Upload
cognitum
View
980
Download
3
Embed Size (px)
Citation preview
1The company, product and service names used in this web site are for identification purposes only.
© Cognitum 2014. All trademarks and registered trademarks are the property of their respective owners.
Cognitum Ontorion: Knowledge Representation and Reasoning
System
2The company, product and service names used in this web site are for identification purposes only.
© Cognitum 2014. All trademarks and registered trademarks are the property of their respective owners.
Motivation
Modern KRR based on the set of W3C OWL/RDF technologies can be seen as a Smart Graph Database
However if extended with certain functionalities it starts to fulfill the definition of Multi Agent System
We can use it for simulations
3The company, product and service names used in this web site are for identification purposes only.
© Cognitum 2014. All trademarks and registered trademarks are the property of their respective owners.
Multi-Agent & Simulation systems (MASS)
Deductive Reasoning Agents:1. use logic to encode a theory defining the best
action to perform in a given situation,
2. are the “purest” in terms of their formal specification.
Unfortunately:1. the complexity of theorem proofs
2. the boundaries of expressivity
Making deductive reasoning requires the selection of underlying logics
4The company, product and service names used in this web site are for identification purposes only.
© Cognitum 2014. All trademarks and registered trademarks are the property of their respective owners.
Multi-Agent & Simulation systems (MASS)
“Reactive agent movement” - an era of reactive agent architecture.
The reactive agent movement manifested in the form of requirements for so-called behavior languages
1. Intelligent behavior can be generated without explicit representations of the kind that symbolic AI proposes.
2. Intelligent behavior can be generated without explicit abstract reasoning of the kind that symbolic AI proposes.
3. Intelligence is an emergent property of certain complex systems.
Agent Oriented Programming (AOP), e.g. JADE
Unfortunately: Lack formal foundations => hard to analyze with formal methods and tools.
5The company, product and service names used in this web site are for identification purposes only.
© Cognitum 2014. All trademarks and registered trademarks are the property of their respective owners.
Multi-Agent & Simulation systems (MASS)
Hybrid Agent Architecture, Attempts to combine the best of symbolic and reactive
architectures.
Two subsystems: (1) a symbolic world model that allows plans to be developed and
decisions made
(2) a reactive engine which is capable of reacting to events without involving complex reasoning.
Ferguson’s “TouringMachines” - three separate control layers: a reactive layer,
a planning layer,
a modelling layer.
The three layers are concurrently-operating, independently-motivated, and activity
6The company, product and service names used in this web site are for identification purposes only.
© Cognitum 2014. All trademarks and registered trademarks are the property of their respective owners.
Our approach
We present an approach to hybrid agent architecture, which is implemented on top of a scalable Knowledge Representation & Reasoning (KRR) system.
KRR here:1. Allows environment to be described formally
2. A reactive agent system is based on a rule triggering subsystem.
3. Reasoning itself allows for agent synthesis
7The company, product and service names used in this web site are for identification purposes only.
© Cognitum 2014. All trademarks and registered trademarks are the property of their respective owners.
Our approach
Here, we consider a reactive agent that is able to maintain its state.
This agent has an internal stand-alone data structure, which is typically used to:
record information about the state and history of the environment
and to store a set of all the internal states of an agent.
Agent
see
next
action
Environment
state
8The company, product and service names used in this web site are for identification purposes only.
© Cognitum 2014. All trademarks and registered trademarks are the property of their respective owners.
OntorionSmart Knowledge Management System
9The company, product and service names used in this web site are for identification purposes only.
© Cognitum 2014. All trademarks and registered trademarks are the property of their respective owners.
Ontorion
Ontorion is a Distributed Knowledge Management System that allows semi-natural language to be used to specify and query the knowledge base.
It also has a built-in engine trigger which fires the rules each time if the corresponding knowledge is modified.
By design, Ontorion allows one to build large, scalable solutions for Semantic Web.
The symmetry of the architecture of the cluster provides system scalability and flexibility:
Ontorion can be deployed and executed in a computing cloud environment, where the total number of nodes can be changed on request, depending on user requirements.
