John Doyle John G Braun Professor Control and Dynamical...
Transcript of John Doyle John G Braun Professor Control and Dynamical...
John DoyleJohn G Braun Professor
Control and Dynamical Systems, EE, BioECaltech
Doyle and Csete, Proc Nat Acad Sci USA, online JULY 25 2011
This paper aims to bridge progress in neuroscience involving
sophisticated quantitative analysis of behavior, including the use
of robust control, with other relevant conceptual and theoretical
frameworks from systems engineering, systems biology, and
mathematics.
Doyle and Csete, Proc Nat Acad Sci USA, online JULY 25 2011
This paper aims to bridge progress in neuroscience involving
sophisticated quantitative analysis of behavior, including the use
of robust control, with other relevant conceptual and theoretical
frameworks from systems engineering, systems biology, and
mathematics.
Familiar and accessible case studies are used to illustrate
concepts of robustness, organization, and architecture
(modularity and protocols) that are central to understanding
complex networks.
These essential organizational features are hidden during normal
function of a system, but fundamental for understanding the
nature, design, and function of complex biologic and technologic
systems.
Doyle and Csete, Proc Nat Acad Sci USA, online JULY 25 2011
Starting points
large,thin,nonconvex
lobster somatogastric ganglia
Helen S. Mayberg, MDProfessor, Psychiatry and Neurology
Dorothy C. Fuqua Chair in Psychiatric Neuroimaging and Therapeutics
Emory University School of Medicine
Last week
moF11pACC24
mF9/10
PCCMCC
PF9/46 Par40PM6
sACC25
hth bstema-ins
amg mb-vta
hc
na-vst thal
SalienceMotivation
Moodstate
EmotionRegulation
Cognitionattention-appraisal-action
Interoceptiondrive-autonomic-circadianMayberg, Br Med Bul 65:193-207, 2003
Mayberg, J Clin Invest 119:717, 2009
Putative “Depression” Network ~ 2001defined using functional imaging
PF
P
Medications
Cognitive
Behavior
Therapy
Where might
DBS fit into
this schema?
Mayberg, J Neuropsych Clin NS, 1997
cg25
Tracts thru SeedCg25WM Target Fiber Assignment by
Continuous Tracking
along adjacent pixels
Fornix
Probablistic Tractography
DTI: Blind ACC ParcellationSVD 2 clusters: sACC Pacc (n=18)
Cingulate
ROI
Johansen-Berg et al
Cerebral Cortex 2008
a-hc
Overlap
Unique inferior ROI
Unique superior ROI
na-hth
Fx
Diffusion Tensor
Imaging
Rethinking Critical PathwaysMapping Fibers of Passage thru SCC25
cg
mF
cg
25
both
4dmF9MCC
nA
25
3
oF11
nA
25
nA
Differences between
Adjacent contacts
mACCmF9
oF11
mF10
Define tracts affected by stimulation
4
3
2
1
nAc
Am/hcSt-Thdm-Th
HcAm
Midline Cortex
limbicIns
oF11 mF10
pACC MCC PCC
vtaCg25
Cingulum
Hth DR LC pag
nac-vs
brainstem
mF9
Model
4321
Post-op CT/MRI merge
Rethinking Critical PathwaysMapping Fibers of Passage thru SCC25
Sensors
Effecto
rs
Reacti
ve
Ad
ap
tive
Co
nte
xtu
al
So
ma
Value/utility
Long Term Memory
Allostatic control
Needs
Internal states
(valence/arousal)
World
Working memory
Sequence/Interval
memory
Episodic memory
PerceptionAction
shaping/
selection
BehaviorsSensation
Plans
Autobiographical memory / Meta learning
Event
memory
Goals
The
Distributed
Adaptive
Control
Architecture
Verschure & Voegtlin (1998) Neural Netw
Verschure et al (1992) Rob. Aut. Sys.
Verschure & Althaus (2003) Cogn. Sci. (27) 561
Verschure et al (2003) Nature (425) 620
Duff et al (2010) Neurocomputing
Mathews et al (2009;2010) IROS09;ICRA10
Verschure & Coolen (1991) Network
Duff et al (2011) Br.Res.Bull.
