A Cloud-Based Bayesian Smart Agent Architecture for Internet-of-Things Applications
1. A Cloud-Based Bayesian Smart
Agent Architecture
for Internet-of-Things Applications
Authors: Veselin Pizurica, Piet Vandaele @waylay
Rome, 27/10/2014
2. IoT early years (technology) view
• IoT was about devices, protocols and data flows
• “gateway centric”
• “Liner logic”: left devices, right services…
3. IoT today: business point of view
• You see marketing departments taking over
• Picture more fuzzy, devices and services all over the
place
5. Why NOT intelligence in the cloud?
• Latency
• Failure (in)tolerance (lack of redundancy) – general issue
in internet, adding more blocks system even less stable
• Cost of pushing data in the cloud
– Energy (battery)
– Data storage (data can be of a huge volume)
– SW cost of integration
– Lack of standardization
• Security concerns: Authentication/Authorization
• Privacy concerns
6. Why intelligence in the cloud?
• Device-agnostic and decouples logic from the
presentation layer
• Combination of the sensor data with API “economy”
• Integrating multiple IoT vertical solutions
• Cloud-capacity scales horizontally, while distributed HW
often needs to be swapped when HW resources are no
longer sufficient
• Cloud intelligence also allows easy generation of analytics
regarding the usage of the logic itself. Which rules fired
and why? How often?
• An architectural model arises where logic is built once
together with a REST API
7. A Cloud-Based Smart Agent
Sense
Transmit
Store
Analyze offline
PresentReason
Act
Artificial Intelligence provides us the framework and tools to
go beyond trivial real-time decision and automation use
cases for IoT.
In this presentation, we present a cloud-based smart agent
architecture for real-time decision taking in IoT applications
8. Rational Agent
* Russell S., Norvig P.: Artificial Intelligence A Modern Approach, Third Edition, Pearson (2014)
Rational Agent Architecture *
9. Agent architecture choices
• The choice for a particular type of agent logic is
influenced by the characteristics of the environment in
which an agent needs to operate
• Type of agents (using software language to express the
logic):
– ‘if-then-else’ constructs that are available in any programming
language or rules engine
– flowchart models
– CEP (complex event processing) engines
– Graph models (Markov, Bayesian nets)
10. Why Bayesian Networks in IOT?
• Environments that cannot be completely observed, i.e.
when not all aspects that could impact a choice of action
are observable.
• Unreliable, noisy or incomplete data or when domain
knowledge is incomplete such that probabilistic reasoning
is required
• Use cases where the number of causes for a particular
observation is so large, that it is nearly impossible to
enumerate them explicitly
• Well suited to model expert-knowledge together with
knowledge that is retrieved from accumulated data
• Use cases where there are asynchronous information flows
11. • Belief propagation algorithm was introduced by Judea Pearl, 1982
• Pearl was inspired by the paper of cognitive psychologist Rumelhart on how
children comprehend text
• Generalization of the Kalman’s algorithm
• Became very popular after it was shown that the same computations are in
turbo codes and the same principles in the Viterbi algorithm
• Main idea: inference by local message passing among neighboring nodes
The message can loosely be interpreted as “I (node i ) think that you
(node j) are that much likely to be in a given state”.
Belief propagation
12. Example: Car diagnosis
• Initial evidence: car won't start
• Testable variables (green), “broken, so fix it” variables
(orange)
• Hidden variables (gray) ensure sparse structure, reduce
parameters
16. Cloud Smart
Agent Platform Environment
SW-defined
Sensors
Graph
Modeling
SW-defined
Actuators
Percepts
Actions
Physical Sensors
IoT platforms
Social media
Location
Open Data
Big Data
API economy
REST
API
LOB apps
Proposed architecture
Vertical
Specific
End-user
Interface