Post-Moores Law Computation

The end of Moores Law is in sight! Physical limits to:

density of binary logic devices speed of operation

Requires a new approach to computation Significant challenges Will broaden & deepen concept of computation in natural

& artificial systems

2

Differences in Spatial Scale

2.71828

0010111001100010100

(Images from Wikipedia)

X := Y / Z

Differences in Time Scale

P[0] := N i := 0 while i < n do if P[i] >= 0 then q[n-(i+1)] := 1 P[i+1] := 2*P[i] - D else q[n-(i+1)] := -1 P[i+1] := 2*P[i] + D end if i := i + 1 end while

(Images from Wikipedia)

Convergence of Scales

Implications of Convergence
Computation on scale of physical processes Fewer levels between computation & realization Less time for implementation of operations Computation will be more like underlying physical

processes
Post-Moores Law computing

greater assimilation of computation to physics

Computation is Physical
Computation is physical; it is necessarily embodied in a device whose behaviour is guided by the laws of physics and cannot be completely captured by a closed mathematical model. This fact of embodiment is becoming ever more apparent as we push the bounds of those physical laws. Susan Stepney (2004)

Cartesian Duality in CS
Programs as idealized mathematical objects Software treated independently of hardware Focus on formal rather than material Post-Moores Law computing:
less idealized more dependent on physical realization

More difficult But also presents opportunities

8

Embodied Cognition
Rooted in pragmatism of James & Dewey Deweys Principle of Continuity:
no break from most abstract cognitive activities down thru sensory/motor engagement with physical world to foundation in biological & physical processes

Cognition: emergent pattern of purposeful interactions

between organism & environment

Cf. also Piaget, Gibson, Heidegger, Polanyi, Merleau-

Ponty
10

Embodiment, AI & Robotics

Dreyfus & al.:
importance & benefits of embodiment in cognition there are many things we know merely by virtue of having a

body
embodiment essential to cognition, not incidental to cognition (& info. processing)

Brooks, Pfeifer, Iida & al.: increasing understanding of

value & exploitation of embodiment in AI & robotics

11

Embodiment & Computation

Embodiment = the interplay of information and

physical processes Pfeifer, Lungarella & Iida (2007)

Embodied computation =

information processing in which physical realization & physical environment play unavoidable & essential role

12

Embodied Computing
Includes computational processes:
that directly exploit physical processes for

computational ends

in which information representations and

processes are implicit in physics of system and environment growth, assembly, development, transformation, reconfiguration, or disassembly of the physical system embodying the computation

in which intended effects of computation include

13

Related Work: Hamann and Wrn (2007)

An embodied computation system has at least two levels
adaptive SO & collective behavior at higher levels results

from
spatially local interactions of microscopic control devices

aspects of embodiment:
lack of separation between processor and memory essential dependence of computation on physical world

Seems to conflate embodiment with other issues

14

Related Work: Susan Stepney (2008)

Material computation and in materio computers Systems in which physical substrate naturally computes Focus on non-living substrates Primarily concerned with use of physical materials to implement

computations
Less concerned with use of computational processes to organize &

control matter and energy

Cautions against ill-advised application of notions from Turing

computation
15

Three Modes of Computation

Offline computation Embedded computation Embodied computation

16

Offline Computation

Physical input converted to computational medium Abstract computation Physical representation of results Computation as evaluation of function
(Images from Wikipedia)

17

Embedded Computation

In ongoing interaction with environment Non-terminating Real-time feedback through environment

(Images from Wikipedia)

18

Embedded Computation
Sensors & actuators still convert to/from

computational medium
Computation is effectively abstract Physical considerations confined to:
embedding device environment transducers basic physical characteristics of processor

19

Embodied Computation
Embodied (vs. embedded)

computation:
little or no abstract

computation
computation as physical

process in continuing interaction with other physical processes

(Images from Wikipedia)

20

Embedded Computation
environment purely formal

agent

interface

21

Embodied Computation
environment purely formal

agent

interface

22

23

Strengths of Embodied Computation

Information often implicit in:
its physical realization its physical environment

Many computations performed for free by physical

substrate
Representation & info. processing emerge as regularities in

dynamics of physical system

24

Example: Diffusion

Occurs naturally in many

fluids
Can be used for many

computational tasks
broadcasting information massively parallel search

for optimization, constraint satisfaction etc.

