Increasing ARPU Through Personalization

1
Agent Oriented Design
Agents, Mobility, Ubiquity Virtuality
COMP 30240 Multi-Agent Systems
Lectures Gregory OHare School of Computer
Science Informatics, University College Dublin
(UCD)
Gregory OHare Department of Computer
Science, University College Dublin
2
A Modal Logic Classification
Notation A B Means that B is a strict
superset of A that is every theorem of A is a
theorem of B but not vice versa A B Means
that A and B contain exactly the same theorems
KT4 KDT4
3
Common Pitfalls of Agent oriented Design

• Over selling of the agent metaphor and a failure
  to fully appreciate where agents can usefully be
  deployed in the design
• Overly dogmatic and evangelical in the adoption
  of AOD.
• Unsure as to why agents ought to be used
• A desire to build generic solutions for what is
  in essence a one off problem
• The believe agents are a panacea or silver
  bullet
• One forgets that agent systems are indeed but
  one class of software and demand all the same
  methodological approaches. This is compounded by
  the immature nature of AOD methodologies.

4
Common Pitfalls of Agent oriented Design

• Agent designs are inherently multi-threaded and
  as such must be respectful of the dangers of
  making the same mistakes that dog multi-threaded
  software
• Designers often fail to exploit concurrency
• An unwillingness to exploit off the shelf
  architectures and software components
• The designer can see agents everywhere
• The designer can see too few agents
• Agents interact too freely and in an
  undisciplined manner

5
The Logical Omniscience Problem

• The logical omniscience problem occurs as a
  consequence of
• Knowing all valid formulae and all
  knowledge/belief deriving from such under closed
  logical consequence.
• Refers to the fact that such logics suffer from
  the fact that according to this model that if
  agent X believes/knows a set of axioms to be true
  then they must therefore believe that all the
  logical consequences of this set of axioms to be
  true. Thus the associated problems of
  computational tractability.
• Given a implies b
• Believes(w,a)
• Then Believes(w, b).

6
The Logical Omniscience Problem

• Two well recognized approaches to representing
  Models of Belief are those of-
• (i) Possible Worlds Model and
• (ii) Sentential Semantics.
• Several approaches have been taken to try and
  circumvent the problem of Logical Omniscience
• These can be characterised as -
• Syntactic Approach - Moore and Henderix (1979),
  Konolige (1986)
• Semantic Approach- Levesque (1984)

7
Possible Worlds Model

• Differentiate between the
• Necessarily true operator (P) is true at wi in
  the frame ltw0,w1... Rgt
• iff P is true in all worlds wj in this frame
  such that wj is accessible from wi (written
  R(wi,wj))
• Possibly true operator (P) is true at wi in the
  frame ltw0,w1... Rgt
• iff P is true in wi or a world(s) wj in this
  frame such that wj is accessible from wi
  (written R(wi,wj))

8
Possible Worlds Model

• Within the possible worlds model there is not
  merely one world but multiple worlds. Each world
  represents a different conceptualisation of the
  same base symbols and sentences.
• Associated with each world w1..wn there is an
  Interpretation I1..In.
• Within this model a wff is either true or false
  wrt a given world.
• A wff has a value true with respect to a world wi
  when it evaluates to be true using the
  interpretation associated with that world.
• An accessibility relation R exists which
• R(a, wi, wj) is satisfied when a world wj is
  accessible from world wi for agent a.
• Differentiate between the
• Necessarily true operator possibly
• (P) is true at wi in the frame ltw0,w1... Rgt
• iff P is true in all worlds wj in this frame such
  that wj is accessible from wi (written R(wi,wj))
• Possibly true operator (P) is true at wi in the
  frame ltw0,w1... Rgt
• iff P is true in wi or a world(s) wj in this
  frame such that wj is accessible from wi (written
  R(wi,wj))

9
Agents and Wireless Sensor Networks
Agents, Mobility, Ubiquity Virtuality
COMP 30240 Multi-Agent Systems
Lectures Gregory OHare School of Computer
Science Informatics, University College Dublin
(UCD)
Gregory OHare Department of Computer
Science, University College Dublin
10
Agilla

• Agilla is a middleware that provides a
  mobile-agent paradigm for programming and using
  wireless sensor networks (WSNs) it was developed
  at the University of Washington St Louis.
• Agilla applications consist of mobile agents that
  can proactively migrate their code and state
  across the network.
• Agilla runs on top of TinyOS and allows multiple
  agents to execute on each node. The number of
  agents is variable and is determined primarily by
  the amount of memory available. Each agent is
  autonomous, but shares middleware resources with
  other agents in the system.

