=Paper=
{{Paper
|id=Vol-2204/paper2
|storemode=property
|title=Improving Probabilistic Rules Compilation using PRM
|pdfUrl=https://ceur-ws.org/Vol-2204/paper2.pdf
|volume=Vol-2204
|authors=Gaspard Ducamp,Philippe Bonnard,Christian De Sainte Marie,Christophe Gonzales,Pierre-Henri Wuillemin
|dblpUrl=https://dblp.org/rec/conf/ruleml/DucampBMGW18
}}
==Improving Probabilistic Rules Compilation using PRM==
https://ceur-ws.org/Vol-2204/paper2.pdf
Improving Probabilistic Rules Compilation using
PRM
Gaspard Ducamp1,2 , Philippe Bonnard2 , Christian De Sainte Marie2 ,
Christophe Gonzales1 , and Pierre-Henri Wuillemin1
1
LIP6 (UMR 7606), Sorbonne Université, 4 place Jussieu, 75005 Paris, France
prenom.nom@lip6.fr
https://www.lip6.fr/
2
IBM France Lab, 9 rue de Verdun, 94250 Gentilly, France
philippe.bonnard@fr.ibm.com, csma@fr.ibm.com, gaspard.ducamp@ibm.com
Abstract. Widely adopted for more than 20 years in industrial fields,
business rules offer the opportunity to non-IT users to define decision-
making policies in a simple and intuitive way. To facilitate their use,
systems known as Business Rule Management Systems have been de-
veloped, separating the business logic from the application one. While
suitable for processing structured and complete data, BRMS face diffi-
culties when those are incomplete or uncertain.
This study proposes a new approach for the integration of probabilistic
reasoning in IBM Operational Decision Manager (ODM), IBMs BRMS,
especially through the introduction of a notion of risk, making the compi-
lation phase more complex but increasing the expressiveness of business
rules.
Keywords: Business Rule · Business Rule Management System · Deci-
sion Making · Bayesian Networks · Bayesian Inference · Object Model ·
Probabilistic Relational Models.
2 G. Ducamp et al.
1 General presentation
Business Rules Management Systems (BRMS), such as IBM Operational De-
cision Manager (ODM), have been introduced in the 90’s to facilitate editing,
authoring, deploying and executing business policies by domain users, in the
form of conditions/actions rules. Syntactically close to the business language,
these ease the translation of decision-making and business strategies, making
them accessible to users with no programming experience. Besides the rule set,
an object data model describes the different objects concerned by the rules, they
are dynamically instantiated in a working memory during the execution of the
program. The activation/execution of rules is managed by inference algorithms
such as RETE [7, 5, 13].
The PhD thesis presented here builds upon work already carried out within
IBM France Lab and the LIP6. It is based on the hybridization of business
rules with probabilistic graphical models such as Bayesian networks [6] or as
Probabilistic Relational Models [1].
A Bayesian network is a compact representation of a joint probability dis-
tribution over a set of random variables. These appear in the form of nodes in
a direct acyclic graph (DAG) where the absence of arcs represent conditional
independences. This type of structure is used as a decision-making tool in many
expert systems and applications [10, 14].
Probabilistic Relational Models (PRM), on the other hand, are combining
notions from Bayesian networks and from the paradigm of object-oriented lan-
guages [12]. The notions of random variables and conditional probabilities are
added to those of classes, relations, interface, inheritance and instantiations.
Such expressiveness makes it possible to answer the problems of reusability and
scalability of graphical models [11].
Several studies including a thesis [4, 1] showed that a loose coupling between
IBM’s BRMS, on the one hand, and probabilistic graphical models (Bayesian
networks initially, then PRM), on the other hand, allowed reasoning about un-
certain data by introducing the notion of probabilistic production rules.
Figure 1 shows how those works fit into the ODM toolchain. Several seman-
tic trees are generated after analyzing and checking the syntax of the user file
describing the rules and the object model (ARL). In addition to producing the
one describing the rules (SemRuleset) these studies proposed to use the object
model to generate a probabilistic graphical model semantic tree (SemPGM ). The
probabilistic expression of rules are then rewritten to link ODM to an inference
engine (aGrUM1 , jSMILE2 , ...) using the generated file (PGM file). Once the
rules are rewritten, the models are compiled before being translated into byte-
code in an archive (JAR), facilitating deployment and execution on the target
machines.
1
aGrUM (A GRaphical Universal Modeler) [15] is an open source C ++ library
for manipulating (learning, modeling, inferring) graphical models and implementing
O3PRM: http: //agrum.gitlab .io
2
SMILE is a reasoning and learning/causal discovery engine for graphical models
https://www.bayesfusion.com/smile-engine
PROCOP 3
Fig. 1. ODM toolchain using probabilistic handling modules
A working example used in those studies is that of a fraud detection applica-
tion, an organization being responsible for managing the reimbursement requests
made by its clients according to their nature (type and cost of reimbursement).