Expensive with

conventional computation
Free in many physical

systems
25

Example: Saturation
Sigmoids in ANNs &

universal approx.
Many physical sys. have

sigmoidal behavior
Growth process

saturates
Resources become

saturated or depleted
EC uses free sigmoidal

behavior
26

(Images from Bar-Yam & Wikipedia)

Example: Negative Feedback

Pos. feedback for growth

& extension
Neg. feedback for:
stabilization delimitation separation creation of structure

Free from
evaporation dispersion degradation
27

 Many algs. use

randomness
escape from local

Example: Randomness

optima
symmetry breaking deadlock avoidance exploration

For free from:

noise uncertainty imprecision
28

(Image from Anderson)

Respect the Medium

Conventional computer technology tortures the medium

to implement computation
Embodied computation respects the medium Goal of embodied computation:

Exploit the physics, dont circumvent it

29

30

p abs. comp. d

Embodied Computation for Action

EC uses physics for

information processing
Information system

governs matter & energy in physical computer

P phys. comp. D

EC uses information

processing to govern physical processes

31

Embodied Computation for Physical Effect

Natural Embodied Computation: governs physical processes in organisms body physical interactions with other organisms & environment Often, result of Embodied Computation is not information,

but action, including:

self-action self-transformation self-construction self-repair self-reconfiguration
32

Disembodied Computation
If purpose is information processing Then represent information with small quantities of matter

or energy
Objective: state change involves small change of matter or

energy
Limit: disembodied computation & communication Pure form without need for matter

33

EC Controlling Matter & Energy

May want to move more rather than less Physical effects may be direct results of computation No clear distinction between

processors & actuators

Examples: Algorithmic assembly by
(figure from Rothemund)

DNA computation (Winfree)

Nanostructure synthesis & control by

molecular combinator reduction (MacLennan)

34

Function and Structure

Structure

Function

structure implements function function creates structure Regulation: hetero- and auto Causation: formal: transformation of form, formal processes material: substratum of form & transformation, complex

media
efficient: transforming agent, dissipative system final: function or purpose of computation
35

Example: Balancing Exploration and Exploitation

How do we balance
the gathering of information (exploration) with the use of the information we have already

gathered (exploitation)
E.g., ant foraging Random wandering leads to exploration Positive feedback biases toward exploitation Negative feedback biases toward exploration
36

Some Non-Turing Characteristics of Embodied Computation

Operates in real time and real space with real matter and real energy and hence non-ideal aspects of physical realization Often does not terminate Often has no distinct inputs or outputs Often purpose is not to get an answer from an input Often purpose is not to control fixed agent Different notions of equivalence and universality
37

Embodied Computation is Non-Turing Computation

It is computing because processes are formal All models exist in a frame of relevance determined by:
questions it is intended to answer and corresponding idealizing assumptions

The needs and concerns of embodied computation are

incompatible with frame of relevance of Church-Turing model of computation

Embodied computation has a different frame of relevance and

needs different models

38

Disadvantages of Embodied Computation

Less idealized Energy issues Lack of commonly accepted and widely applicable

models of computation
But nature provides good examples of how:
computation can exploit physics without opposing it information processing systems can interact fruitfully

with physical embodiment of selves & other systems

39

Design of Emergent Computation Systems

1) Understand 2) Abstract 3) Realize

40

41

Motivation for Artificial Morphogenesis

Nanotechnology challenge: how to organize millions of relatively

simple units to self-assemble into complex, hierarchical structures

It can be done: embryological development Morphogenesis: creation of 3D form Characteristics: structure implements function function creates structure no fixed coordinate framework soft matter sequential (overlapping) phases temporal structure creates spatial structure
42

Artificial Morphogenesis

Morphogenesis can

coordinate:

proliferation movement disassembly

to produce complex,

hierarchical systems
Future nanotechnology:

use artificial morphogenesis for multiphase selforganization of complex, functional, active hierarchical systems

(Images from Wikipedia)

43

Reconfiguration & Metamorphosis

Degrees of metamorphosis: incomplete complete Phase 1: partial or complete

dissolution
Phase 2: morphogenetic

reconfiguration

(Images from Wikipedia)

Microrobots, Cells, and Macromolecules

45

Components
Both active and passive Simple, local sensors (chemical, etc.) Simple effectors local action (motion, shape, adhesion) signal production (chemical, etc.) Simple regulatory circuits (need not be electrical) Self-reproducing or not Ambient energy and/or fuel
46

Metaphors for Morphogenesis

Donna Haraway: Crystals, Fabrics, and Fields:

Metaphors that Shape Embryos (1976) a history of embryology

The fourth metaphor is soft matter: 1. crystals 2. fabrics 3. fields 4. soft matter
47

Self-Organization of Physical Pattern and 3D Form

(Images from Wikipedia)