Sanches et al (2010) Adv Compl Sys
Eng et al (2003;2005) ICRA; IEEE Tr Sys, Man,
Paul
Verschure
Sensors
Effecto
rs
Reactive
Adaptive
Contextual
Soma
World
Layered
Distributed
Adaptive
Control
Architecture
Sensors
Effecto
rs
Reacti
ve
Ad
ap
tiv
eC
on
textu
al
So
ma
Value/utility
Long Term Memory
Allostatic control
Needs
Internal states
(valence/arousal)
World
Working memory
Sequence/Interval
memory
Episodic memory
PerceptionAction
shaping/
selection
BehaviorsSensation
Plans
Autobiographical memory / Meta learning
Self:
Interoception
World:
Exteroception
Action
Event
memory
Goals
The Distributed
Adaptive
Control
Architecture
Weak 1st
person
perspective
1PP
minimally
phenomenal self
(MPS)
Sensors
Effecto
rs
Reacti
ve
Ad
ap
tive
Co
nte
xtu
al
So
ma
Value/utility
Long Term Memory
Allostatic control
Needs
Internal states
(valence/arousal)
World
Working memory
Sequence/Interval
memory
Episodic memory
PerceptionAction
shaping/
selection
BehaviorsSensation
Plans
Autobiographical memory / Meta learning
Self:
Interoception
World:
Exteroception
Action
Event
memory
Goals
Levels of Conciousness
following Metzinger
(2003)
The Distributed
Adaptive
Control
Architecture
“Laws” and architectures
• Theory of hard limits, constraints,… (“laws” )
• Theory of “architecture”?
– From platforms to
– “systems of systems” to
– Architecture
• Case studies: Internet, turbulence, smartgrid, cell biology, wildfire ecology, earthquakes, stat mech, brain architecture, UAVs, …
IEEE TRANS ON SYSTEMS, MAN, AND CYBERNETICS,
JULY 2010, Alderson and Doyle
Robust Fragile
Human complexity
Metabolism
Regeneration & repair
Healing wound /infect
Obesity, diabetes
Cancer
AutoImmune/Inflame
Robust Fragile
Mechanism?
Metabolism
Regeneration & repair
Healing wound /infect
Fat accumulation
Insulin resistance
Proliferation
Inflammation
Obesity, diabetes
Cancer
AutoImmune/Inflame
Fat accumulation
Insulin resistance
Proliferation
Inflammation
Robust Fragile
What’s the difference?
Metabolism
Regeneration & repair
Healing wound /infect
Obesity, diabetes
Cancer
AutoImmune/Inflame
Accident or necessity?
Fat accumulation
Insulin resistance
Proliferation
Inflammation
Fluctuating
energy
Static
energy
Robust Fragile
What’s the difference?
Metabolism
Regeneration & repair
Healing wound /infect
Obesity, diabetes
Cancer
AutoImmune/Inflame
Fat accumulation
Insulin resistance
Proliferation
Inflammation
Controlled
Dynamic
Uncontrolled
Chronic
Low mean
High variabilityHigh mean
Low variability
Robust Fragile
Restoring robustness
Controlled
Dynamic
Uncontrolled
Chronic
Low mean
High variabilityHigh mean
Low variability
Robust
Metabolism
Regeneration & repair
Healing wound /infect
Fat accumulation
Insulin resistance
Proliferation
Inflammation
Fluctuating
energy
Controlled
Dynamic
Low mean
High variability
Mechanism?
Brain
Heart
Muscle
Liver
GI
Glu
Triglyc
Fat
Glyc
Glyc
FFA
Glycerol
Oxy
Lac/ph
Out
fast slow
high
low
pri
ori
tydynamics
Control?
• Energy• Inflammation• Coagulation
Evolved for large energy variation and
moderate trauma
source receiver
signaling
gene expression
metabolism
lineage
Biological
pathways
Brain
Heart
Muscle
Liver
GI
Glu
Triglyc
Fat
Glyc
Glyc
FFA
Glycerol
Oxy
Lac/ph
Out
fast slow
high
low
pri
ori
tydynamics
source receiverBiological
pathways
source receiver
control
energy
materials
signaling
gene expression
metabolism
lineage
More
complex
feedback
food intake
Glucose
Oxygen
Amino acids
Fatty acids
Organs
Tissues
Cells
Molecules
Universal metabolic system
Blood
Peter Sterling and Allostasis
food intake
Organs
Tissues
Cells
Molecules
Robust
Highly
variable
supply
Highly
variable
demand
Evolvableevolving
diet
evolving
function
Efficient
Organs
Tissues
Cells
Molecules
Highly
variable
demand
food intake
Highly
variable
supply
evolving
dietevolving
function
Glucose
Oxygen
Blood
Conserved
core
building
blocks
VTAPrefrontalcortexAccumbens
dopamine
Universal reward systems