11
Agilla Middleware Architecture
12
Agilla

• Agilla provides two fundamental resources on each
  node
• a neighbour list. The neighbor list contains the
  addresses of neighboring nodes. This is necessary
  for agents to decide where they want to move or
  clone to next.
• a tuple space The tuple space provides an elegant
  decoupled-style of communication between agents.
  It is a shared memory architecture that is
  addressed by field-matching rather than memory
  addresses. A tuple is a sequence of typed data
  objects that is inserted into the tuple space.
  The tuple is remains in the tuple space even if
  the agent that inserted it dies or moves away.
  Later, another agent may retrieve the tuple by
  issuing a query for a tuple with the same
  sequence of fields. Note that tuple spaces
  decouples the sending agent from the receiving
  agent they do not have to be co-located, or even
  aware of each other's existence, for them to
  communicate.

13
Agilla

• Agilla is the first mobile agent middleware for
  WSNs that is implemented entirely in TinyOS.
• It has been tested on Mica2 and MicaZ motes. It
  abstracts away complexities associated with
  developing WSN applications, and provides
  mechanisms that overcome the challenges
  associated with limited resources and unreliable
  network communication.
• It demonstrates the feasibility of using mobile
  agents within WSNs and, furthermore, it takes the
  first steps at identifying a minimal set of
  primitives that should be provided for
  facilitating highly flexible WSN applications.

14
Motivation for Mobile Agents within WSNs

• Mobile agents offer more flexibility by allowing
  applications to control the way they spread. They
  can position themselves in the optimal locations
  for performing application-specific tasks.
• They can save energy by bringing computation to
  the data rather than requiring that the data be
  sent over unreliable wireless links.
• They can increase the utility of a WSN by
  constraining themselves to the specific locals
  that are relevant to their application's
  requirements (in contrast to spreading throughout
  an entire network), and sharing the resources of
  a single node, i.e., multiple mobile agents can
  reside on each WSN node.
• Other systems like Deluge and Maté allow
  in-network reprogramming. Agilla, however, goes
  one step further by allowing programs to control
  where they go and to maintain both their code and
  state across migrations.

15
Motivation for Mobile Agents within WSNs

• Since new agents can be injected into a
  pre-existing network, the network can be
  re-tasked. Since each agent executes autonomously
  and multiple agents can simultaneously run on a
  node, multiple applications can co-exist.
• Since mobile agents can move and clone, they can
  quickly morph an application's installation base
  to handle unexpected changes in an environment.
  There are many other inherent advantages of using
  mobile agents, especially in a wireless sensor
  network.

16
Challenges for Agents and WSNs

• However, there are also many challenges
• the foremost being the lack of computational
  resources description of the resource constrained
  nature of a typical mote (Mica Mica2) and
  unreliable network connectivity (lossy and
  unpredictable).
• Issues of serialization should (encoded in an
  appropriate form for transmission),
• Issue of security and protecting against
  malicious mobile code

17
Comparison of AFME Agilla

• Key Comparisons
• AFME supports BDI style strong agents with an
  explicit mental state unlike Agilla agents that
  are of a much weaker genre.
• AFME is not dependent upon TinyOS. Agilla is and
  therefore can only operate on restricted
  platforms
• AFME is implemented in Java and is thus extremely
  portable though it would be argued that JVM
  requirements places an unnecessary expectation on
  the computational power of the host device
• AFME only supports weak migration because of Java
  and the prohibition of access to system
  variables