Tokyo Institute of Technology, Japan
email: sato at mi.cs.titech.ac.jp
Sato (Taisuke) Laboratory
for information not found here
- My research interests:
In the early eighties, I started exploring various aspects of
logic programming such as program analysis, transformation and
synthesis with a purpose to make logic programming more
attractive as a basis for future AI. At that time I found
negation in logic programs quite intriguing and studied
semantics of negation together with applications to logic
program synthesis. Then my interest shifted to non-deductive
information processing such as genetic programming and
inductive/abductive reasoning. In particular uncertainty
inevitable in non-deductive reasoning strongly caught my
attention. I soon began to investigate the possibility of
unifying statistical inference and logical inference in a
general setting. In the mid nineties, I proposed a
probabilistic semantics called the distribution semantics.
Since then I have been developing a probabilistic language
called PRISM based on that semantics, while collaborating
with the B-Prolog project
by professor Zhou at the City University of New York, U.S.A.
and the Lost project by
Professor Christiansen at Roskilde University, Denmark. I'm
also working with Professor Inoue at National Institute of
Informatics for developing logical approaches to systems
- Recent projects:
- JSPS Grant-in-Aid for Scientific Research (B), leader
Knowledge acquisition from uncertain information, 2005.4-2008.3
Uncertainty inference by probabilistic knowledge modeling, 2008.4-2010.3
Uncertainty modeling by statistical abduction, 2011.4-2014.3
- JSPS Grant-in-Aid for Scientific Research (A), member
Inference-based Hypothesis-finding and its Applications to Systems Biology,
- JST-CNRS Japanese-French Cooperative Program, member
Knowledge-based Discovery in Systems Biology, 2007-2009
- J. Logic Programming, editorial advisor (1998-2000)
- New Generation Computing, associate editor (1998-2008),
- The Japanese Society for Artificial Intelligence,
- Theory and Practice of Logic Programming,
advisory board member (2004-)
- program chair, annual national conference on AI, Japan (1998)
- co-chair, FLOPS'99
- PC member of AISC'12, PPDP'12, ECAI'12, ACML'09,10, PPDP'09,04, LLLL'09,07,06,05, ILP'09, SRL'09,
ECML/PKDD'08, FLOPS'08(steering committee), MLG'07,
IJCAI'05, AIAN04, ICML'03,02
- Short CV:
I received M.S. in Electrical Engineering in 1975 and also
received Ph.D. in Computer Science in 1987, both from Tokyo
Institute of Technology. The title of my doctoral thesis was
"Declarative Logic Programming." I joined ETL
(Electrotechnical Laboratory), Japan in 1975 as a researcher
and soon was involved the Fifth Generation Project in various
ways. In those days, my research interest was in equivalence
preserving logic program transformation and its extension to
the case of general programs. In 1995 I moved to Tokyo
Institute of Technology as a professor, and at the same time I
changed my research direction toward non-deductive approaches
to AI. In particular I concentrated on combining the idea of
probabilistic logic programming and statistical learning.
Later in 1997 I designed a probabilistic logic programming
language PRISM. Since then I've been concentrating on the
development of PRISM as a high level probabilistic modeling
language. Currently I'm investigating the possibility of
further generalizing PRISM by a constraint approach.
- Current research topics:
- Statistical abduction/PRISM:
This is an abductive framework for machine leaning that unifies
symbolic computation and statistical learning at semantic level
based on the distribution semantics. It
subsumes most of discrete generative models including Bayesian
networks, HMMs and PCFGs. As an implementation of the
distribution semantics, we have been developing a probabilistic
logic programming language PRISM since
1997. PRISM programs look like Prolog programs containing
probabilistic atoms and we can easily write probabilistic
models declaratively at first-order level. Probability
computation and parameter leaning are automatically carried
out by PRISM, efficiently using dynamic programming combined
with tabling. We recently introduced
variational Bayes inference
and a general MCMC method
to perform Bayesian inference.