sportsmusicdancecrafts arttoolmaking sexfood Dopamine,
Ghrelin,
Leptin,…
Universal reward systems
sportsmusicdancecrafts arttoolmaking sexfood
PFC
CG
OFC
NAcc
Amyg
STR
TH PIT
HIP
SN
VTA dopamine
Reward
Drive
Control
Memory
Robust and evolvable
VTA dopamine
Universal reward systemssportsmusicdancecrafts arttoolmaking sex
food
Glucose
Oxygen
Organs
Tissues
Cells
Molecules
Universal metabolic system
Blood
Constraints
that
deconstrain
Reward Drive Control Memory
dopamine
Blood
Glucose
Oxygen
Constraints
Modularity 2.0
sportsmusicdancecrafts arttoolmaking sex
food
Organs
Tissues
Cells
Molecules
Reward Drive Control Memory
that
deconstrain
Modularity 2.0
Layered architectures
“bow-tie”
diverse diverse
dopamine
workfamily communitynature
Universal reward/metabolic systems
Robust and adaptive, yet …
foodsextoolmakingsportsmusicdancecrafts art
Blood Organs
Tissues
Cells
Molecules
Reward Drive Control Memory
workfamily communitynature
foodsextoolmakingsportsmusicdancecrafts art
Organs
Tissues
Cells
Molecules
Reward Drive Control Memory
Modularity 1.0
“Weak linkage”
dopamine
Most important “modules”
Blood
Not weakly connected to others,
but highly connected
Modularity 2.0
dopamine
workfamily communitynature
Universal reward/metabolic systems
Robust and adaptive, yet …
foodsextoolmakingsportsmusicdancecrafts art
Blood Organs
Tissues
Cells
Molecules
Reward Drive Control Memory
sexfoodtoolmakingsportsmusicdancecrafts art
workfamily communitynature
cocaineamphetamine
dopamine
Blood Organs
Tissues
Cells
Molecules
Reward Drive Control Memory
workfamily communitynature
sextoolmakingsportsmusicdancecrafts art
dopamine
Vicarious
money
saltsugar/fatnicotinealcohol
industrialagriculture
market/consumerculture
OrgansTissuesCellsMolecules
Reward Drive Control Memory
Vicarious
money
saltsugar/fatnicotinealcohol
high sodium
obesity
overwork
smoking
alcoholism
drug abuse
hyper-tension
athero-sclerosis
diabetes
inflammation
immunesuppression
coronary,cerebro-vascular,reno-vascular
cancer
cirrhosis
accidents/homicide/suicide
From Sterling
VTA
Prefrontalcortex
Accumbens
dopamine
Universal reward systemssportsmusicdancecrafts arttoolmaking sexfood
Glucose
Oxygen
Organs
Tissues
Cells
Molecules
Universal metabolic system
Blood
food
VTA dopamine
Glucose
Oxygen
Vicarious
money
saltsugar/fatnicotinealcohol
high sodium
obesity
overwork
smoking
alcoholism
drug abuse
hyper-tension
athero-sclerosis
diabetes
inflammation
immunesuppression
coronary,cerebro-vascular,reno-vascular
cancer
cirrhosis
accidents/homicide/suicide
Yet Fragile
Robust Yet Fragile
Human complexity
Metabolism
Regeneration & repair
Microbe symbionts
Immune/inflammation
Neuro-endocrine
Complex societies
Advanced technologies
Risk “management”
Obesity, diabetes
Cancer
Parasites, infection
AutoImmune/Inflame
Addiction, psychosis…
Epidemics, war…
Disasters, global &!%$#
Obfuscate, amplify,…
Accident or necessity?
Robust Fragile Metabolism
Regeneration & repair
Healing wound /infect
Obesity, diabetes
Cancer
AutoImmune/Inflame Fat accumulation
Insulin resistance
Proliferation
Inflammation
• Fragility Hijacking, side effects, unintended…
• Of mechanisms evolved for robustness
• Complexity control, robust/fragile tradeoffs
• Math: New robust/fragile “conservation laws”
Accident or necessity?
Both
Chandra, Buzi, and Doyle
Yet Fragile
Fix?
Obesity, diabetes
Cancer
Parasites, infection
AutoImmune/Inflame
Addiction, psychosis…
Epidemics, war…
Catastrophes
Obfuscate, amplify,…
“Architecture”
Don’t worry ...
• “There’s an app for that.”
• “The rapture is near.”
• “There’s a gene…
• “The market will...”
• “The new sciences of …”
• “Order for free..”
• “Like, dude, like, chill…”
In the real (vs virtual) world
What matters:
• Action
What doesn’t:
• Data
• Information
• Computation
• Learning
• Decision
• …
sense
move Spine
sense
move Spine
Reflex
Reflect
sense
move Spine
Reflex
Reflect
Yet Fragile
Fix?
