• Use other tools like NNs to reduce the size of the data set or problem complexity before applying EC technique.
• A good approach for future research is to hybridize different EC algorithms as well as to hybridize with other techniques – more work could be done although there are some existing ones.
  • For pure EC research, it could be for the purpose of better understanding of the EC behaviors.

• Focus more on what industries are interested on, e.g. real-time, fast answer, etc.
• Hybridize with existing software, extending from old system if applicable.
• Sequence alignment –hybridization by applying EC after some initial investigation - understand the problem.
• Domain expert + EC expert =>solution for the problem
• Important to help others to better understand our field, e.g. the biologist community.

• Answer the question - why EC?
• There are existing algorithms – should be clear on when and where to use EC.
• One particular area EC is useful could be for optimization in dynamic environment, e.g., dynamic fitness feedback is not static but involves noisy signal. For this problem, other algorithms may not work as well as EC as found by the BT.
• Build systems to illustrate the concepts of EC and to convince the industry the capability of EC.
• Hybrid system – BT has applied the graph theory to preprocess data and apply EC to further improve the results obtained.
• Current approach is mainly focused on offline optimization – online evolution where adaptation and evolution are performed at the same time are important.

Unification:

• To establish a common framework and use the framework to achieve a problem oriented approach instead of algorithm oriented approach for solving problems.
• Unified view of “simple EA” is not sufficient.
• Principled extensions are required.

Expansion:

• Adaptation and reduce knob twiddling
• Exploiting parallelism
  • Coarsely grained network models or finely grained diffusion models
• Understanding co-evolutionary models
  • Competitive co-evolution or cooperative co- evolution?
• Exploiting morphogenesis:
  • Sophisticated genotype –phenotype mappings
  • Evolve plans for building complex objects rather that the objects themselves
• Understand multi-objective optimization better
  • standard feature of industrial problems
  • need deeper theoretical understanding
• Understand time varying environment better
• Agent oriented problems
  • Decentralize the genetic algorithms
• Need stronger analysis tools
  • Markov models
  • Statistical mechanics
  • Evolutionary game theory
  • Test problems
  • Visualization
• Need better hybrid systems
  • Memetic algorithms
  • EA and ANN and others

• International collaboration
• Many groups too small to have sufficient internal diversity of expertise

Hybrid –vigour

• interacting groups generate more ideas more quickly

Diversity promotion

• avoiding premature convergence in ideas

Super-phenotypic behaviour

Challenging the individual with a variety of environments

• Especially important for our graduate students (postdoc opportunities etc)

Science advances from chaotic behaviour

Generation of paradigm shifts

• international linkages boost our profile to funding agencies, benefiting EA research as a whole

• non-targeted funding
  • doctoral scholarship schemes
  • postdoctoral fellowship schemes
  • university exchange schemes
• 
  
• inter-Govermental schemes
  • Australia –
    • Germany, UK, Eira, USA, China, Japan, Korea, Malaysia, Singapore, …
  • Development Aid Schemes
    • Doctoral scholarship
• 
  
• international funding agencies
  • UN, trans-national corporations
• 
  
• Australian targeted funding for complex/intelligent systems
• Centre of excellence for complex/intelligent systems

• Why a student with PhD is incapable of solving problem? - blind leads blind
• Teach academic solutions - solve industrial problems???
• Teach the fundamental and how to apply it to solve real problems.
• Teach students on what should be learnt, how to solve problems and apply what he/she has learnt.