10The company, product and service names used in this web site are for identification purposes only.
© Cognitum 2014. All trademarks and registered trademarks are the property of their respective owners.
General Ontorion Architecture
Ontorion
nodenodenodenode A
lert
sExt
ern
al
Stre
am
CN
L O
nto
log
y
11The company, product and service names used in this web site are for identification purposes only.
© Cognitum 2014. All trademarks and registered trademarks are the property of their respective owners.
Controlled Natural Language in FECNL is a subset of natural language with restricted grammar and vocabulary
in order to reduce the ambiguity and complexity inherent in full natural language
Ontology OWL 2 + SWRLControlled Englishin FluentEditor
Controlled English (CE) in Fluent Editoris automatically translated into and from description
logic OWL 2, SWRL
12The company, product and service names used in this web site are for identification purposes only.
© Cognitum 2014. All trademarks and registered trademarks are the property of their respective owners.
Exmple of ontology in CNL
January is a month and is-in-quarter equal-to 1.February is a month and is-in-quarter equal-to 1.March is a month and is-in-quarter equal-to 1.April is a month and is-in-quarter equal-to 2.May is a month and is-in-quarter equal-to 2.June is a month and is-in-quarter equal-to 2.July is a month and is-in-quarter equal-to 3.August is a month and is-in-quarter equal-to 3.September is a month and is-in-quarter equal-to 3.October is a month and is-in-quarter equal-to 4.November is a month and is-in-quarter equal-to 4.December is a month and is-in-quarter equal-to 4.
Year-2011 is a year.Year-2012 is a year.Year-2013 is a year.Year-2014 is a year.
If X is-inside Y then Y contains X.
Usa is a country.Canada is a country.
California is a place and is-inside Usa.New-York is a place and is-inside Usa.Washington is a place and is-inside Usa.Ontario is a place and is-inside Canada.Quebec is a place and is-inside Canada.
Every chromebook is a computer-type.Every sleekbook is a computer-type.Every laptop is a computer-type.Samsung is a vendor.Toshiba is a vendor.Gateway is a vendor.Lenovo is a vendor.Dell is a vendor.Acer is a vendor.Asus is a vendor.Hp is a vendor.Touchsmart is a family.Satellite is a family.Elitebook is a family.Alienware is a family.Inspiron is a family.Pavilion is a family.Thinkpad is a family.Qosimo is a family.Aspire is a family.Envy is a family.Intel is a cpu-vendor.Amd is a cpu-vendor.Ssd is a disk-type.Hdd is a disk-type.Solid-State-Drive is a disk-type.Flash-Drive is a disk-type.Hard-Drive is a disk-type.The-"windows-8.1" is an os.Chrome-Os is an os.Windows-7 is an os.Windows-8 is an os.
Computer-1 is a laptop.Computer-1 is-produced-by Hp.Computer-1 has-diagonal-in-inches equal-to 15.6.Computer-1 has-cpu-produced-by Intel.Computer-1 has-cpu-model equal-to 'Intel Pentium N3520'.Computer-1 has-ram-in-gb equal-to 4.Computer-1 has-disk-capacity-in-gb equal-to 500.Computer-1 has-disk-type Hdd.Computer-1 has-os The-"windows-8.1".Computer-1 has-color equal-to 'black licorice'.
13The company, product and service names used in this web site are for identification purposes only.
© Cognitum 2014. All trademarks and registered trademarks are the property of their respective owners.