Obesity, diabetes
Cancer
Parasites, infection
AutoImmune/Inflame
Addiction, psychosis…
Epidemics, war…
Catastrophes
Obfuscate, amplify,…
Accident or necessity?
What matters:
• Action
• Automation
• Limits
• Tradeoffs
wasteful?
fragile?
slow?
?
What we want to avoid.
IEEE TRANS ON SYSTEMS, MAN, AND CYBERNETICS,
JULY 2010, Alderson and Doyle
robust
efficient wasteful
fragile
Want robust
and efficient
wasteful
fragile
robust
efficient
At best we
get one
wasteful
fragile
robust
efficient
Often
neither
???
Bad
theory?
???
?
?
Bad
architectures?
wasteful
fragile
gap?
robust
efficient
Control Comms
Compute Physics
ShannonBode
Turing
Godel
Einstein
Heisenberg
Carnot
Boltzmann
Theory?Deep, but fragmented, incoherent, incomplete
Control Comms
Compute Physics
ShannonBode
Turing
Godel
Einstein
Heisenberg
Carnot
Boltzmann
wasteful?
fragile?
slow?
?
• Each theory one dimension
• Important tradeoffs acrossdimensions
• Progress is encouraging but…
Case studies
wasteful
fragile
Sharpen
hard bounds
Conservation
“laws”?
Case studies
wasteful
fragile
Sharpen
hard bounds
bad Find and
fix bugs
Case studies
wasteful
fragile
Sharpen
hard bounds
bad Find and
fix bugsBad
architectures?
J. Fluid Mech (2010)
Streamlined
Laminar Flow
Turbulent Flow
Turbulence and drag?
Streamlined
Laminar Flow
Turbulent Flow
Turbulence and drag?
uz
x
yFlow
Coherent structures
wU
z x
y
wU
z x
y
Blunted turbulent velocity profile
Laminar
Turbulent
wU
0u
1uu u p u
t R
“turbulence is a highly nonlinear
phenomena”
0u
1uu u p u
t R
Small Large
Robust Simple Organized
Fragile chaocritical Irreducibile
Complexity?
0u
1uu u p u
t R
Small Large
RobustSimple
2d, linear
Organized
Computer
Fragilechaocritical
3d, nonlinearIrreducibile?
Complexity?
mildlynonlinear
highly nonlinear
Physics of Fluids (2011)
wU
z x
y
uz
x
yFlow
upflowhigh-speed
region
downflowlow speed
streak
Blunted turbulent velocity profile
Laminar
Turbulent
wU3D coupling
Coherent structures and turbulent drag
Case studies
wasteful
fragile
Sharpen
hard bounds
bad
Find and
fix bugs
Bad
architectures?
Case studies
wasteful
fragile
Sharpen
hard bounds
Theory + biology case study
Chandra, Buzi, and Doyle
Theory + biology case study
• Universal issues
• Longstanding mystery (century? millennia?)
• Accessible, components “well-known”
• Evolution + physiology + “CDS/CME”
• Broadly relevant
• Science paper in press (w/ Fiona Chandra, Genti Buzi)
• Extreme responses typical
Glycolytic oscillations
Hard tradeoffs between
1. Fragility (disturbance rejection)
2. Amount (of enzymes)
3. Complexity (of enzymes)
• Most ubiquitous/studied “circuit” in science/engineering
• New insights and experiments
Metabolic overhead
Fragility
simple
Metabolic overhead
complex
Hard tradeoffs between
1. Fragility (disturbance rejection)
2. Amount (of enzymes)
3. Complexity (of enzymes)
K Nielsen, PG Sorensen, F Hynne, H-G Busse. Sustained oscillations in glycolysis: an experimental and theoretical study of chaotic and complex periodic behavior and of quenching of simple oscillations. Biophys Chem 72:49-62 (1998).
Experiments
Figure S4. Simulation of two state model (S7.1) qualitatively recapitulates experimental observation from CSTR studies [5] and [12]. As the flow of material in/out of the system is increased, the system enters a limit cycle and then stabilizes again. For this simulation, we take q=a=Vm=1, k=0.2, g=1, u=0.01, h=2.5.
0 20 40 60 80 100 120 140 160 180 2000
1
2
3
4
v=0.03
0 20 40 60 80 100 120 140 160 180 2000
0.5
1
1.5
2
v=0.1
0 20 40 60 80 100 120 140 160 180 2000.2
0.4
0.6
0.8
1
v=0.2
“Standard” Simulation
Figure S4. Simulation of two state model (S7.1) qualitatively recapitulates experimental observation from CSTR studies [5] and [12]. As the flow of material in/out of the system is increased, the system enters a limit cycle and then stabilizes again. For this simulation, we take q=a=Vm=1, k=0.2, g=1, u=0.01, h=2.5.
