19 vues

Titre original : 0807.3790v1

Transféré par nemismd

- PRP Anna University QA
- kk
- Exponential&Possion
- poger_pem1
- 2.MARKOV CHAAINS.pptx
- Rr411501 Performance Evaluation of Computer Systems
- Markov Chains Chap 5
- Chapter_2.pdf
- Stochastic Processes
- Stochast
- psappha
- ds
- Final
- revision.pdf
- Atcs Exercises8
- stochastic calculus
- Nata Textb
- 10.1.1.116
- chap2
- Locally Associative Planes of Combinatorially Liouville Systems and Convergence Methods

Vous êtes sur la page 1sur 29

arXiv:0807.3790v1 [nlin.AO] 24 Jul 2008

Abstract

We have proposed a novel model of general quantum, stochastic and chaotic psychody-

namics. The model is based on the previously developed Life–Space Foam (LSF) framework

to motivational and cognitive dynamics. The present model extends the LSF–approach

by incorporating chaotic and topological non-equilibrium phase transitions. Such extended

LSF–model is applied for rigorous description of multi–agent joint action. The present model

is related to Haken–Kelso–Bunz model of self-organization in the human motor system (in-

cluding: multi-stability, phase transitions and hysteresis effects, presenting a contrary view

to the purely feedback driven neural systems), as well as the entropy–approach to adaptation

in human goal–directed motor control.

Keywords: Quantum probability, Life–Space Foam, noisy decision making, chaos, topolog-

ical phase transitions, multi–agent joint action, goal–directed motor control

∗

Human Systems Integration, Land Operations Division, Defence Science & Technology Organisation, P.O.

Box 1500, Edinburgh SA 5111, Australia (Vladimir.Ivancevic@dsto.defence.gov.au)

†

School of Electrical and Information Engineering, University of South Australia, Mawson Lakes, S.A. 5095,

Australia (Tijana.Ivancevic@unisa.edu.au)

1

1 Introduction modules.

According to Freeman and Vitiello (2006),

Classical physics has provided a strong founda- many–body quantum field theory appears to be

tion for understanding brain function through the only existing theoretical tool capable to ex-

measuring brain activity, modelling the func- plain the dynamic origin of long–range correla-

tional connectivity of networks of neurons with tions, their rapid and efficient formation and

algebraic matrices, and modelling the dynamics dissolution, their interim stability in ground

of neurons and neural populations with sets of states, the multiplicity of coexisting and pos-

coupled differential equations (Freeman, 1975, sibly noninterfering ground states, their degree

2000). Various tools from classical physics en- of ordering, and their rich textures relating to

abled recognition and documentation of aspects sensory and motor facets of behaviors. It is his-

of the physical states of the brain; the struc- torical fact that many–body quantum field the-

tures and dynamics of neurons, the operations ory has been devised and constructed in past

of membranes and organelles that generate and decades exactly to understand features like or-

channel electric currents; and the molecular and dered pattern formation and phase transitions

ionic carriers that implement the neural ma- in condensed matter physics that could not be

chineries of electrogenesis and learning. They understood in classical physics, similar to those

support description of brain functions at sev- in the brain.

eral levels of complexity through measuring neu- The domain of validity of the ‘quantum’ is

ral activity in the brains of animal and human not restricted to the microscopic world (Umeza-

subjects engaged in behavioral exchanges with va, 1993). There are macroscopic features of

their environments. One of the key properties classically behaving systems, which cannot be

of brain dynamics are the coordinated oscil- explained without recourse to the quantum dy-

lations of populations of neurons that change namics. This field theoretic model leads to the

rapidly in concert with changes in the environ- view of the phase transition as a condensation

ment (Freeman and Vitiello, 2006; Ivancevic, that is comparable to the formation of fog and

2006a, 2007b). Also, most experimental neuro- rain drops from water vapor, and that might

biologists and neural theorists have focused on serve to model both the gamma and beta phase

sensorimotor functions and their adaptations transitions. According to such a model, the

through various forms of learning and memory. production of activity with long–range correla-

