Engineering Complexity Nicolas Demassieux © Nicolas Demassieux
Complexity of Interacting Systems Internet Security Hubble mirror ARIANE 5 first flight  –  Space shuttle © Nicolas Demassieux
Engineering Environment PAST Simple predictable environment Low interactions Low complexity Expensive Control Cheap Energy A priori validation Strict plans  (“design”) FUTURE Complex environments Complex interactions Really High complexity Cheap Control Costly Energy In situ validation Indirect plans (“specification”) © Nicolas Demassieux
Life “know-how” CELL Self Assembly Self Repair Self Recycling DNA Robustness Replication Evolution ORGANISM Energy Supply Intelligence Adaptation ECOSYSTEMS Co-evolution Diversity Long term survival © Nicolas Demassieux
Engineering SUBSYSTEMS Self Assembly :  Boot, Pilot channels ,  Molecular engineering,   Quantum dots, Nanotubes, Shape memory alloys, ... Self Repair :  Disk unfragmentation,   Self check, Fault tolerance, ... Self Recycling :  Garbage collectors,   Self disassembly, biodegradable materials ... DESIGN Robustness :  DIGITAL Replication :  Software   Objects (Inheritance)... Evolution :  Monte Carlo,   Genetic Algorithms, Artificial life... SYSTEMS Energy Supply : Intelligence :  Neural networks, Agent technology, Cellular Automatas, AI... Adaptation :  Self optimizing systems, Software radios, Adaptive techniques TECHNOSPHERE Co-evolution :  the Internet, Radio ecosystems, service pyramid…. Diversity :  Non convergence Long term survival :  Non optimality © Nicolas Demassieux
One Example : Optimisation CONTINUOUS OPTIMISATION Energy landscape Local minima Gradient algorithms, simulated annealing BOOLEAN OPTIMISATION Boolean function : {0,1} n     {0,1} Problem : Synthesis using boolean gates with minimal cost …. a NP hard problem Solutions : Iterative heuristics, genetic programming, ... 0 1 0 1 1 0 1 0 A  B  C © Nicolas Demassieux
One Example : Optimisation (2) © Nicolas Demassieux COMPLEX MULTIPARAMETER SYSTEMS CONTROL Problem : “tweak the knobs of a system” to optimize its behavior in a given complex environment Solutions : genetic programming, optimal rate distorsion, self optimizing systemes, self awareness... DCT Q image reconstruite image source  VLC mémoire d'image + - + + Q -1 DCT -1 VLC -1 mémoire d'image estimation mouvement Q -1 DCT -1 Buffer Buffer COMPLEX MULTIDIMENSIONAL “SHAPES” IDENTIFICATION Problem : Classification, learning Solutions : VQ, Neural network, fractal ...
New Engineering Culture PAST Models, black Boxes “ Hard Specifications” Optimization Engineered control Independent Systems Confidence FUTURE No more complete models Evolving specifications Overprovision Autonomous control Interdependent Systems Humility Engineering Optimized Systems Managing  Technical  Environments © Nicolas Demassieux

Engineering Complexity

  • 1.
    Engineering Complexity NicolasDemassieux © Nicolas Demassieux
  • 2.
    Complexity of InteractingSystems Internet Security Hubble mirror ARIANE 5 first flight – Space shuttle © Nicolas Demassieux
  • 3.
    Engineering Environment PASTSimple predictable environment Low interactions Low complexity Expensive Control Cheap Energy A priori validation Strict plans (“design”) FUTURE Complex environments Complex interactions Really High complexity Cheap Control Costly Energy In situ validation Indirect plans (“specification”) © Nicolas Demassieux
  • 4.
    Life “know-how” CELLSelf Assembly Self Repair Self Recycling DNA Robustness Replication Evolution ORGANISM Energy Supply Intelligence Adaptation ECOSYSTEMS Co-evolution Diversity Long term survival © Nicolas Demassieux
  • 5.
    Engineering SUBSYSTEMS SelfAssembly : Boot, Pilot channels , Molecular engineering, Quantum dots, Nanotubes, Shape memory alloys, ... Self Repair : Disk unfragmentation, Self check, Fault tolerance, ... Self Recycling : Garbage collectors, Self disassembly, biodegradable materials ... DESIGN Robustness : DIGITAL Replication : Software Objects (Inheritance)... Evolution : Monte Carlo, Genetic Algorithms, Artificial life... SYSTEMS Energy Supply : Intelligence : Neural networks, Agent technology, Cellular Automatas, AI... Adaptation : Self optimizing systems, Software radios, Adaptive techniques TECHNOSPHERE Co-evolution : the Internet, Radio ecosystems, service pyramid…. Diversity : Non convergence Long term survival : Non optimality © Nicolas Demassieux
  • 6.
    One Example :Optimisation CONTINUOUS OPTIMISATION Energy landscape Local minima Gradient algorithms, simulated annealing BOOLEAN OPTIMISATION Boolean function : {0,1} n  {0,1} Problem : Synthesis using boolean gates with minimal cost …. a NP hard problem Solutions : Iterative heuristics, genetic programming, ... 0 1 0 1 1 0 1 0 A  B  C © Nicolas Demassieux
  • 7.
    One Example :Optimisation (2) © Nicolas Demassieux COMPLEX MULTIPARAMETER SYSTEMS CONTROL Problem : “tweak the knobs of a system” to optimize its behavior in a given complex environment Solutions : genetic programming, optimal rate distorsion, self optimizing systemes, self awareness... DCT Q image reconstruite image source VLC mémoire d'image + - + + Q -1 DCT -1 VLC -1 mémoire d'image estimation mouvement Q -1 DCT -1 Buffer Buffer COMPLEX MULTIDIMENSIONAL “SHAPES” IDENTIFICATION Problem : Classification, learning Solutions : VQ, Neural network, fractal ...
  • 8.
    New Engineering CulturePAST Models, black Boxes “ Hard Specifications” Optimization Engineered control Independent Systems Confidence FUTURE No more complete models Evolving specifications Overprovision Autonomous control Interdependent Systems Humility Engineering Optimized Systems Managing Technical Environments © Nicolas Demassieux