For this purpose, a set of healthcare professionals are in charge of validating the
reimbursement requests made by customers.
Figure 2 shows an example of relation schema for PRM classes of this prob-
lem. A HealthcareProfessional class describes healthcare professionals based on
several characteristics, including their age, location, gender, and the list of clients
they manage (Subscriber), themselves characterized by an age and a list of re-
imbursements requests (Reimbursement).
Fig. 2. Class dependency schema for the insurance example
Generated during compilation, this model allows to answer a first set of
probabilistic rules. In the example below, we will trigger the action part when
we detect a trusted healthcare professional based in Paris who has a client with
a high risk of fraud.
4 G. Ducamp et al.
Ex. 1. An example of probabilistic rule
The previous works raised two major issues:
– Business user friendliness: such rules can be difficult to define and to un-
derstand by a business user, expressing a probability on particular conditions
requiring a deep level of knowledge of the probabilistic models used.
– Performance: ODM provides the ability for users to define different types
of conditions (for example, using filters, aggregators, and nested conditions).
Neither these constructions, more complex, nor their impacts on the perfor-
mances have been studied.
Incidentally, a new inference algorithm [3] has been studied during the pre-
vious thesis but:
– the interest of the use of this algorithm comes from the assumption that the
incremental modifications of the working memory were not going to have big
impacts on the structure of the junction tree used for the inferences, impor-
tant savings of calculations could therefore be realized. It is, consequently,
especially adapted to large connected structures;
– the algorithm works with PRM, but retranscribed in the form of Bayesian
networks (so-called ”grounded”). The interest of PRMs is greatly reduced
since the structural redundancies of these grounded networks, which are
encoded in the PRMs that generated them, are not exploited, which reduces
the efficiency of the inferences;
We notice that the graphical model built during the compilation of ODM
takes into account a probabilized model of the relations between the objects
of the working memory but it does not adapt itself to the probabilistic queries
defined in the rules.
2 A new definition of probabilistic production rules
To address the business user friendliness issue raised above, we need to redefine
the treatment of uncertainty in the expression of rules by replacing the probabil-
ity thresholds attached to single variables by a notion of acceptable risk on the
evaluation of the conditions of the rule as a whole. The action part of a rule will
therefore be executed only if the set of conditions is verified with a probability
greater than the defined acceptable risk. This will allow our probabilistic rules
to be both more complex and intuitive, but it requires a redefinition of the rules
compilation phase to redistribute the overall risk to each individual condition.
PROCOP 5
In the rule below, a rewriting of the previous one, the then part will only
be triggered if the probability that conditions c1 and c2 are true is greater than
80%. Using such kind of formulation will ease the use of probabilistic rules by
business users.
Ex. 2. A new form of probabilistic rule
To achieve this, it is necessary to generate the PRM not only from the object
models but also from the set of rules (see Figure 3). All the predicates over
attributes of a probabilistic object, such as hp.risk==low, will be added in our
graphical model in an extension of the concerned class (red classes in Figure 3).
This requires studying the different forms that conditions can take (aggregations,
filters, nesting, ...) and their predicates. Hence, a feasibility study must be carried
out to define the theoretical and practical limits of such a method.
Fig. 3. Class dependency schema after enhancement based on example 2
The rules will then be rewritten to take into account our new probabilistic
queries (on the probability attribute of the detectInvoiceAnomaly class, for ex-
ample). Such manipulations could, however, change the efficiency of the pattern
matching performed by ODM for the selection of rules to be performed, therefore
their impact should be studied.
6 G. Ducamp et al.
On the PRM side, the dynamic instantiation of classes representing rules (in
green in Figure 3) as well as the simplification of the model generated during
compilation will be at the center of the project (by taking into account that
some of the data sources will be deterministic). In our example the location of
a healthcare professional is considered certain, thus we don’t need to include it
in our new PRM.
Due to the highly deterministic aspect of the user models, we will assume that
the instantiation of our PRMs will yield a large number of small disjoint sub-
graphs (one per health professional, for example), which would require working
on a new inference algorithm taking advantage of this structural redundancy.
This development work is based on what has been done in the Bayesian
Insight Service (BIS) plugin developed as part of the previous thesis. The fig-
ure 4 illustrates how our new module, PRIME (Probabilistic Reasoning Insight
ModulE ), fits into the ODM compilation chain. It intervenes directly during the
process of rewriting the semantic tree describing the rules (SemRuleset) but,
unlike BIS, extends the definition and optimization of the graphical model from
the rules before rewriting them (via the PRM enhancement process).
Once the rules are rewritten, the graphical model is serialized in order to be
usable by a probabilistic engine in parallel with ODM.
Fig. 4. ODM toolchain using PRIME
As part of our work, ODM uses aGrUM for probabilistic calculations and
manipulation of PRMs. The complexity of the compilation, as well as perfor-
mance problems during the execution, might require working on an extension of
this library (advanced aggregations, functional nodes, parametric nodes, ...)