Ontology of dimentions
Satellite
Computer-1
os
June
January
Acer
Computer-39
Computer-38
Computer-49
Computer-48
Computer-31
Computer-30
Computer-33
Computer-32
Computer-35
Computer-34
Computer-37
Computer-36
Computer-42
Computer-45
Hdd
Computer-47
Computer-46
laptop
Computer-67
New-York
September
May
Computer-7
Year-2011
Computer-8
July
Pavilion
Elitebook
Computer-2
Inspiron
computer-type
February
Touchsmart
December
Computer-41
Intel
Qosimo
Computer-59
Computer-58
cpu-vendor
Lenovo
Computer-51
Computer-50
Computer-53
Computer-52
Computer-55
Computer-54
Computer-57
Computer-56
Computer-44
Computer-77
vendor
Samsung
Computer-9
August
placeyear
"thing"
Computer-4
sleekbook
Asus
Usa
Computer-80
Computer-5
Computer-21
month
Computer-69
Computer-68
Thinkpad
Computer-61
Computer-60Computer-63
Computer-62Computer-65
Computer-64
November
Computer-66
Computer-19
Computer-18
Computer-76
Computer-11
Computer-10
Computer-13
Computer-12
Computer-15
Computer-14
Computer-17Computer-16
Hard-Drive
Flash-Drive
Windows-8
Windows-7
Year-2014
Year-2013
April
Dell
Solid-State-Drive
Hp
disk-type
October
Quebec
Computer-3
Chrome-Os
Ssd
Computer-40
Ontario
Computer-43
Amd
Aspire
Computer-79
Computer-78
"windows-8.1"
Computer-71
Computer-70
Computer-73
Computer-72
Computer-75
Computer-74
California
country
Toshiba
Computer-29
Computer-28
Canada
Washington
Envy
Computer-20
Computer-23
Computer-22
Computer-25
Computer-24
Computer-27
Computer-26
chromebook
March
family
Alienware
Computer-6
Year-2012
Gateway
14The company, product and service names used in this web site are for identification purposes only.
© Cognitum 2014. All trademarks and registered trademarks are the property of their respective owners.
The Agent-Move Rule (MASS definition)
A „move” function as a parameter takes a percept which is a result of a reasoning process over the current state of the environment
We consider reasoning as an implementation of the abstract „see” function
A single agent is determined by its state and all the „move” functions that can be ever executed in the context of its state, therefore here, the agent synthesis process, is a process of the assignment of the „move” functions to the single agent-individual.
The body the „move” function determine the specific environment state that allows the system to assign the function to the agent.
If ... then ... execute <? Move(agent, "current-state", message, "expected-message", ()=> {
/* the agent action */ return "new-state"; });?>.
Agent
see
next
action
Environment
state
15The company, product and service names used in this web site are for identification purposes only.
© Cognitum 2014. All trademarks and registered trademarks are the property of their respective owners.
Non deterministic behaviour
The overall behaviour of MASS is non-deterministic.
This is due to the fact that the concrete run (the MASS run) needs the selection of agent instances made in runtime – and runtime (in opposition to reasoning time) is a part of the reactive model that is non-deterministic by nature.
The non-deterministic selection of choices, often by use of pseudo-random number generators, and the parallel execution of different threads, are required by underlying technologies to provide an efficient computational model.
Therefore, we need to keep in mind, that simulations based on the reactive/hybrid approach are non-deterministic.
In this case, we have to perform a large set of experiments with the same initial state and then use analytical tools and methods to verify the statistical hypothesis.
16The company, product and service names used in this web site are for identification purposes only.
© Cognitum 2014. All trademarks and registered trademarks are the property of their respective owners.
Experimental Setup
Software Process Simulation Modelling (SPSM) is widely used nowadays to support planning and control during software development.
MA systems play a very important role here as they naturally can be used to simulate social behaviors in the software testing phase.
In our approach, the SPSM is divided into two components: ontology and knowledge modification triggers. In the example given below, the
ontology defines (with CNL) the core concepts such as: competency, task, developer, manager:
We also defined agent-rules by making use of knowledge modification triggers,
Those triggers implement the following scenario: a developer with certain competencies starts to realize a task.
After the task is finished, new knowledge about the task realization process is added, and a “Busy” state is set on the developer.
The second trigger is fired when the task is finished and a “Ready” state is set back.
17The company, product and service names used in this web site are for identification purposes only.
© Cognitum 2014. All trademarks and registered trademarks are the property of their respective owners.
