realistic human behaviours, Bayesian priors and inference must additionally account for cognitive goals and intentions, such as inference of intent for the pedestrian. This project aims to develop a framework for probabilistic causal reasoning with cognitive aspects to study accountability and blame in autonomous scenarios, together with demonstrator case studies and associated software tools.
The successful applicants will join an internationally
leading research group of Professor Marta Kwiatkowska, who has
an extensive track record in probabilistic verification and
pioneering research on safety verification for neural networks
and trust in human-robot collaborations. More information
about Professor Kwiatkowska’s research and PRISM model checker
can be found here:
http://www.cs.ox.ac.uk/marta.kwiatkowska/
https://royalsociety.org/science-events-and-lectures/2018/11/milner-lecture/
https://www.prismmodelchecker.org/
The closing date for all applications is 8 July
2019 (note different procedures for postdocs and
studentships) .
Interviews are expected to be held on 23-24th
July 2019.
Enquiries to Professor Marta Kwiatkowska (marta.kw...@cs.ox.ac.uk) are welcome.
Our staff and students come from all over the world and we proudly promote a friendly and inclusive culture. Diversity is positively encouraged, through diversity groups and champions, for example http://www.cs.ox.ac.uk/aboutus/women-cs-oxford/index.html, as well as a number of family-friendly policies, such as the right to apply for flexible working and support for staff returning from periods of extended absence, for example maternity leave.
-- Professor Marta Kwiatkowska FRS Fellow of Trinity College Department of Computer Science University of Oxford Wolfson Building, Parks Road Oxford, OX1 3QD Tel: +44 (0)1865 283509 Email: Marta.Kw...@cs.ox.ac.uk URL: http://www.cs.ox.ac.uk/people/marta.kwiatkowska/