0 20 40 60 80 100 120 140 160 180 2000
1
2
3
4
v=0.03
0 20 40 60 80 100 120 140 160 180 2000
0.5
1
1.5
2
v=0.1
0 20 40 60 80 100 120 140 160 180 2000.2
0.4
0.6
0.8
1
v=0.2
SimulationExperiments
Why?
simple enzyme
Fragility
Enzyme amount
complex enzyme
lnz p
z p
2 2
0
1ln ln
z z pS j d
z z p
Theorem
z and p functions of enzyme complexity
and amount
L
u
x
qm
M
m
M
Linearized pendulum
on a cart
TCA
Gly
G1P
G6P
F6P
F1-6BP
PEP Pyr
Gly3p
13BPG
3PG
2PG
ATP
NADH
Oxa
Cit
ACA
Autocatalysis
Control
• Autocatalytic • Complex vssimple enzymes
Feedbacks
Translation
L
m
M
• Up• eyes vsno eyes
0
1ln 0S j d
Easy, even with eyes closed
No matter what the length
0
ln 0
Gratuitous fragility
Fragile robustness
S j d
Gratuitous fragility
versus
fragile robustness
0
1ln S j d p
1 1g
p z r rL
small largep L
Up is hard for shorter lengths
Down easy, even with
• eyes closed
• all lengths
Fragility
complex
This is a cartoon,
but can be made
precise.
L
0
1ln S j d p
1p
L
L
Too
fragile Why oscillations?
Side effects of
hard tradeoffs
L
1
1 1
1 1
g mz p z r r
L M
p z r
p z r
m
M
Eyes closed
2 2
0
1ln ln
z z pS j d
z z p
Want r and z large (but p small).
Fragility
up + eyes
lnz p
z p
2 2
0
1ln ln
z z pS j d
z z p
Theorem
up, no eyes
This is a cartoon, but can be made
precise.
L
hopeless
down
TCA
Gly
G1P
G6P
F6P
F1-6BP
PEP Pyr
Gly3p
13BPG
3PG
2PG
ATP
NADH
Oxa
Cit
ACA
Autocatalysis
ControlL
m
M
• Autocatalytic • Complex vssimple enzymes
Feedbacks
Translation
• Up• eyes vsno eyes
Fragility Different “architectures”
up-eyes
This is a cartoon, but can be made
precise.
Ldown
2 2
0
1ln ln
z z pS j d
z z p
Theorem
lnz p
z p
Fragility Different “architectures”
up-eyes
This is a cartoon, but can be made
precise.
Ldown
2 2
0
1ln ln
z z pS j d
z z p
Theorem
complex enzyme
lnz p
z p
simple enzyme
Fragility
up-eyes
This is a cartoon, but can be made
precise.
L
2 2
0
1ln ln
z z pS j d
z z p
Theorem
simple enzyme
lnz p
z p
Too fragile
Fragility
Enzyme amount
complex enzyme
lnz p
z p
2 2
0
1ln ln
z z pS j d
z z p
Theorem
Why oscillations?Side effects of hard tradeoffs
k
z p
z p
10-1
100
10110
0
101
Fragility Biological architectures
achieve hard limits and
use complex enzymes
and networks
2 2
0
1ln ln
z z pS j d
z z p
complex enzyme
Enzyme amount
• Microfluidic experiments
• Yeast strain W303 grown in Ethanol
• Glucose and KCN added anaerobic glycolysis
• NADH measured every 3 s
0 60 120 180 240 3000
50
100
150
t ( seconds)
NADH
Architecture Good architectures
allow for effective
tradeoffs
wasteful
fragile
2 2
0
1ln ln
z z pS j d
z z p
Theorem
• z and p are functions of enzyme complexity and amount• standard biochemistry models• phenomenological
• first principles?
What reviewers say• “If such oscillations are indeed optimal, why are they not
universally present?” • “The approach to establish universality for all biological and
physiological systems is simply wrong. It cannot be done…” • “While the notion of universality is well justified in physics, it is
perhaps not so useful in biological sciences and medicine. To develop a set of universal principles for biological and physiological systems is mostly likely a dream that will never be realized, due to the vast diversity in such systems.”
• “…does not seem to have an understanding or appreciation of the vast diversity of biological and physiological systems…”
• “…desire to develop rigorous framework is understandable, but usually this can be done only by imposing a high degree of abstraction, which would then make the model useless …”
• “… a mathematical scheme without any real connections to biological or medical problems…”
Fragilityhard limits
simple
Overhead, waste
complex
• General• Rigorous• First principle
• Domain specific• Ad hoc• Phenomenological
Plugging in domain details
?
