Download 11-mobihoc

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Convolutional neural network wikipedia , lookup

Ecological interface design wikipedia , lookup

Embodied cognitive science wikipedia , lookup

Catastrophic interference wikipedia , lookup

Transcript
AI Technologies for
Tactical Edge Networks
Karen Zita Haigh
Raytheon BBN Technologies
May 2011
Page 1
What is AI?
Economics
Mathematics
Psychology
Artificial
Intelligence
Natural
Language
Processing
Speech
Recognition
Machine Vision
Control Theory
Robotics
The Odd Paradox
Practical AI successes … were soon assimilated into whatever
application domain they were found to be useful in, and became silent
partners …, which left AI researchers to deal only with the failures.”
[McCorduck, 2004]
Karen Zita Haigh
Page 2
Joe Mitola’s OOPDAL Loop
Joseph Mitola III, Cognitive Radio: An Integrated Agent Architecture for Software
Defined Radio, Phd Thesis, Royal Institute of Technology (KTH), 2000
Karen Zita Haigh
Page 3
Joe Mitola’s OOPDAL Loop (2)
Orient
Assess situation
Infer Intent
Impact Analysis
Observe
Plan
Learn
Collect
Validate
Update
Models
Act
Select Goals
Generate Plans
Schedule
Decide
Select Plan
Allocate Resources
Implement
Karen Zita Haigh
Page 4
Roles for AI in Networking
• Cyber Security
• Network Configuration
(which modules to use)
• Network Control (which
parameter settings to
use)
• Policy Management
• Traffic Analysis
• Performance Analysis
• Sensor fusion /
situation assessment
• Planning
• Coordination
• Optimization
• Constraint reasoning
• Learning (Modelling)
– Complex Domain
– Dynamic Domain
 Unpredictable by
Experts
AI enables real-time, context-aware adaptivity
Page 5
MANET Characteristics
What AI is good at
•
•
•
•
Dynamic
Diverse
Massive Scale
Complex Parameter
Interactions
• Partially-observable
feedback
• Complex Access Policies
• Multi-objective
performance requirements
Main challenges for AI
• Ambiguous feedback
• High-latency feedback
• Resource Constrained
• Heterogeneous
Intercommunication
Cross-Layer Optimization on Steroids
Karen Zita Haigh
Page 6
Knowledge Engineering
• Captures knowledge so that a computer system
can solve complex problems, e.g.
– models of physics and signal propagation, constraints
on the system, analysis of interactions, and rules of
thumb (e.g., about how to configure the system).
• A formal ontology may help a cognitive system
reason about how and when capabilities are
interchangeable
• Knowledge bases can help optimize the network
– e.g. By biasing a learning algorithm
– e.g. By constraining a planner
Karen Zita Haigh
Page 7
Planning and Scheduling
• Organizes tasks to meet performance objectives
under resource constraints
– Multi-agent planning, dynamic programming, constraint
satisfaction, and distributed or combinatorial
optimization algorithms
• Planning and scheduling techniques in networks
can decide what content to move, where, when,
and how
–
–
–
–
–
Prefetch / prepush data
Power-aware computing
Node activity and task scheduling
Network management
Server placement; when to handle queries
Karen Zita Haigh
Page 8
Multi-Agent Systems
• Traditional MAS approaches fail in MANET
because they assume that communications are (a)
infinite and (b) always available
• Biologically-inspired approaches have done better.
• Demonstrated Applications:
– Routing: AntHocNet uses both proactive and reactive
schemes to update the routing tables, and outperforms
AODV.
– Network connectivity
– Dynamic load balancing
– Service placement
Karen Zita Haigh
Page 9
Machine Learning
• ML improves the performance of a system by
observing the environment and updating models
– the learner must generalize so that the learned model is
useful for new (previously unseen) situations.
– Artificial neural networks, support vector machines,
clustering, explanation-based learning, induction,
reinforcement learning, genetic algorithms, nearest
neighbour methods, and case-based learning.
• Demonstrated Applications
–
–
–
–
Routing
Energy management
Node mobility
Parameter interaction
Karen Zita Haigh
Page 10
Concrete Example: ML in ADROIT
• Adaptive Dynamic Radio Open-source Intelligent
Team (ADROIT)
• Create cognitive radio teams that
– Recognize that the situation has changed
– Anticipates changes in networking needs
– Adapts the network, in real-time, for improved
performance
• Real-time composability of the stack
• Real-time control of parameters
• On one node and across the network
Page 11
ADROIT’s Experimental Testbed
Maximize %
of shared
map of the
environment
Page 12
Experimental Results
Training Run:
• In first run nodes learn
about environment
• Train neural nets with
(Conditions,Strategy)Performance
tuples
– Every 5s, measure and
record progress, conditions,
& strategy
– Observations are local, so