PROCOP 7
3 Application and extensions
This work will provide an extension of the ODM business rule definition lan-
guage to probabilistic concepts. Syntactically and semantically close to natural
language, it will allow non-computer scientists to define probabilistic business
rules in an non-ambiguous and intuitive way.
A study on the use of temporal models via an extension of the PRM model-
ing capabilities in O3PRM3 could be done in a second step, which would allow
a similar reasoning but in the context of the treatment of complex events (Com-
plex Event Processing, CEP [9]). The use of rules with temporal expressions
is at the core of the reasoning capabilities of IBM Decision Server Insights, a
CEP application developed by IBM. This study would therefore be useful at the
theoretical as well as the technical and industrial levels.
Finally, one of the advantages of the expressiveness of PRM is that it is
possible to take into account the structural uncertainty of a model [8]. In doing
so, a study on a predictive extension of ODM could be carried out. The system
would then be able to reason beyond the working memory, allowing to define
new types of rules (anticipation ones, for example).
References
1. Agli, H.: Uncertain Reasoning for Business Rules. PhD thesis. Sorbonne University
(2017)
2. Agli, H., Bonnard, P., Gonzales, C., Wuillemin, P.-H.: Business Rules Uncertainty
Management with Probabilistic Relational Models. RuleML16, Stony Brook, New
York, US (2016). https://doi.org/10.1007/978-3-319-42019-6 4
3. Agli, H., Bonnard, P., Gonzales, C., Wuillemin, P.-H.: Incremental Junction Tree
Inference. IPMU16, Eindhoven, Netherlands (2016). https://doi.org/10.1007/978-
3-319-40596-4 28
4. Ait-Kaci, H. and Bonnard, P.: Probabilistic Production Rules. Technical report.
IBM. (2011)
5. Berstel - Da Silva, B.: Verification of Business Rules Program. PhD thesis. University
of Freiburg (2012)
6. El Ghali, A. and Bonnard, P. and El Ghali, K. and Hromada, D. and Ait-Kaci, H.:
Règles de production et réseaux bayésiens pour lextraction de mots clés. Technical
Report. (2012)
7. Forgy, C. L.: Rete: A fast algorithm for the many pattern/many object
pattern match problem. In: Artificial Intelligence, vol 19, pp.17–37. (1982)
https://doi.org/10.1016/0004-3702(82)90020-0
8. Getoor, L. and Koller, D. and Taskar, B. and Friedman, N.: Learning Probabilistic
Relational Models with structural uncertainty. In: Proc. of the AAAI-2000 Work-
shop on Learning Statistical Models from Relational Data,Technical Report WS-
0006, pp. 13–20. AAAI Press, Menlo Park, CA (200)
9. Luckham, D.: The Power of Events: An Introduction to Complex Event Processing
in Distributed Enterprise Systems. Addison Wesley (2002)
3
O3PRM is a framework developed in the framework of aGrUM allowing the manip-
ulation of PRM: http://o3prm.gitlab.io
8 G. Ducamp et al.
10. Pearl, J.: Probabilistic Reasoning in Intelligent Systems: Networks of Plausible
Inference. Morgan Kaufman (1988).
11. Medina-Oliva, G., Weber, P., Levrat, E., Iung, B.: Use of probabilistic relational
model (PRM) for dependability analysis of complex systems. In: IFAC Symposium
on Large Scale Systems: Theory and Applications, LSS 2010. Villeneuve d’Ascq,
France. (2010)
12. Torti, L., Wuillemin, P.-H., Gonzales, C.: Reinforcing the Object-Oriented Aspect
of Probabilistic Relational Model. In: PGM 2010 - The Fifth European Workshop
on Probabilistic Graphical Models, pp. 273–280. Helsinki, Finland. (2010)
13. Reteplus, the ODM rule execution mode based on the Rete Algorithm.
https://www.ibm.com/support/knowledgecenter/SSQP76 8.9.0/com.ibm.odm.dser
ver.rules.designer.run/optimizing topics/tpc opt reteplusalgo.html
14. Weber, P. and Medina-Oliva, G. and Simon, C. and Iung, B.: Overview on Bayesian
networks applications for dependability, risk analysis and maintenance areas. In:
Engineering Applications of Artificial Intelligence, vol. 25, n. 4, pp. 671 – 682. (2012)
https://doi.org/10.1016/j.engappai.2010.06.002
15. Gonzales, C. and Torti, L. and Wuillemin, P.-H.: aGrUM: a Graphical Universal
Model framework. In: Proceedings of the 30th International Conference on Indus-
trial Engineering, Other Applications of Applied Intelligent Systems. International
Conference on Industrial Engineering, Other Applications of Applied Intelligent
Systems, Arras, France. (2017)