- Constraint based probabilistic modeling:
CBPMs (constraint based probabilistic models) are a recently
introduced new probabilistic modeling framework. They are
propositional probabilistic models that define conditional
distributions P(x | KB) in which KB, a set of clauses,
constrain probabilistic behaviors of independent boolean atoms
as we desire. They can represent cyclic interdependencies in
probabilistic events, thus suitable for describing cyclic
systems such as metabolic networks. We
adopt probabilistic BDDs as a base data
structure for CBPMs and derived an EM algorithm for them.
- Genetic programming:
Genetic programming provides a convenient way to the acquisition
of control knowledge. We applied it to the synthesis of
semaphores and the construction of regional control of traffic
- Selected publications :
(see Sato Lab.
for a complete list of publications)
- (Logic Programming)
Enumeration of Success Patterns in Logic Programs,
Sato,T. and Tamaki,H.,
Theoretical Computer Science 34, 1984, pp.227-240.
Unfold/fold Transformation of Logic Programs, Tamaki,H. and Sato,T.,
Proc. 2nd Int'l Conf. on Logic Programming,
Uppsala, 1984, pp.127-137.
OLD Resolution with Tabulation, Tamaki,H. and Sato,T.,
Proc. 3rd Int'l Conf. on Logic Programming,
London, 1986, pp.84-98.
First Order Compiler: A Deterministic Logic Program Synthesis Algorithm,
Sato,T. and Tamaki,H.,
J. Symbolic Computation 8, 1989, pp.605-627.
Completed Logic Programs and Their Consistency, Sato,T.,
J. Logic Programming 9, 1990, pp.33-44.
Equivalence-Preserving First Order Unfold/fold Transformation Systems,
Theoretical Computer Science 105, 1992, pp.57-84.
Linear tabling strategies and optimization,
Zhou, N.-F., Sato, T. and Shen, Y.-D.,
Theory and Practice of Logic Programming, Vol.8, No.1, 2008, pp.81-109.
- (Genetic Programming)
BUGS: A Bug-Based Search Strategy using Genetic Algorithms,
Iba,H., Akiba,S., Higuchi,T. and Sato,T.,
Proc. PPSN (Parallel Problem Solving from Nature),
Belgium, 1992, pp.165-174.
Evolutionary Learning Strategy using Bug-Based Search,
Iba,H., Higuchi,T., de Garis,H and Sato,T.,
Proc. IJCAI'93, Shambery, 1993, pp.960-966.
A Numerical Approach to Genetic Programming for System Identification,
Iba,H., deGaris,H. and Sato,T.,
Evolutionary Computation 3(4), 1996, pp.417-452.
- (Statistical abduction/PRISM)
A Statistical Learning Method for Logic Programs with
Proc. ICLP'95, 1995, pp.715-729.
PRISM:A Language for Symbolic-Statistical Modeling,
Sato,T. and Kameya,Y.,
Proc. IJCAI'97, 1997, pp.1330-1335.
Parameter Learning of Logic Programs for
Sato,T. and Kameya,Y.
Journal of Artificial Intelligence Research Vol.15, 2001, pp.391-454.
Variational Bayes via propositionalized probability computation in PRISM,
Sato, T., Kameya, Y., Kurihara, K.,
Annals of Mathematics and Artificial Intelligence, Vol.54, No.1-3,
A general MCMC method for Bayesian inference in logic-based probabilistic modeling,
Proc. IJCAI'11, 2011, pp.1472-1477.
- See also
- (Constraint based Probabilistic Modeling)
Evaluating abductive hypotheses using an EM algorithm on BDDs,
Inoue, K., Sato, T., Ishihata, M., Kameya, Y. and Nabeshima,H.,
Proc. IJCAI'09, 2009, pp.810-815.
Constraint-based probabilistic modeling for statistical abduction,
Sato, T., Ishihata, M. and Inoue, K.,
Machine Learning, Vol.83, No.2, pp.241-264, 2011.