Control Comms
Physics
Wiener
Bode
Kalman
Heisenberg
Carnot
Boltzmann
robust control
• Fundamental multiscale physics• Foundations, origins of
– noise – dissipation– amplification
• General• Rigorous• First principle
?
IEEE TRANS ON AUTOMATIC CONTROL,
FEBRUARY, 2011
Sandberg, Delvenne, and Doyle
http://arxiv.org/abs/1009.2830
Gen
es
Co-factors
Pro
teins
Pre
curs
ors
Autocatalytic feedback
Nu
trie
nts
Core metabolism
DNA
replication
Trans*
Cat
aboli
sm
Carriers
Environment Environment
Huge
Variety
Huge
Variety
Bacterial cell
TCA
Gly
G1P
G6P
F6P
F1-6BP
PEP Pyr
Gly3p
13BPG
3PG
2PG
ATP
NADH
Oxa
Cit
ACA
Autocatalysis
Control
• Autocatalytic • Complex vssimple enzymes
Feedbacks
Catabolism
TCAPyr
Oxa
Cit
ACA
Gly
G1P
G6P
F6P
F1-6BP
PEP
Gly3p
13BPG
3PG
2PG
ATP
NADH
Cat
aboli
sm
AA
Ribosome
transl. Proteins
Pre
curs
ors
ATP
ATP
Inside every cell
Ribosomes
make
ribosomes
Translation: Amino acids
polymerized into proteins
Cat
aboli
sm
AA
Ribosome
RNARNAp
transl. Proteins
xRNAtransc.
Pre
curs
ors
DNADNAp
Repl. Gene
ATP
ATP
Building
Blocks
• Translation
• Transcription
• DNA Replication
Cat
aboli
sm
AA
Ribosome
RNARNAp
transl. Proteins
xRNAtransc.
Pre
curs
ors
DNADNAp
Repl. Gene
ATP
ATP
Enzymes
Building
Blocks
Crosslayer
autocatalysis
Macro-layers
Inside every cell
NucleotidesCatabolism
Pre
curs
ors
Taxis and
transport
Nu
trie
nts
Carriers
Core metabolism
Same
12
in all
cells
Same
8
in all
cells
Gen
es
Co-factors
Polymerization and complex
assemblyP
rotein
sP
recu
rsors
Autocatalytic feedback
Taxis and
transport
Nu
trie
nts
Core metabolism
DNA
replication
Trans*
Cat
aboli
sm
Carriers
100 104 to ∞in one
organisms
Huge
Variety
12
8
Other examples
Clothing
Lego
Money
Cell biology
Shirt
Slacks
Jacket Tie
T-Shirt
Socks
ShoesCoatShorts
Soft layering
Wool
Cotton
Nylon
Silk
Polyester
Rayon
Shirt
Slacks
Jacket Tie
T-Shirt
Socks
ShoesCoatShorts
Modularity?
Shirt
Slacks
Jacket Tie
T-Shirt
Socks
ShoesCoatShorts
Given a wardrobe (set of garments)
1 << # outfits << # non-outfits
(random heaps are of garments are never outfits)
Shirt
Slacks
Jacket Tie
T-Shirt
Socks
ShoesCoatShorts
1 << # outfits << # non-outfits
(random heaps are of garments are never outfits)
large thin
large thin
1 << # toys << # non-toys
“order for free”
Letters and words
• 9 letters: adeginorz
• 9!= 362,880 sequences of 9 letters
• Only “organized” is a word
1 << (# words) << (# non-words)
large thin
Shirt
Slacks
Jacket sweater
T-Shirt
Socks
ShoesCoatShorts
Inne
r
Mid
dle
Oute
r
Skin
Hidden
Robust to variations in
• weather
• activity
• appearance requirements
• wear and tear
• cleaning
System
constraints
Wool Cotton NylonSilk Polyester Rayon
Cloth
Shirt
Slacks
Jacket Tie
T-Shirt
Socks
ShoesCoatShorts
Sewing
Cloth
Thread
Fiber
Garments
Xform
Xform
Xform
Modularity 2.0
Prevents unraveling of lower layers
Cloth
Thread
Fiber
Garments
Xform
Xform
Xform
Hidden,
large, thin,
nonconvex
Outfits
Assemble
Wool
Cotton
Nylon
Silk
Polyester
Rayon
Shirt
Slacks
Jacket Ti
e
T-Shirt
Socks
ShoesCoatBoxers
The
hourglass?