Reliance has been placed on measurements of tion in the brain takes place through the mech-

the rates and intervals of trains of action poten- anism of spontaneous breakdown of symmetry

tials of small numbers of neurons that are tuned (SBS), which has for decades been shown to de-

to perceptual invariances and modelling neu- scribe long-range correlation in condensed mat-

ral interactions with discrete networks of simu- ter physics. The adoption of such a field theo-

lated neurons. These and related studies have retic approach enables modelling of the whole

given a vivid picture of the cortex as a mosaic cerebral hemisphere and its hierarchy of com-

of modules, each of which performs a sensory ponents down to the atomic level as a fully in-

or motor function; they have not given a pic- tegrated macroscopic quantum system, namely

ture of comparable clarity of the integration of as a macroscopic system which is a quantum

2

system not in the trivial sense that it is made, with that of motor variability (biomechanical

like all existing matter, by quantum compo- degrees of freedom). Moreover, their theory

nents such as atoms and molecules, but in the accommodates the idea that the human mo-

sense that some of its macroscopic properties tor control mechanism uses internal ‘functional

can best be described with recourse to quan- synergies’ to regulate task–irrelevant (redun-

tum dynamics (see Freeman and Vitiello, 2006 dant) movement.

and references therein). Until recently, research concerning sensory

It is well–known that non-equilibrium phase processing and research concerning motor con-

transitions (Haken, 1983, 1993, 1996) are phe- trol have followed parallel but independent paths.

nomena which bring about qualitative physical The partitioning of the two lines of research in

changes at the macroscopic level in presence practice partly derived from and partly fostered

of the same microscopic forces acting among a bipartite view of sensorimotor processing in

the constituents of a system. Phase transi- the brain – that a sensory/perceptual system

tions can also be associated with autonomous creates a general purpose representation of the

robot competence levels, as informal specifica- world which serves as the input to the motor

tions of desired classes of behaviors for robots systems (and other cognitive systems) that gen-

over all environments they will encounter, as erate action/behavior as an output. Recent re-

described by Brooks’ subsumption architecture sults from research on vision in natural tasks

approach (Brooks, 1986, 1989, 1990). The dis- have seriously challenged this view, suggest-

tributed network of augmented finite–state ma- ing that the visual system does not generate

chines can exist in different phases or modal- a general–purpose representation of the world,

ities of their state–space variables, which de- but rather extracts information relevant to the

termine the systems intrinsic behavior. The task at hand (Droll et al, 2005; Land and Hay-

phase transition represented by this approach hoe, 2001). At the same time, researchers in

is triggered by either internal (a set–point) or motor control have developed an increasing un-

external (a command) control stimuli, such as derstanding of how sensory limitations and sen-

a command to transition from a sleep mode to sory uncertainty can shape the motor strategies

awake mode, or walking to running. that humans employ to perform tasks. More-

On the other hand, it is well–known that over, many aspects of the problem of sensori-

humans possess more degrees of freedom than motor control are specific to the mapping from

are needed to perform any defined motor task, sensory signals to motor outputs and do not

but are required to co-ordinate them in order to exist in either domain in isolation. Sensory

reliably accomplish high-level goals, while faced feedback control of hand movements, coordi-

with intense motor variability. In an attempt nate transformations of spatial representations

to explain how this takes place, Todorov and and the influence of processing speed and atten-

Jordan (2002) formulated an alternative the- tion on sensory contributions to motor control

ory of human motor coordination based on the are just a few of these. In short, to under-

concept of stochastic optimal feedback control. stand how human (and animal) actors use sen-

They were able to conciliate the requirement sory information to guide motor behavior, we

of goal achievement (e.g., grasping an object) must study sensory and motor systems as an

3

integrated whole rather than as decomposable the authors monitored the interactions of eight

modules in a sequence of discrete processing pairs of subjects as they moved their fingers

steps (Knill et al, 2007). with and without a view of the other individ-

Cognitive neuroscience investigations, inclu- ual in the pair.

ding fMRI studies of human co–action, sug- Recently developed Life Space Foam (LSF)

gest that cognitive and neural processes sup- model (Ivancevic and Aidman, 2007) is an inte-

porting co–action include joint attention, ac- gration of two modern approaches to cognition:

tion observation, task sharing, and action co- (i) dynamical field theory (DFT, Amari, 1977;

ordination (Fogassi et al, 2005; Knoblich and Schöner, 2007) and (ii) quantum–probabilistic

