Applying Artificial Immune Systems to RealTime Embedded Systems

1
Applying Artificial Immune Systems to Real-Time
Embedded Systems

• Nicholas Lay, Iain Bate
• Real-Time Systems Group Department of Computer
  Science University of York, UK
• nlay,ijb_at_cs.york.ac.uk

2
Overview

• Background - real-time and embedded systems
• Using AIS techniques to enhance real-time
  embedded systems development
• Applying AIS to a real-time systems problem

3
Real-time systems

• A real-time system is one where the correctness
  of an operation depends on the time a result is
  produced as well as the outcome of that operation
• Often associated with safety-critical or
  high-integrity applications vital that
  real-time requirements are met
• The RTS domain has been widely researched and is
  well understood
• A variety of specialist techniques have been
  developed which guarantee the correctness of
  real-time systems

4
Embedded systems

• An embedded system is a computer system which is
  encapsulated inside another device
• Originally employed to replace custom control
  logic
• Often perform a small set of specific functions
  within that device
• Complexity often less than traditional computer
  systems, but is increasing

5
Real-time embedded systems

• A variety of systems exhibit both real-time and
  embedded properties
• A range of application domains
• Aerospace, automotive safety critical
• Consumer electronics (CE) non-safety critical
• We are particularly interested in the use of
  real-time embedded systems in the CE industry

6
CE development

• The CE marketplace is highly competitive
• The development of CE devices is influenced by
  market considerations in addition to the system
  requirements
• Development costs must be as low as possible
• Often, system components are reused or modified
  for use in similar systems
• Development time must be short
• System must reach the market on time
• Final system must be reliable
• A product which is unreliable wont sell

7
Real-time development

• Real-time development traditionally relies on
  static analysis techniques to guarantee the
  correctness of a system
• These methods are effective, but are
• Inflexible analysis results are only applicable
  to the exact system they were generated with
• Time-consuming every change requires the
  analysis to be completely re-done
• Expensive largely as a result of the above!

8
How can AIS help?

• Traditional real-time development is not feasible
  for the CE industry
• Need a technique to reduce real-time anomalies in
  CE systems
• AIS has been applied successfully to anomaly
  detection problems
• The adaptability of AIS-based systems gives the
  potential for a complete turnkey solution

9
AIS in embedded systems

• The application of AIS techniques to embedded
  systems raises a number of issues
• In particular, how AIS-based systems can be
  engineered to run effectively with constrained
  resources
• Traditional AIS-based systems are based around
  adaptive immune concepts
• Effective, but generally not suited to
  constrained resource environments
• New techniques, based on innate immunity, may
  provide a viable solution

10
The Danger Model

• Classic immunology theory is based on the concept
  of self-nonself discrimination
• The application of this has been problematic in
  artificial systems
• The Danger Model is an alternative theory for the
  functioning of the immune system
• Immune system components respond to danger
  signals produced when cells die unexpectedly
• The concepts of the Danger Model can be readily
  applied to a variety of anomaly detection problems

11
The Dendritic Cell Algorithm (DCA)

• A relatively new algorithm developed by
  Greensmith et al, University of Nottingham
• Based on the Danger Model and principles of
  innate immunity
• Exact operation derived from observation of
  immune system components in vivo
• A population of DCs collect input signals from a
  set of system components over a period of time,
  and output a danger level for that set of
  components

12
Applying AIS to a real-time problem

• The functionality of most computer systems is
  divided into a number of specific tasks
• Each task has a set of properties which affect
  its execution within the system
• In a real-time system, each task has a deadline
  by which its execution must be complete
• Failure to meet these deadlines causes errors in
  the system
• We are investigating the use of the DCA to detect
  missed deadlines in a real-time system. The
  results can be compared with those derived from
  traditional static analysis techniques

13
Applying AIS to task scheduling

• In order to test the DCA, we use a set of tasks
  which are arranged such that the execution of one
  task can cause an overrun in another

Increasing pririty
14
Applying AIS to task scheduling

• We assign various properties of each task to the
  input signals of the DCA
• Each DC in the system is associated with a subset
  of the tasks present. Output of multiple DCs can
  be combined to deduce the state of the whole
  system

15
Applying AIS to task scheduling
16
Evolving the DCA parameters

• The operation of the DCA relies heavily on
  parameters which are control how the output is
  derived from the chosen input signals
• These parameters are currently based on in vivo
  observations of living dendritic cells
• The DCA in its current form assumes that these
  parameters are valid for all problems
• This may not be the case

17
Evolving the DCA parameters

• How do these parameters affect the operation of
  the DCA?
• Is it feasible to use an evolutionary strategy to
  tune the DCA to a specific problem?
• Could this be achieved automatically, allowing a
  DCA-based solution to be incorporated into a
  system during its development process?
• By randomly mutating these parameters, we can
  establish how they affect the operation of the
  DCA with a view to employing an evolutionary
  technique to optimise them
• Allow us to produce a version of the DCA which is
  correct, responsive and robust

18
Evolving the DCA parameters
19
Conclusions and further work

• The DCA has potential to improve the reliability
  of real-time embedded systems
• There is the possibility of further enhancements
  to the DCA, allowing its parameters to be tuned
  according to the problem
• Such parameters include weights, danger
  thresholds, number of allocated DCs etc
• Examine the effects of constrained resources on
  the operation of the DCA
• Ideal size of DC population?