Experimental Setup
If a wake-up-message has-target a developer and the wake-up-message has-origin the developer and the wake-up-message has-task-realization-query a task-realization-query and the task-realization-query has-origin a task then for the wake-up-message and the developer and the task-realization-query and the task execute <? Move(developer, "Busy", wake_up_message, "WakeUp", ()=> {// modify the status of task KnowledgeInsert(task+" has-status Done.");
// mark the agent state as ready return "Ready"; });
?>.
Cpp-Programming is a competency.Java-Programming is a competency....Task-0 is a task.Task-1 is a task.Task-1 is-dependent-on Task-0.Task-1 requires-competency Cpp-Programming.Task-1 has-estimated-realization-md equal-to 500.Task-2 is a task.Task-2 is-dependent-on Task-1.Task-2 requires-competency Java-Programming.Task-2 has-estimated-realization-md equal-to 500....Anna is a developer .Anna has-competency Cpp-Programming .Anna has-competency Java-Programming .
John is a developer .John has-competency Java-Programming .John has-competency Web-Programming ....
If a task-realization-query requires-competency a competency and a developer has-competency the competency and the task-realization-query has-origin a task then for the task-realization-query and the developer and the task execute <? Move(developer, "Ready", task_realization_query, "Programming", ()=> {// read the realization timevar realizationTime = (from v in Values where v.source==InstanceDL(task) && v.datarole=="have-estimated-realization-md" select v.value). FirstOrDefault(). SetConsistencyLevel(ConsistencyLevel.Quorum). Execute();
// create the wake-up message var msgid = CreateMessage(developer, "WakeUp",task); // delayed (by the realization time) modification // of KB KnowledgeInsertWithDelay( msgid + " is a wake-up-message."+ msgid + " has-origin " + developer + "."+ msgid + " has-task-realization-query " + task_realization_query + "."+ msgid + " has-target "+ developer + ".", int.Parse(realizationTime));
// mark the agent state as busy return "Busy"; });?>.
18The company, product and service names used in this web site are for identification purposes only.
© Cognitum 2014. All trademarks and registered trademarks are the property of their respective owners.
Trigger the simulation start
If a start-event exists then for the start-event execute <? Once("Lets the simulation start.", ()=> { CreateAgent("Mark","Ready"); CreateAgent("John","Ready"); CreateAgent("Tom","Ready"); CreateAgent("Gabi","Ready"); CreateAgent("Anna","Ready");
KnowledgeInsert("Task-0 has-status Done."); });?>.
Start-Event is a start-event.
19The company, product and service names used in this web site are for identification purposes only.
© Cognitum 2014. All trademarks and registered trademarks are the property of their respective owners.
SPSM Simulation Results
20The company, product and service names used in this web site are for identification purposes only.
© Cognitum 2014. All trademarks and registered trademarks are the property of their respective owners.
Conclusion
1. In this paper, we showed that the Ontorion server is able to execute, maintain and control massive simulations based on a hybrid MASS approach.
2. The resulting MASS fits well into the definition of a Hybrid MASS.
3. An expressive and distinct Ontorion functionality is the ability to encode agent logics in semi-natural language in the interest of a less professional user.
4. We also observed that this allows end users to understand actions taken by an agent, even if a user is not well trained in formal representation systems.
21The company, product and service names used in this web site are for identification purposes only.
© Cognitum 2014. All trademarks and registered trademarks are the property of their respective owners.
Future work
An agent synthesis benefits from a formal reasoning engine (being a central component of KRR) and is based on an action-selection procedure
22The company, product and service names used in this web site are for identification purposes only.
© Cognitum 2014. All trademarks and registered trademarks are the property of their respective owners.
Try it!
Ontorion is free of charge for academic institutions and independent researches. For more information, please visit our website available at http://www.conitum.eu/semantics/.
23The company, product and service names used in this web site are for identification purposes only.
© Cognitum 2014. All trademarks and registered trademarks are the property of their respective owners.
Thank you