Jordan, 2003; Newman et al, 2007; Sebanz at dynamics (QP, Glimcher, 2005; Busemayer et

al, 2006). For example, when two actors are al, 2006). In this paper we expand the LSF–

given a joint control task (e.g., tracking a mov- concept to model decision making process in

ing target on screen) and potentially conflicting human–robot joint action and related LSF–phase

controls (e.g., one person in charge of accelera- transitions.

tion, the other – deceleration), their joint per-

formance depends on how well they can antici-

pate each other’s actions. In particular, better

2 Classical versus Quantum

coordination is achieved when individuals re- Probability

ceive real–time feedback about the timing of

each other’s actions (Sebanz at al, 2006). As quantum probability in human cognition and

A developing field in coordination dynam- decision making has recently become popular,

ics involves the theory of social coordination, let us briefly describe this fundamental concept

which attempts to relate the DC to normal hu- (for more details, see Ivancevic, 2007a, 2007c,

man development of complex social cues follow- 2008b).

ing certain patterns of interaction. This work is

aimed at understanding how human social in- 2.1 Classical Probability and

teraction is mediated by meta-stability of neu- Stochastic Dynamics

ral networks. fMRI and EEG are particularly Recall that a random variable X is defined by

useful in mapping thalamocortical response to its distribution function f (x). Its probabilistic

social cues in experimental studies. In partic- description is based on the following rules: (i)

ular, a new theory called the Phi complex has P (X = xi ) is the probability that X = xi ;

been developed by S. Kelso and collaborators, and (ii) P (a ≤ X ≤ b) is the probability that

to provide experimental results for the theory X lies in a closed interval [a, b]. Its statistical

of social coordination dynamics (see the recent description is based on: (i) µX or E(X) is the

nonlinear dynamics paper discussing social co- mean or expectation of X; and (ii) σ X is the

ordination and EEG dynamics of Tognoli et al, standard deviation of X. There are two cases

2007). According to this theory, a pair of phi of random variables: discrete and continuous,

rhythms, likely generated in the mirror neuron each having its own probability (and statistics)

system, is the hallmark of human social coor- theory.

dination. Using a dual-EEG recording system,

4

A discrete random variable X has only a To emphasize this similarity even further,

countable number of values {xi }. Its distribu- as well as to set–up the stage for the path inte-

tion function f (xi ) has the following properties: gral, recall the notion of a cumulative distribu-

tion function of a random variable X, that is a

P (X = xi ) = f (xi ), f (xi ) ≥ 0, function F :R− → R, defined by

X

f (xi ) dx = 1.

i

F (a) = P (X) ≤ a.

Statistical description of X is based on its In particular, suppose that f (x) is the distri-

discrete mean value µX and standard deviation bution function of X. Then

σ X , given respectively by X

F (x) = f (xi ), or

X

µX = E(X) = xi f (xi ), x i ≤x

Z ∞

i

q F (x) = f (t) dt,

σX = E(X 2 ) − µ2X . −∞

Here f (x) is a piecewise continuous function random variable. In either case, F (a) ≤ F (b)

such that: whenever a ≤ b. Also,

Z b

P (a ≤ X ≤ b) = f (x) dx, f (x) ≥ 0, x− lim F (x) = 0 and lim F (x) = 1,

→−∞ x−

→∞

a

Z ∞ Z

f (x) dx = f (x) dx = 1. that is, F (x) is monotonic and its limit to the

−∞ R left is 0 and the limit to the right is 1. Further-

more, its cumulative probability is given by

Statistical description of X is based on its

continuous mean µX and standard deviation P (a ≤ X ≤ b) = F (b) − F (a),

σ X , given respectively by

Z ∞ and the Fundamental Theorem of Calculus tells

µX = E(X) = xf (x) dx, us that, in the continuum case,

−∞

f (x) = ∂x F (x).

q

σ X = E(X 2 ) − µ2X .