Dress Shirt Slacks Lingerie Coat Scarf Tie
Garments
Cloth
Sewing
Wool Cotton NylonSilk Polyester
Material technologies
Rayon
Horizontal networks of fibers
Ver
tica
l d
ecom
posi
tion
Horizontal networks of garments
Ver
tica
l d
ecom
posi
tion
Inne
r
Mid
dle
Ou
ter
Horizontal networks of fibers
Horizontal networks of garments
Cloth
Thread
Fiber
Garments
Xform
Xform
Xform
Cloth
Thread
Fiber
Garments
Xform
Xform
Xform
Universal strategies?
Garments have
limited access to
threads and fibers
constraints on
cross-layer
interactions
quantization
for robustness
Even though
garments seem
analog/continuous
Prevents unraveling of lower layers
Fiber
Geographically diverse sources
Diverse fabric
Functionally diverse garments
General
purpose
machines Diverse Thread
sew
knit, weave
spin
Functionally diverse garments
Cloth
Thread
Fiber
Garments
Fragilities?
Cloth
Thread
Fiber
Garments
Scalable
Sustainable?
Embedded
virtual
actuator/
sensor
Network
cable
Controller
Lib
App
DIF
Networked/embedded/layered
Lib
Physical
plant
Embedded
virtual
actuator/
sensor
Network
cable
Controller
DIF
Physical
plant
Meta-layering of cyber-phys control
Physiology
Organs
Cells
Layered architectures
Physiology
Organs
Layered architectures
Cells
Physiology
Organs
Meta-layers
Prediction
Goals
Actions
errors
ActionsCo
rte
x
Which blue line is longer?
Which blue line is longer?
Which blue line is longer?
Which blue line is longer?
Which blue line is longer?
Meta-layers
Fast,
Limited
scope
Slow,
Broad
scope
Unfortunately, we’re not
sure how this all works.
Shirt
Slacks
Jacket sweater
T-Shirt
Socks
ShoesCoatShorts
Inne
r
Mid
dle
Oute
r
Skin
Selection
assembly
Cloth
Thread
Fiber
Garments
Xform Ctrl Mgmt
Networked,
universal,
layeredXform Ctrl Mgmt
Xform Ctrl Mgmt
Xform Ctrl Mgmt
Do
ma
in s
pe
cific
Cloth
Thread
Fiber
Garments
Xform Ctrl Mgmt
Networked,
universal,
layeredXform Ctrl Mgmt
Xform Ctrl Mgmt
Xform Ctrl Mgmt
Dem
an
d
Su
pp
ly
Cloth
Thread
Fiber
Garments
Xform Ctrl Mgmt
Networked,
universal,
layeredXform Ctrl Mgmt
Xform Ctrl Mgmt
Xform Ctrl Mgmt
Co
ntro
l
Su
pp
ly
Complexity?
Yet Fragile
Fix?
Obesity, diabetes
Cancer
Parasites, infection
AutoImmune/Inflame
Addiction, psychosis…
Epidemics, war…
Catastrophes
Obfuscate, amplify,…
Accident or necessity?
What matters:
• Action
• Automation
• Limits
• Tradeoffs
THE END OF THEORYScientists have always relied on hypothesis and experimentation. Now, in the era of
massive data, there’s a better way.
“All models are wrong, and increasingly
you can succeed without them.”
Save our
children
Peta-
philia
There is a
treatment.
New words
• Peta-philia: Perverse love
of data and computation
• Peta-fop: Someone who
profits from peta-philia
• Exa-duhs: Loss of clue
from excessive peta-philia
Fortunately
there seems
to be a
treatment
Not yet in
widespread use
Peta-
philia
Case studies
wasteful
fragile
Sharpen
hard bounds
bad
Find and
fix bugs
wasteful
fragile
bad
Find and
fix bugs
App AppApplications
Router
Chiang, Low, Calderbank, and Doyle
TCPIP
Physical
MAC
Switch
MAC MAC
Pt to Pt Pt to Pt
Diverse applications
Layered architectures
Control
Compute
Internet
optimization
operating systems
“Layering as optimization”
robust control
• 10+ years of progress & impact…
• Static optimization dynamic control
• Wireless, scheduling, net coding, …
• But….
• Something is wrong architecturally
• Better protocols/control won’t fix it
• Design: from protocols to architectures
Layered architectures everywhere
• Computers, Internet, software… cyber-physical?
