Universal laws and architectures: Theory and lessons from

Transcription

1 Universal laws and architectures: Theory and lessons from brains, bugs, nets, grids, planes, docs, fire, bodies, fashion, earthquakes, turbulence, music, buildings, cities, art, running, throwing, Synesthesia, spacecraft, statistical mechanics John Doyle Jean-Lou Chameau Professor Control and Dynamical Systems, EE, & BioE # Ca 1 tech

2 A minimal prologue

3 Diverse applications Aerospace Robotics Biology Fluids Physics Medicine Internet Smartgrid Ecology Neuroscience Tuesday Wednesday Thursday Monday

4 Diverse applications Diverse mathematics Aerospace Robotics Biology Fluids Physics Medicine Internet Smartgrid Ecology Neuroscience Details are not serious ODE/PDE Automata Analysis Optimization Prob&Stats Operator Th Diff Geom Alg Geom Cmp Cplxty

5 Diverse applications (not just engineering) Diverse mathematics (not just applied ) Aerospace Robotics Biology Fluids Physics Medicine Internet Smartgrid Ecology Neuroscience CDS Universal laws and architectures ODE/PDE Automata Analysis Optimization Prob&Stats Operator Th Diff Geom Alg Geom Cmp Cplxty

6 Diverse applications Aerospace Robotics Biology Fluids Physics Medicine Internet Smartgrid Ecology Neuroscience Domain experts have core interface to diverse math and tools CDS

7 Math (not just applied) has core interface to diverse application drivers CDS Diverse mathematics ODE/PDE Automata Analysis Optimization Prob&Stats Operator Th Diff Geom Alg Geom Cmp Cplxty

8 CDS as architecture Architecture= constraints that deconstrain Constraint = CDS core concepts and math Deconstraint = connections between new applications and (also possibly new) math How to explain this? Laws: Universal constraints on achievable robust performance and efficiency Architectures: Universal organizational strategies to flexibly achieve what is possible Case studies: concrete and familiar examples to illustrate universals 8

9 Core protocols From Architecture slides, understanding the OS is difficult and essential But term OS seems to confuse Replace with core protocols? Highly conserved/constrained core (knot) Highly diverse/deconstrained edges What is CDS in this context? 9

10 Slow Apps Horizontal App Transfer OS HW Fast Flexible Horizontal Hardware Transfer Inflexible General Special

11 Tradeoffs Slow Apps OS HW Fast Flexible General Inflexible Special

12 Tradeoffs = laws? Slow Apps OS HW Fast Flexible General Inflexible Special

13 Layered architectures Prefrontal vision Slow Fast HGT Motor DNA repair Mutation Apps OS HW Dig. Lump. Distrib. Flexible General OS HW Dig. Signal Lump. Distrib. DNA replication Transcription Translation Metabolism Digital Lump. Distrib. Fast Lumped Distrib. Distrib. Inflexible Special Sense VOR Reflex

14 Architecture = Constraints that deconstrain Horizontal App Transfer Deconstrained OS Constrained Horizontal Hardware Transfer Deconstrained

15 Universal architectures Horizontal Transfer Hourglass Deconstrained Constrained Horizontal Transfer Deconstrained

16 Deconstrained Constrained Deconstrained Bowtie Horizontal versus vertical is an arbitrary convention.

17 Diverse applications (not just engineering) Diverse mathematics (not just applied ) Aerospace Robotics Biology Fluids Physics Medicine Internet Smartgrid Ecology Neuroscience CDS Universal laws and architectures ODE/PDE Automata Analysis Optimization Prob&Stats Operator Th Diff Geom Alg Geom Cmp Cplxty

18 Pasteur s quadrant? (Donald Stokes) Typically poor choice of colors and font sizes

19 Pasteur s quadrant? (Donald Stokes) High Fundamental Understanding Low Bohr Use inspired basic research (Pasteur) Edison Relative emphasis Low Practical use High

20 Pasteur s quadrant? Redrawn Low Fundamental Understanding High Edison Pasteur Pasteur aimed for both Bohr Relative emphasis High Practical use Low

21 Apps CDS Math Practical use Fundamental Understanding

22 CDS quadrant? Specific Apps Scope Deep, Universal CDS Math Now Practical impact Later

23 Related but somewhat flipped Slow SW Apps OS Fast HW Specific Apps Flexible General Inflexible Special Scope Deep, Universal Now CDS Math Later Practical impact

24 Circa 1975 Specific Scope Apps Modern control theory Deep, Universal Math Now Later Practical impact

25 Circa 1990 Complexity science Specific Scope Deep, Universal Apps Now Robust control theory Math Modern control theory Later Practical impact

26 Nothing Specific Scope Deep, Universal Apps 2014 CDS Complexity science Math Network science Now Later Practical impact Never

27 Nothing Network science Complexity science Specific Scope Humans love this corner! Deep, Universal Now Later Practical impact Never

28 Near future? Network science Complexity science Specific Apps Deep, Universal CMS CDS Math Now Practical impact Later

29 The dream? Network science Complexity science Specific Apps Deep, Universal CDS Math Now Practical impact Later

30 Slow SW Apps Universal laws Fast OS HW Flexible General Inflexible Special Universal architectures Horizontal Transfer Deconstrained Constrained Horizontal Transfer Deconstrained

31 Layered architectures Prefrontal vision Slow Fast HGT Motor DNA repair Mutation Apps OS HW Dig. Lump. Distrib. Flexible General OS HW Dig. Signal Lump. Distrib. DNA replication Transcription Translation Metabolism Digital Lump. Distrib. Fast Lumped Distrib. Distrib. Inflexible Special Sense VOR Reflex

32 Diverse applications Aerospace Robotics Biology Fluids Physics Medicine Internet Smartgrid Ecology Neuroscience Morning core interface to minimal math and tools CDS Universal laws and architectures Undergrad math

33 Afternoon Diverse mathematics Networks, Distributed CDS Cutting edge theory foundations ODE/PDE Automata Analysis Optimization Prob&Stats Operator Th Diff Geom Alg Geom Cmp Cplxty

34 The dream? Network science Complexity science Specific Apps Deep, Universal CDS Math Now Practical impact Later

35 Caveats Many ideas are classic robust control, but much of the organization (the architecture) is fairly new Motivating case studies also new Many rough edges Nobody is working on it quite like this We want to change that