Dr Nicolas Verstaevel presented a seminar titled "Keep it SMART, keep it simple! – Challenging complexity with self-organising software" as part of the SMART Seminar Series on 24th July 2018.
More information: https://news.eis.uow.edu.au/event/keep-it-smart-keep-it-simple-challenging-complexity-with-self-organising-software/
Keep updated with future events: http://www.uoweis.co/events/category/smart-infrastructure-facility/
Similar to SMART Seminar Series: "Keep it SMART, keep it simple! – Challenging complexity with self-organising software". Presented by Dr Nicolas Verstaevel
Similar to SMART Seminar Series: "Keep it SMART, keep it simple! – Challenging complexity with self-organising software". Presented by Dr Nicolas Verstaevel (20)
7. The computer of the 21st Century
7
« The most profound technologies are those that
disappear. They weave themselves into the fabric of
everyday life until they are from it. »
Mark Weiser. ‘The computer of the 21t century’
Scientific American, pp 94-10, September 1991
8. Smart Cities
• Usage of ICT to transform the way people live and work
• Addressing citizen needs
• Community involvement
8
10. Challenging properties
10
• Multipleassets to manage
– Energy, water, comfort,waste,…
– Interdisciplinary
• Composed of variousheterogeneoussystems
• Strong politico-socialcomponents
• Open
• Heterogeneity of devices
• Unpredictabledynamics
• Users with dynamic and contradictoryneeds
➢ Impossibilityto specify a priori all interactions
Me
Me
Me
12. — There are many undecidable problems
— No general method will ever be found to solve
them
— (without simplifying the problem or
employing a more ‘powerful’ theory)
Complex artificial systems
Notion of undecidability1
1
an undecidable problem is a decision problem for which it is known to be impossible to constructa single algorithm that
always leads to a correct yes-or-no answer.
12
13. Halting problem: there is no algorithm that
correctly determines whether arbitrary
programs eventually halt when run.
“Program testing can be used to show the presence
of bugs, but never to show their absence! ”
Verification & Proof
No guaranty
Edsger Dijkstra
Alan Turing
13
14. Ashby’s Law
The variety in the control
system must be equal to or
larger than the variety of the
perturbationsin order to
achieve control
System Controller
No guaranty
Control System
Command
State
W. Ross Ashby
14
15. No free lunch theorem for
search and optimization
Any two algorithms are
equivalent when their
performance is averaged
across all possible problems
Machine Learning
No guaranty
David H. Wolpert
Learning
Control System
Command
State
Objective
15
16. Meta learning is a subfield of Machine learning where automaticlearning
algorithms are applied on meta-data about machine learning experiments
➢ It thereforeaims for a universal controlmethod able to learn the
matching “situation-learningalgorithm” pairs
Meta-Learning
No guaranty
Learning
Control System
Command
State
Algorithm
selection
16
17. Impact of complexity in artificial systems
➢ More and more energy is
spent to build systems that
are less and less generic and
which must be regularly
revalidated
➢ Without ever having the
guaranteeof an impeccable
operation.
➢ A conceptual breakthrough:
➢ The discovery of micro level
theories (where complexity is
reduced) which lead to the
appearance of the desired
phenomena at the macro level
17
18. Self-organising software
18
Serugendo, Giovanna Di Marzo, Marie-
Pierre Gleizes, and AnthonyKarageorgos,
eds. Self-organising software: From
natural to artificial adaptation. Springer
Science & Business Media, 2011.
Definition: Self-organization is the process enablinga
software to dynamicallyalter its internal organization
(structure and functionality)during its execution time
without any explicit external directing mechanism.
20. 20
https://scratch.mit.edu/projects/26099509/
- 2 Light sensors
- 2 motors
Speed of the motor is
proportional to the
sensed light according to
the connection
Braitenberg vehicles
ValentinoBraitenberg(1984)
Braitenberg, V. (1984). Vehicles: Experiments in synthetic psychology. Cambridge, MA: MIT
Press.
22. Multi-agent system and Complexity
• Solution to overcome this complexity?
