A/Prof Kee Siong Ng
Research School of Computer Science,
College of Engineering and Computer Science,
The Australian National University.
keesiong dot ng at anu dot edu dot au
Blog: http://mentalmodels4life.net
About Me
Kee Siong is an experienced data scientist with more than 10 years of
postPhD experience in multiple domains, including government,
financial services, retail, telco, energy, and manufacturing. He has
consulted for many large enterprises in AsiaPacificJapan and acts as
a trusted advisor for many industry colleagues.
Kee Siong has a parttime appointment at ANU. He is also the Chief Data Scientist
at AUSTRAC, Australia's Financial Intelligence Agency.
Research Interests
 Confidential Computing
 InDatabase Machine Learning
 Integrating logic and probability
 Symbolic or knowledgebased machine learning
 Functional and logic programming paradigms
 Uncertainty modelling in AI
 Universal reinforcement learning
Publications
Journal papers
 The MADlib Analytics Library or MAD Skills, the SQL
J.M. Hellerstein, C. Re, F. Schoppmann, Z.D. Wang, E. Fratkin, A, Gorajek, K.S. Ng, C. Welton, X. Feng, K. Li, A. Kumar,
In PVLDB, 2012.
 Probabilities on Sentences in an Expressive Logic
M. Hutter, J.W. Lloyd, K.S. Ng, W. Uther, Journal of
Applied Logic, vol 11(4), pp. 386420, 2011
 A Monte
Carlo AIXI Approximation
J. Veness, K.S. Ng, M. Hutter, W. Uther, D. Silver,
Journal of Artificial Intelligence Research, vol 40, pp. 95142, 2011
 This paper describes a computationally feasible approximation to the universal reinforcement
learning agent AIXI.
 Declarative Programming for Agent Applications
J.W. Lloyd, K.S. Ng, Journal of Autonomous Agents and
MultiAgent Systems, vol 23(2), pp. 224272, 2011.
 This paper describes the computational model of the Bach programming language.
The design of Bach builds on
the lessons learned from the design
of the Godel and
Escher programming languages.
 Probabilistic Reasoning in a Classical Logic
K.S. Ng, J.W. Lloyd, Journal of Applied Logic, vol 7,
pp. 218238, 2009.
 Probabilistic Modelling, Inference and Learning using Logical
Theories
K.S. Ng, J.W. Lloyd, W.T.B. Uther, Annals of Mathematics and
Artificial Intelligence, vol 54(1), pp. 159205, 2008.
 This paper describes how different forms of probabilistic
modelling, inference, and learning described in the
probabilistic logics literature
can be elegantly and simply handled in a standard classical logic:
higherorder logic.
The problem of doing approximate inference in this logic, which
was open at the time of publication, has now been
partially resolved; see Ramana Kumar's project report.
Conference/Workshop papers
 Scalable Entity Resolution Using Probabilistic Signatures on
Parallel Databases
Y. Zhang, K.S. Ng, M. Walker, P. Chou, T. Churchill, P. Christen,
Submitted, 2018.

Exploiting Redundancy, Recurrence and Parallelism: How to Link
Millions of Addresses with Ten Lines of Code in Ten
Minutes
Y. Zhang, T. Churchill, K.S. Ng, In AusDM, 2017.
 Predicting High Impact Academic Papers Using Citation Network Features
D. McNamara, P. Wong, P. Christen, K.S. Ng, In DMApps, 2013.
 Context Tree Switching
J. Veness, K.S. Ng, M. Hutter, M. Bowling,
In IEEE Data Compression Conference, 2012.
 Detecting Noncompliant Consumers in SpatioTemporal Health Data: A Case Study from Medicare Australia
K.S. Ng, Y. Shan, D.W. Murray, A. Sutinen, B. Schwarz, D. Jeacocke, J. Farrugia,
In DMAGI, ICDM Workshops 2010, pp. 613622, 2010.
 Reinforcement Learning via AIXI Approximation
J. Veness, K.S. Ng, M. Hutter, D. Silver, AAAI 2010, pp. 605611.
 Probabilistic and Logical Beliefs
J.W. Lloyd, K.S. Ng, In M. Dastani et al (Eds), LADS 2007, LNAI
5118, pp. 1936, 2008.