each node learns different
model!
Real-time learning run:
• In second run, nodes adapt
behaviour to perform
better.
• Adapt each minute by
changing strategy
according to current
conditions
Real-time cognitive control of a
real-world wireless network
Page 13
Observations from Learning
System performed better with learning
• Selected configurations explainable but not
predictable
– Farthest-refraining was usually better
• congestion, not loss dominated
– Unicast/Multicast was far more complex
• close: unicast wins (high data rates)
• medium: multicast wins (sharing gain)
• far: unicast wins (reliability)
14
Page 14
Biggest remaining challenges
• Social engineering
– the human-to-human interaction of the AI
community differs dramatically from that of the
networking community
• Software architecture
– Network architectures are traditionally tightly
coupled; we need to provide hooks
Module 2
Module 1
Module 1
May 2011
Karen Zita Haigh
Broker
Module 2
Page 15
SOFTWARE ARCHITECTURE
May 2011
Karen Zita Haigh
Page 16
A Need for Restructuring
• SDR gives opportunity to create
highly-adaptable systems, BUT
– They usually require network experts to
exploit the capabilities!
– They usually rely on module APIs that are
carefully designed to expose each
parameter separately.
Module 2
Module 1
• This approach is not maintainable
– e.g. as protocols are redesigned or new
parameters are exposed.
• This approach is not amenable to
real-time cognitive control
– Hard to upgrade
– Conflicts between module & AI
Page 17
A Need for Restructuring
• We need one consistent, generic, interface
for all modules to expose their parameters
and dependencies.
Module 2
Module 1
Page 18
A Generic Network Architecture
Network Module
Network Stack
Network Module
Broker
Registering
Modules
Re/Setting
Modules
Observing
Params
- Assigns
handles
- Provides
directory
services
- Sets up event
monitors
- Pass through
get/set
Applications / QoS
Registering
Modules &
Parameters
Re/Setting
Modules
Observing
Params
Cognitive Control
Network Management
Command Line
Interface
exposeParameter( parameter_name, parameter_properties )
setValue( parameter_handle, parameter_value )
getValue( parameter_handle )
Page 19
Benefits of a Generic Architecture
• It supports network architecture design &
maintenance
– Solves the nхm problem (upgrades or
replacements of network modules)
• It doesn’t restrict the form of cognition
– Open to just about any form of cognition you
can imagine
– Supports multiple forms of cognition on each
node
– Supports different forms across nodes
• It doesn’t mandate cognition
20
Page 20
SOCIAL ENGINEERING
May 2011
Karen Zita Haigh
Page 21
Cultural Issues: But why?
• Benefits and scope of
cross-layer design:
– More than 2 layers!
– More than 2-3
parameters per layer
 Drill-down walkthroughs
highlighted benefits to
networking folks;
explained restrictions to
AI folks
 Simulation results for
specific scenarios
demonstrated the power
• Traditional network
design includes
adaptation
– But this works against
cognition: it is hard to
manage global scope
– AI people want to
control everything
– But network module
may be better at doing
something focussed
 Design must include
constraining how a
protocol adapts
Page 22
Cultural Issues: But how?
• Reliance on centralized
Broker:
– Networking folks don’t
like the single bottleneck
 Design must have failsafe default operation
• Asynchrony and
Threading:
– AI people tend to like
blocking calls.
• e.g. to ensure that
everything is consistent
– Networking folks outright
rejected it.
 Design must include
reporting and alerting
Page 23
Cultural Issues: But it’ll break!?!
• Relinquishing control
outside the stack:
– Outside controller
making decisions scares
networking folks
– AI folks say “give me
everything & I’ll solve
your problem”
 Architecture includes
“failsafe” mechanisms to
limit both sides
• Heterogeneous and
non-interoperable
nodes
– Networks usually have
homogeneous
configurations to
maintain
communications
– AI likes heterogeneity
because of the benefit
• But always assumes safe
communications!
 “Orderwire” bootstrap
channel as backup
Page 24
Cultural Issues: New horizons?
• Capability Boundaries
– Traditional Networking has very clear boundary
between “network” and “application”
– Generic architecture blurs that boundary
• AI folks like the benefit
• Networking folks have concerns about complexity
Removing this conceptual restriction will
result in interesting and significant new
ideas.
Page 25
Conclusion
• AI techniques are ready to be
challenged with this complex real-world
domain, just as Networking
requirements are reaching the limits of
what can be done without AI.
• To demonstrate the power of cognitive
networking, both AI folks & Networking
folks need to recognize and adapt
Page 26