• Agents’ complexity is lower
– Behaviours can be expressed in few
rules
• MAS as an engineering methodology for
complex systems
– Bottom-up design
– Can exhibits self-* properties
– Emergent properties
– Openness
– Heterogeneity
22
23. Perceive
Act
System
Environment
An agent
1. Define the systemand
the environment
2. Identify the interaction
between the systemand
the environment
3. Identify agents and their
interactions
4. Discover failures in the
interactions (NCS)
5. Design simple local rules
to repair NCS
AMAS Designer
TO DO LIST
An interaction
AMAS - Adaptive
Multi-Agent Systems
Davy Capera, Jean-Pierre Georgé, Marie Pierre Gleizes, Pierre Glize:
The AMAS theory for complex problem solving based on self-organizing cooperative
agents. WETICE2003: 383-388
24. Cooperation as the motor of
self-organization
24
Perception
•Incomprehension
•Ambiguity
Decision
•Incompetence
•Unproductivity
Action
•Concurrence
•Conflict
•Inutility
7 Types of non cooperative situations
Anticipateor resolve NCS by:
- Tuning: adjusts its internal
parameters.
- Reorganization:changing
interactionwith other
- Openness: addingor removing
agents
26. Problem: Challenges:
Surveillance of an area by a robotic fleet
• Robotic fleet composed of 𝑛
robots
• Evolves in a grid environment
• How to optimise space
occupancy?
• How to deal with emergencies?
• Scalability
– Number of robots
– Different environments
• Dynamic
– Unpredictableemergencies
26
28. Surveillance of an area by a robotic fleet
• 2 type of agents:
– Robots
– Cells
• Local perception of a shared
environment
• Cells compute criticality
accordingto own knowledge
and time sine last visit
– Low criticality
– High criticality
28
1 0 0 0
3 2 6
1 2 1 2
0 4 0
Robotsbehaviouris described as a
simple set of cooperativerules to
maintainfunctionaladequacy
𝐶 = 𝑡𝑙𝑎𝑠𝑡_𝑣𝑖𝑠𝑖𝑡
29. 29
Robots
Low criticality cell
High criticalitycell
Alexandre Perles, Fabrice
Crasnier, Jean-Pierre Georgé:
AMAK - A Framework for
Developing Robustand Open
Adaptive Multi-agent
Systems. PAAMS
(Workshops) 2018: 468-479
38. Linear optimisation
38
𝑥1 𝑥2
2𝑥1 + 𝑥3 ≥ 2𝑥1 + 𝑥2 ≤ 2
𝑥31 0 1
-1.5 -1 𝑥3 ≤ 2-1
== =
Jorquera, Tom. An adaptive multi-agent
system for self-organizing continuous
optimization. Diss. Université de Toulouse,
Université Toulouse III-Paul Sabatier, 2013.
39. Detection of anomalies in the context of elderly people
40
∑
𝑥1
𝑥2
𝑥 𝑛
ω1
ω2
ω 𝑛
> T
A
B
yesno
*
*
*
𝑦 = 𝑓(ω. റ𝑥) = 𝑓(
𝑛
𝑥 𝑛 ∗ ω 𝑛)
Nicolas Verstaevel, Carole Bernon, Jean-Pierre Georgé and Marie-Pierre Gleizes. A Self-OrganizedLearning
Model for AnomaliesDetection:Applicationto Elderly People.12th IEEE International Conference on Self-Adaptive
and Self-Organizing Systems, SASO 2018 [to be published]
Feedback
40. Detection of anomalies in the context of elderly people
41
Nicolas Verstaevel, Carole Bernon, Jean-Pierre Georgé and Marie-Pierre Gleizes. A Self-OrganizedLearning
Model for AnomaliesDetection:Applicationto Elderly People.12th IEEE International Conference on Self-Adaptive
and Self-Organizing Systems, SASO 2018 [to be published]
Steps
Number of sensors (0-100)
Number of sensors (0-100)
Numberofexamples
Number of step to reach 1% or precision in weightestimation Number of example needed to reach 1% of precision
42. 43
Boes,Jérémy,et al. "The self-adaptive context learning pattern: Overview and
proposal." International and InterdisciplinaryConference on Modelingand UsingContext.
Springer, Cham, 2015.
Control Buildingmodels
43. 44
Verstaevel, Nicolas, et al. "A Distributed User-Centered Approach For Control inAmbient Robotic." 8th European
Congress on Embedded Real Time Software and Systems (ERTS 2016). 2016.
100
44. 45
• Addressing complex problems by design
• Simple set of rules can lead to complex behaviours
• Self-organization enables systems to change its
internal organisation to deal with dynamics
without external control
• Studying how the system organizes enables to
extract knowledge
45. 46
Simplicity through locality
Complexity by emergence
KEEP IT SMART, KEEP IT SIMPLE!
Dr. Nicolas Verstaevel, Associate Research
Fellow
SMART Infrastructure Facility,
University of Wollongong, Australia