 This paper argues for a unified treatment of the notion of
beliefs as traditionally used in probabilistic robotics and
logicbased BDI agents.
 Reflections on Agent Beliefs
J.W. Lloyd, K.S. Ng, In M. Baldoni et al (Eds), DALT 2007, LNCS
4897, pp. 122139, 2007.
 This paper provides some reflections on the form beliefs
should take in logicbased agents that both reason and learn.
 Learning Modal Theories
J.W. Lloyd, K.S. Ng, In S. Muggleton, R. Otero and
A. TamaddoniNezhad (Eds.): ILP 2006, LNAI 4455, pp. 320334, 2007.
 This paper presents a general framework for learning theories in a
higherorder multimodal logic.
 (Agnostic) PAC Learning Concepts in
Higherorder Logic, (A longer preprint.)
K.S. Ng, In J. Furnkranz, T. Scheffer and M. Spiliopoulou (Eds.):
ECML 2006, LNAI 4212, pp. 711718, 2006.
 This paper studies the PAC and agnostic PAC learnability of some
function classes expressible in higherorder logic.
 Generalization Behaviour of Alkemic Decision Trees,
K.S. Ng,
In S. Kramer and B. Pfahringer (Eds.): ILP 2005, LNAI 3625,
pp. 246263, 2005.
 This paper studies the VC dimensions of some common function
classes defined on structured data, including sets, multisets,
trees, graphs, etc.
 Predicate Selection for
Structural Decision Trees, Additional notes
K.S. Ng, J.W. Lloyd,
In S. Kramer and B. Pfahringer (Eds.): ILP 2005, LNAI 3625,
pp. 264278, 2005.
 This paper gives efficient algorithms for the problem of
picking from a structured search space a predicate that partitions
a set of examples well.
 Personalisation for User Agents
J.J.Cole, M.Gray, J.W. Lloyd, K.S. Ng,
In Proceedings of the 4th International Joint
Conference on Autonomous Agents and Multi Agent Systems
(AAMAS05), pp. 603610,
2005.
 This paper presents a symbolic machine learning framework for
achieving personalisation in intelligent user agents.
 Symbolic Learning for Adaptive Agents
J.J. Cole, J.W. Lloyd, K.S. Ng, Proceedings of the Annual Partner
Conference, Smart Internet Technology Cooperative Research Centre,
pp 139148, 2003
 This paper presents an interesting new perspective on relational
reinforcement learning.
 Predictive Toxicology using a Decisiontree
Learner
K.S. Ng, J.W. Lloyd, A.W. Slater, The 20001 Predictive
Toxicology Challenge Workshop, 5th European Conference on Principles
and Practice of Knowledge Discovery in Databases (PKDD01), 2001
Theses
Tech Reports
 A Simple Explanation of Partial Least Squares
K.S. Ng, 2013
 Approximate Inference in Structured Bayesian
Networks
R. Kumar, K.S. Ng, 2009
 Modal Functional Logic Programming
J.W. Lloyd, K.S. Ng, J.W. Veness, Tech Report, 2007
 Agnostic PAC Learning Decision Lists is Hard
K.S. Ng, 2005
 Generalization Bounds for Structural
Decision Trees,
K.S. Ng, 2005
Software
I was a committer for the
opensource Apache MADlib project,
which seeks to foster innovation in largescale indatabase analytics.
Download the latest version of Alkemy.
(Escher is now incorported into Alkemy!!)
Download the latest version of
Escher! See also
the Wikipedia
entry on Escher.
The Godel programming language can be found here.
Bach is a workinprogress, but a
version of the system is available for experimentation on request.
Here's the literate program An Implementation of Bach.
Community Service
 PC Member for the 2013 AAAI Workshop on Statistical Relational Artificial Intelligence
 PC Member for the 2012 ICML Workshop on Statistical Relational Learning
 Member of the Leadership Group for the ACT Chapter of the Analyst First Society
 PC Member for the Solomonoff 85th Memorial Conference
 Coorganiser of a NIPS2009 mini symposium on Partially
Observable Reinforcement Learning
 NICTA Workshop on Integrating Logic and Probability, 2009
 PC Member for 2009 International Workshop on Statistical
Relational Learning (SRL09)
 PC Member for 2006 European Conference on Machine Learning (ECML06)