Now, recall that Markov stochastic process

Now, let us observe the similarity between

is a random process characterized by a lack of

the two descriptions. The same kind of similar-

memory, i.e., the statistical properties of the

ity between discrete and continuous quantum

immediate future are uniquely determined by

spectrum stroke P. Dirac when he suggested the

the present, regardless of the past (Gardiner,

combined

R integral approach, that he denoted 1985; Ivancevic, 2006b).

by Σ – meaning ‘both integral and sum at For example, a random walk is an exam-

once’: summing over a discrete spectrum and ple of the Markov chain, i.e., a discrete–time

integration over a continuous spectrum. Markov process, such that the motion of the

5

system in consideration is viewed as a sequence equation reads (Gardiner, 1985):

of states, in which the transition from one state

to another depends only on the preceding one, mv̇ = R(t) − βv,

or the probability of the system being in state

where m denotes the mass of the particle and

k depends only on the previous state k − 1.

v its velocity. The right–hand side represent

The property of a Markov chain of prime im-

the coupling to a heat bath; the effect of the

portance in biomechanics is the existence of an

random force R(t) is to heat the particle. To

invariant distribution of states: we start with

balance overheating (on the average), the par-

an initial state x0 whose absolute probability is

ticle is subjected to friction β. In humanoid

1. Ultimately the states should be distributed

dynamics this is performed with the Rayleigh–

according to a specified distribution.

Van der Pol’s dissipation. Formally, the solu-

Between the pure deterministic dynamics,

tion to the Langevin equation can be written

in which all DOF of the system in considera-

as

tion are explicitly taken into account, leading

to classical dynamical equations, for example β

v(t) = v(0) exp − t

in Hamiltonian form (using ∂x ≡ ∂/∂x), m

Z t

1

q̇ i = ∂pi H, ṗi = −∂qi H, (1) + exp[−(t − τ )β/m] R(τ ) dτ ,

m 0

(where q i , pi are coordinates and momenta, while where the integral on the right–hand side is

H = H(q, p) is the total system energy) – and a stochastic integral and the solution v(t) is a

pure stochastic dynamics (Markov process), there random variable. The stochastic properties of

is so–called hybrid dynamics, particularly Brow- the solution depend significantly on the stochas-

nian dynamics, in which some of DOF are rep- tic properties of the random force R(t). In the

resented only through their stochastic influence Brownian dynamics the random force R(t) is

on others. As an example, suppose a system Gaussian distributed. Then the problem boils

of particles interacts with a viscous medium. down to finding the solution to the Langevin

Instead of specifying a detailed interaction of stochastic differential equation with the supple-

each particle with the particles of the viscous mentary condition (zero and mean variance)

medium, we represent the medium as a stochas-

tic force acting on the particle. The stochastic < R(t) > = 0, < R(t) R(0) > = 2βkB T δ(t),

force reduces the dimensionally of the dynam-

ics. where < . > denotes the mean value, T is tem-

Recall that the Brownian dynamics repre- perature, kB −equipartition (i.e., uniform dis-

sents the phase–space trajectories of a collec- tribution of energy) coefficient, Dirac δ(t)−func-

tion of particles that individually obey Langevin tion.

rate equations in the field of force (i.e., the par- Algorithm for computer simulation of the

ticles interact with each other via some deter- Brownian dynamics (for a single particle) can

ministic force). For a free particle, the Langevin be written as (Heermann, 1990):

6

2. Draw a random number from a Gaussian or corresponding Ito stochastic integral equa-

distribution with mean zero and variance. tion

Z t

3. Integrate the velocity to get v n+1 . i i

x (t) = x (0) + ds Ai [xi (s), s]

0

4. Add the random component to the veloc- Z t

ity. + dW j (s) Bij [xi (s), s],

0

Another approach to taking account the cou-

pling of the system to a heat bath is to subject in which xi (t) is the variable of interest, the

the particles to collisions with virtual particles vector Ai [x(t), t] denotes deterministic drift, the

(Heermann, 1990). Such collisions are imag- matrix Bij [x(t), t] represents continuous stochas-

ined to affect only momenta of the particles, tic diffusion fluctuations, and W j (t) is an N −

hence they affect the kinetic energy and in- variable Wiener process (i.e., generalized Brow-

troduce fluctuations in the total energy. Each nian motion, see Wiener, 1961) and

stochastic collision is assumed to be an instan-

dW j (t) = W j (t + dt) − W j (t).

taneous event affecting only one particle.