• Bacterial biosphere
• Evo-devo
• Brain
• Lego, clothing, supply chain, …
• Useful “comparative physiology”
TCPIP
Physical
Diverse applications
Layered architectures
Diverse
OS
Physical
Diverse applications
Layered architectures
• OS allocates/shares
– diverse resources among
– diverse applications
• “Strict layering” crucial,
e.g. clearly separate
– Application name space
– Logical (virtual)
name/address space
– Physical (name/) address
space
• Name resolution w/in appls
• Name/addr transl X layers
OS
Physical
Diverse applications
Layered architectures
In operating systems:
Don’t cross layers
(rings)
Direct
access to
physical
memory?
In programming:
No global variables
Problems with leaky layering
Modularity benefits are lost
• Global variables? @$%*&!^%@&
• Poor portability of applications
• Insecurity of physical address space
• Fragile to application crashes
• No scalability of virtual/real addressing
• Limits optimization/control by duality?
App AppApplications
Router
IP addresses
interfaces
(not nodes)
App App
IPC
Global
and direct
access to
physical
address!
Robust?
• Secure
• Scalable
• Verifiable
• Evolvable
• Maintainable
• Designable
• …
DNS
IP addresses
interfaces
(not nodes)
OS
Diverse hardware
Layered architectures
Little diversity
Diverse applications
OS
Diverse hardware
Layered architectures
“Hourglass”
Diverse applications
Physical
IP
TCP
Application
Naming and addressing need to be
• resolved within layer
• translated between layers
• not exposed outside of layer
Related “issues”
• DNS
• NATS
• Firewalls
• Multihoming
• Mobility
• Routing table size
• Overlays
• …
Architecture
is not graph
topology.
Architecture
facilitates
arbitrary graphs.
Persistent
errors and
confusion.
Notices of the AMS, 2009
wasteful
fragile
slowGood case studies
bad
worse
“New sciences” of “complexity” and “networks”?
“New sciences” of “complexity” and “networks”? worse
• Edge of chaos• Self-organized criticality• Scale-free “networks”• Creation “science”• Intelligent design• Financial engineering• Risk management• “Merchants of doubt”• …
• Science as pure fashion• Ideological• Political• Evangelical• Nontech trumps tech
D. Alderson, NPS 178
Complexity = hard to understand, explain
• “Complexity science” = persistently in error
– bionetworks: gene regulation, metabolism, PPI
– wildfires, earthquakes
– Internet, power grid
• Minimal impact on technology
• Diminishing impact on biology
• Of concern in medicine, neuroscience
Complex systems?
Fragile
• Scale
• Dynamics
• Nonlinearity
• Nonequlibrium
• Open
• Feedback
• Adaptation
• Intractability
• Emergence
• …
Even small
amounts can
create
bewildering
complexity
Complex systems?
Fragile
• Scale
• Dynamics
• Nonlinearity
• Nonequlibrium
• Open
• Feedback
• Adaptation
• Intractability
• Emergence
• …
• Scale
• Dynamics
• Nonlinearity
• Nonequlibrium
• Open
• Feedback
• Adaptation
• Intractability
• Emergence
• …
Robust
Complex systems?
• Resources
• Controlled
• Organized
• Structured
• Extreme
• Architected
• …
Robust complexity
• Scale
• Dynamics
• Nonlinearity
• Nonequlibrium
• Open
• Feedback
• Adaptation
• Intractability
• Emergence
• …
Architecture
• Resources
• Controlled
• Organized
• Structured
• Extreme
• Architected
• …
Robust complexity
• Scale
• Dynamics
• Nonlinearity
• Nonequlibrium
• Open
• Feedback
• Adaptation
• Intractability
• Emergence
• …
New
wordsFragile complexity
Emergulent
Emergulence
at the edge of
chaocritiplexity
• Scale
• Dynamics
• Nonlinearity
• Nonequlibrium
• Open
• Feedback
• Adaptation
• Intractability
• Emergence
• …
wU
z x
y
Blunted turbulent velocity profile
Laminar
Turbulent
wU
0u
1uu u p u
t R
“turbulence is a highly nonlinear
phenomena”
0u
1uu u p u
t R
Small Large
RobustSimple
2d, linear
Organized
Computer
Fragilechaocritical
3d, nonlinearIrreducibile?
Complexity?
mildlynonlinear
highly nonlinear
Irreducibility and “intelligent design”
Rube Goldberg
The essential
ID argument
If biology is like this,
then it could not have evolved.
• This is actually true, and in fact…
• If biology is like this,
• then it would be too fragile to persist,
• and would need the constant intervention of
supernatural forces
The flaw
• It is too fragile to actually build.
• Neither biology nor (most of) technology is
anything like this.
• Who said otherwise? Lots of real scientists!
• Oops! (But we are too fragile and unsustainable.)
This is a
cartoon.