Picture of James Cussens James Cussens

Picture: On Craig-y-dorth with Sugar Loaf (Mynydd Pen-y-fâl) in the background.

[Research] [PhD supervision] [Projects] [Software] [Teaching] [Professional Activities] [Pastoral and admin roles]
[Personal history] [Contact information]

Funded PhDs available in Bristol from the Centre for Doctoral Training Practice-Oriented Artificial Intelligence (PrO-AI).

I do not take on summer interns or accept students to do MSc by Research, so if you contact me about either of these you will not get a reply.

There appears to me to be a difficulty in this conclusion: that happenings which depend upon an infinite number of cases cannot be determined by a finite number of experiments; indeed nature has her own habits, born from the return of causes, but only 'in general'. And so, who will say whether a subsequent experiment will not stray somewhat from the rule of all the preceding experiments, because of the very mutabilities of things? [Letter from Leibniz to Bernoulli, 3 December 1703. Quoted in: Cussens, Probability and Statistics in Antognazza (ed.) The Oxford Handbook of Leibniz, OUP, 2018.]

Research

GOBNILP software for Bayesian network structure learning

Google scholar profile

University of Bristol research profile

Recent papers
Recent talks

PhD supervision

I am seeking suitably qualified PhD students and have a page with suggested PhD topics. I am seeking suitably qualified PhD students and If you are interested in probabilistic graphical models, integer programming applications in machine learning or statistical relational learning then contact me.

Current students
Former students

Software

GOBNILP

GOBNILP learns Bayesian network structure by encoding the problem as an integer program. Use the following links to get the C/SCIP and Python/Gurobi versions of GOBNILP, respectively:

Some other GOBNILP-related links:
FASTPOLICYTREE
fastpolicytree is an R package available on CRAN which aims to do the same job (constructing optimal policy trees) as the existing policytree R package, but faster. The source code is available on github. It is also possible to create a standalone executable (as long as you have a C compiler). See here for information on how to do this. fastpolicytree was developed as part of the project Tailoring health policies to improve outcomes using machine learning, causal inference and operations research methods 01/07/20-30/06/23. Funded by MRC .
PYADTREE
When learning Bayesian networks from discrete data it is important to construct contingency tables quickly. AD-trees were introduced by Moore and Lee to do this (Cached Sufficient Statistics for Efficient Machine Learning with Large Datasets , Journal of Artificial Intelligence Research 8 (1998) 67-91). A Python module (wrapping C code) called adtree can be obtained via the link below. To install it you need both SWIG and a C compiler installed. In a comparative test using adtree was about 600 times faster than using the groupby.count() method from pandas and about 2500 times faster than simply traversing the data in Python. If adtree is available the (most recent version of the) Python version of GOBNILP uses it to speed up its computation, although sometimes the speedup is modest.

Projects

Teaching

Currently, at Bristol:

Professional Activities

General chair
Programme chair
Invited speaker
Area chair/Senior PC
Co-organiser
PC member / Reviewer
Miscellaneous

Pastoral and admin roles

Personal history

2020-  Senior Lecturer in Artificial Intelligence University of Bristol
2001-2020 Senior Lecturer in the Artificial Intelligence Group University of York
Oct 1997-2001 Lecturer in the Artificial Intelligence Group University of York
Feb 1996-Sept 1997 Researcher on the ILP2 project University of Oxford
1994-1995 Researcher on the ISSAFE project Glasgow Caledonian University
1991-1993 Researcher on the RUBS project King's College London
1990 Researcher on a stepwise refinement project University of Oxford
1986-1989 PhD student in Philosophy of Science King's College London
1983-1986 BSc student in Mathematics University of Warwick

Contact information

Address School of Computer Science, University of Bristol, Merchant Venturers Building, Woodland Road, Bristol, BS8 1UB, UK
Phone +44 (0)117 455 8723
Email firstname.lastname AT bristol DOT ac DOT uk