The collision–coupling idea is incorporated Now, there are three well–known special cases

into the Hamiltonian model of dynamics (1) by of the Chapman–Kolmogorov equation (see Gar-

adding a stochastic force Ri = Ri (t) to the ṗ diner, 1985):

equation

1. When both Bij [x(t), t] and W (t) are zero,

q̇ i = ∂pi H, ṗi = −∂qi H + Ri (t). i.e., in the case of pure deterministic mo-

tion, it reduces to the Liouville equation

On the other hand, the so–called Ito stochas-

tic integral represents a kind of classical Riemann– ∂t P (x′ , t′ |x′′ , t′′ ) =

Stieltjes integral from linear functional analy- X ∂

Ai [x(t), t] P (x′ , t′ |x′′ , t′′ ) .

sis, which is (in 1D case) for an arbitrary time– − i

∂x

i

function G(t) defined as the mean square limit

Z t 2. When only W (t) is zero, it reduces to the

G(t)dW (t) = Fokker–Planck equation

t0

n

X ∂t P (x′ , t′ |x′′ , t′′ ) =

ms lim { G(ti−1 [W (ti ) − W (ti−1 ]}. X ∂

n→∞

Ai [x(t), t] P (x′ , t′ |x′′ , t′′ ) +

i=1 −

∂x i

i

Now, the general N D Markov process can

be defined by Ito stochastic differential equa- 1X ∂2

Bij [x(t), t] P (x′ , t′ |x′′ , t′′ ) .

2 i

∂x ∂xj

tion (SDE), ij

dxi (t) = Ai [xi (t), t]dt + Bij [xi (t), t] dW j (t), 3. When both Ai [x(t), t] and Bij [x(t), t] are

xi (0) = xi0 , (i, j = 1, . . . , N ) zero, i.e., the state–space consists of inte-

7

gers only, it reduces to the Master equa- 2.2 Quantum Probability Concept

tion of discontinuous jumps

An alternative concept of probability, the so–

′ ′ ′′ ′′

∂t P (x , t |x , t ) = called quantum probability, is based on the fol-

Z lowing physical facts (elaborated in detail in

dx W (x′ |x′′ , t) P (x′ , t′ |x′′ , t′′ ) this section):

Z

− dx W (x′′ |x′ , t) P (x′ , t′ |x′′ , t′′ ). 1. The time–dependent Schrödinger equation

represents a complex–valued generalization

of the real–valued Fokker–Planck equa-

The Markov assumption can now be formu- tion for describing the spatio–temporal

lated in terms of the conditional probabilities probability density function for the sys-

P (xi , ti ): if the times ti increase from right to tem exhibiting continuous–time Markov

left, the conditional probability is determined stochastic process.

entirely by the knowledge of the most recent R

condition. Markov process is generated by a set 2. The Feynman path integral Σ is a gen-

of conditional probabilities whose probability– eralization of the time–dependent Schrö-

density P = P (x′ , t′ |x′′ , t′′ ) evolution obeys the dinger equation, including both continuous–

general Chapman–Kolmogorov integro–differen- time and discrete–time Markov stochastic

tial equation processes.

X ∂

∂t P = − {Ai [x(t), t] P } 3. Both Schrödinger equation and path in-

∂xi tegral give ‘physical description’ of any

i

1X ∂2 system they are modelling in terms of its

+ {Bij [x(t), t] P } physical energy, instead of an abstract

2 ∂xi ∂xj

Z

ij probabilistic description of the Fokker–

Planck equation.

dx W (x′ |x′′ , t) P − W (x′′ |x′ , t) P

+

R

Therefore, the Feynman path integral Σ , as

including deterministic drift, diffusion fluctua- a generalization of the time–dependent Schrö-

tions and discontinuous jumps (given respec- dinger equation, gives a unique physical de-

tively in the first, second and third terms on scription for the general Markov stochastic pro-

the r.h.s.). cess, in terms of the physically based general-

It is this general Chapman–Kolmogorov inte- ized probability density functions, valid both

gro–differential equation, with its conditional for continuous–time and discrete–time Markov

probability density evolution, P = P (x′ , t′ |x′′ , t′′ ),systems.

that we are goingR to model by the Feynman Basic consequence: a different way for cal-

path integral Σ , providing us with the phys- culating probabilities. The difference is rooted

ical insight behind the abstract (conditional) in the fact that sum of squares is different from

probability densities. the square of sums, as is explained in the fol-

lowing text.