48

Fundamental Processes*
directed mitosis differential growth apoptosis differential adhesion condensation contraction matrix modification migration

diffusion chemokinesis chemotaxis haptotaxis

cell-autonomous

modification of cell state

asymmetric mitosis temporal dynamics

inductive modif. of state

hierarchic emergent
49

* Salazar-Ciudad, Jernvall, & Newman (2003)

50

Goals & Requirements

Continuous processes Continuous tissues Complementarity Intensive quantities Embodies computation in Coordinate-independent

behavioral description tation

Mathematical interpre Operational interpretation Influence models Multiple space & time

solids, liquids, gases especially soft matter

scales

Active and passive elements Energetic issues

Stochastic

51

Change Equations
Want to maintain complementarity between discrete

and continuous descriptions:

Neutral change equation:

52

Example of Notation: Simple Diffusion

A Simple Diffusion System

Example Morphogenesis Goal

55

56

Clock-and-Wavefront Model of Segmentation

Vertebrae: humans have 33, chickens 35, mice 65, corn

snake 315 characteristic of species

How does developing embryo count them? Segments also govern development of organs Clock-and-wavefront model of Cooke & Zeeman

(1976), recently confirmed (2008)

Depends on clock, excitable medium (cell-to-cell

signaling), and diffusion

57

Simulated Segmentation by Clock-and-Wavefront Process

58

500

1000

2000

4000

Effect of Growth Rate

5000
59

Timing Signals
Growth duration:

Segmentation clock:

60

Conditional Notation

61

Growth of Undifferentiated Tissue

T = density of tissue in terminal (tailbud) state u = direction of motion r = rate of movement or growth M = density of undifferentiated tissue = length of tailbud

62

Growth of Undifferentiated Tissue

violet = T = density of tail tissue yellow = M = density of undifferentiated tissue

63

Wave Propagation
The tissue is an active medium Clock signal causes a patch of tail tissue to fire: emit

a pulse of (segmentation morphogen)

It diffuses and degrades Sufficiently high stimulates nearby tissue to fire But after tissue fires, it enters a refractory period

(determined by a variable )
Ensures unidirectional propagation
64

Equations

65

Wave Propagation

66

Caudal Morphogen
Rapidly accumulates in tail tissue Diffuses and degrades Represents proximity to tail tissue

67

Rostral Morphogen
Accumulates in differentiated segments (S > 0) Diffuses and degrades Represents proximity to differentiated segments

68

Segment Differentiation
tissue differentiates into segment tissue when:
segmentation signal () passes through sufficiently far from tail (C < threshold) sufficiently far from previous segments (R < threshold)

69

Segmentation Process

green = rostral (head) morphogen blue = caudal (tail) morphogen brown = differentiated segment tissue

70

71

Posterior Segment Border

Segment tissue differentiates into posterior border

tissue when:
segmentation signal () passes through caudal morphogen (C) concentration is high

72

Anterior Segment Border

Segment tissue differentiates into anterior border tissue

when:
segmentation signal () passes through rostral morphogen (R) concentration is high

73

Differentiation of Anterior and Posterior Segment Tissue

lime = anterior segment tissue (A) cyan = posterior segment tissue (P)

74

75

Imaginal Disks

76

Anterior/Posterior Position
Anterior and posterior border tissues emit anterior (a)

and posterior (p) morphogens, which diffuse and degrade

Establish opposing gradients by which position can be

determined

77

Exterior Surface Detection

Inverse quorum sensing: detect when density of

neighbors is below a threshold tissue

Implemented by morphogen diffusing from segment Modeled by convolution with Gaussian kernel

determined by diffusion parameters

78

Differentiation of Imaginal Disks (in 2D)

Imaginal disk tissue differentiates when:
anterior border morphogen in correct range posterior border morphogen in correct range segment density is sufficiently low

79

Formation of Imaginal Disks

orange = anterior border morphogen (a) turquoise = posterior border morphogen (p) red = imaginal disk tissue (I)

80

First Steps Toward Leg Growth

Imaginal disk tissue

differentiates to be in T (terminal) state outward (i.e., grad S) produce undifferentiated leg tissue (ready for clock-and-wavefront)

These cell orient

Begin to move and

Still buggy
81

Conclusions
Post-Moores Law computing will require a greater

assimilation and computation to physics

Novel physical processes can be used for

computational purposes
Computational processes can be used for physical

purposes
Embodied computation takes the physicality of

computation seriously
Artificial morphogenesis is an example of truly

embodied computation
82

More Information?
MacLennan@utk.edu Homepage: http://web.eecs.utk.edu/~mclennan/

[sic]
Emergent Computation Project:

http://web.eecs.utk.edu/~mclennan/EC/

83