8

In Dirac–Feynman quantum formalism, each

possible route from the initial system state A

to the final system state B is called a history.

This history comprises any kind of a route (see

Figure 1), ranging from continuous and smooth

deterministic (mechanical–like) paths to com-

pletely discontinues and random Markov chains

(see e.g., Gardiner, 1985). Each history (la-

belled by index k) is quantitatively described

by iθ k (with i =

√ a complex number, zk = rk e

−1), called the ‘individual transition ampli-

tude’. Its absolute square, |zk |2 , is called the Figure 1: Two ways of physical transition from an

individual transition probability. Now, the to- initial state A to the corresponding final state B.

tal transition amplitude is thePsum of all indi- (a) Classical physics proposes a single determinis-

vidual transition amplitudes, k zk , called the tic trajectory, minimizing the total system’s energy.

sum–over–histo-ries. ThePabsolute square of (b) Quantum physics proposes a family of Markov

this sum–over–histories, | k zk |2 , is the total stochastic histories, namely all possible routes from

transition probability. A to B, both continuous–time and discrete–time

In this way, the overall probability of the Markov chains, each giving an equal contribution

system’s transition from some initial state A to the total transition probability.

to some final state B is given not by adding up

the probabilities for each history–route, but by

‘head–to–tail’ adding up the sequence of am- foam (LSF) as a natural medium for motiva-

plitudes making–up each route first (i.e., per- tional and cognitive psychodynamics. Accord-

forming the sum–over–histories) – to get the to- ing to the LSF–formalism, the classic Lewinian

tal amplitude as a ‘resultant vector’, and then life space can be macroscopically represented as

squaring the total amplitude to get the overall a smooth manifold with steady force–fields and

transition probability. behavioral paths, while at the microscopic level

it is more realistically represented as a collec-

tion of wildly fluctuating force–fields, (loco)mo-

3 The Life Space Foam tion paths and local geometries (and topologies

with holes).

General nonlinear attractor dynamics, both de-

We have used the new LSF concept to de-

terministic and stochastic, as well as possibly

velop modelling framework for motivational dy-

chaotic, developed in the framework of Feyn-

namics (MD) and induced cognitive dynamics

man path integrals, have recently been applied

(CD). Motivation processes both precede and

by Ivancevic and Aidman (2007) to formalize

coincide with every goal–directed action. Usu-

classical Lewinian field–theoretic psychodynam-

ally these motivation processes include the se-

ics (Lewin, 1951, 1997; Gold, 1999), resulting in

quence of the following four feedforward phases

the development of a new concept of life–space

(Ivancevic and Aidman, 2007): (*)

9

1. Intention Formation F, including: deci- where the integral is taken over all n coordi-

sion making, commitment building, etc. nates xj = xj (t) of the LSF, and ∂xj Φi are time

and space partial derivatives of the Φi −variables

2. Action Initiation I, including: handling

over coordinates. The standard least action

conflict of motives, resistance to alterna- principle

tives, etc. δS[Φ] = 0, (3)

3. Maintaining the Action M, including: re- gives, in the form of the so–called Euler–Lagran-

sistance to fatigue, distractions, etc. gian equations, a shortest (loco)motion path,

4. Termination T , including parking and an extreme force–field, and a life–space geom-

avoiding addiction, i.e., staying in con- etry of minimal curvature (and without holes).

trol. In this way, we effectively derive a unique glob-

ally smooth transition map

With each of the phases {F, I, M, T } in (*), we

can associate a transition propagator – an en- F : IN T EN T IONtini - ACT ION

tf in ,

semble of (possibly crossing) feedforward paths (4)

propagating through the ‘wood of obstacles’ performed at a macroscopic (global) time–level

(including topological holes in the LSF, see Fig- from some initial time tini to the final time

ure 3), so that the complete transition is a tf in . In this way, we have obtained macro–

product of propagators (as well as sum over objects in the global LSF: a single path de-

paths). All the phases–propagators are con- scribed by Newtonian–like equation of motion,

trolled by a unique M onitor feedback process. a single force–field described by Maxwellian–

A set of least–action principles is used to like field equations, and a single obstacle–free

model the smoothness of global, macro–level Riemannian geometry (with global topology

LSF paths, fields and geometry, according to without holes).

the following prescription. The action S[Φ], To model the corresponding local, micro–

psycho–physical dimensions of level LSF structures of rapidly fluctuating cog-

nitive dynamics, an adaptive path integral is

Energy × T ime = Effort formulated, defining a multi–phase and multi–

and depending on macroscopic paths, fields path (multi–field and multi–geometry) transi-

and geometries (commonly denoted by an ab- tion amplitude from the state of Intention to

stract field symbol Φi ) is defined as a temporal the state of Action,

integral from the initial time instant tini to the R

final time instant tf in , hAction|Intentioni total := Σ D[wΦ] eiS[Φ] ,

Z tf in (5)

S[Φ] = L[Φ] dt, (2) where the Lebesgue integration is performed

tini over all continuous Φicon = paths + f ields +

geometries, while summation is performed over

with Lagrangian density given by

all discrete processes and regional topologies

Φjdis . The symbolic differential D[wΦ] in the

Z

L[Φ] = dn x L(Φ , ∂ j Φi ),

i x

10

general path integral (5), represents an adap- On the macro–level in LSF we have the

tive path measure, defined as a weighted prod- (loco)motion action principle

uct (with i = 1, ..., n = con + dis)

δS[x] = 0,

YN

D[wΦ] = lim ws dΦis . (6) with the Newtonian–like action S[x] given by

N−→∞

s=1 Z tf in

1

The adaptive path integral (5)–(6) represents S[x] = dt [ gij ẋi ẋj + ϕi (xi )], (7)

tini 2

an ∞−dimensional neural network, with weights

w updating by the general rule (Ivancevic and where ẋi represents motivational (loco)motion

Aidman, 2007): velocity vector with cognitive processing speed.

The first bracket term in (7) represents the ki-

new value(t+1) = old value(t) + innovation(t).

netic energy T ,

The adaptive path integral (5) incorporates 1

the local Bernstein adaptation process (Bern- T = gij ẋi ẋj ,

2

stein, 1967, 1982):

generated by the Riemannian metric tensor gij ,

desired state SW (t + 1) = while the second bracket term, ϕi (xi ), denotes

current state IW (t) + adjustment step ∆W (t) the family of potential force–fields, driving the

(loco)motions xi = xi (t) (the strengths of the

as well as the augmented finite state machine fields ϕi (xi ) depend on their positions xi in

of Brooks’ subsumption architecture (Brooks, LSF. The corresponding Euler–Lagrangian equa-

1986, 1989, 1990), with a networked behavior tion gives the Newtonian–like equation of mo-

function: tion

d

T i − Txi = −ϕixi , (8)

f inal state w(t + 1) = dt ẋ

current state w(t) + (subscripts denote the partial derivatives), which

adjustment behavior f (∆w(t)). can be put into the standard Lagrangian form

d

We remark here that the traditional neural L i = Lx i , with L = T − ϕi (xi ).

networks approaches are known for their classes dt ẋ

of functions they can represent. This limitation Now, according to Lewin, the life space also

has been attributed to their low-dimensionality has a sophisticated topological structure. As

(the largest neural networks are limited to the a Riemannian smooth n−manifold, the LSF–

order of 105 dimensions, see Izhikevich and Edel- manifold Σ gives rise to its fundamental n−

man, 2008). The proposed path integral ap- groupoid, or n−category Πn (Σ) (see Ivancevic,

proach represents a new family of function-re- 2006b, 2007a). In Πn (Σ), 0–cells are points in

presentation methods, which potentially offers Σ; 1–cells are paths in Σ (i.e., parameterized

a basis for a fundamentally more expansive so- smooth maps f : [0, 1] → Σ); 2–cells are smooth

lution. homotopies (denoted by ≃) of paths relative

11

to endpoints (i.e., parameterized smooth maps

h : [0, 1]×[0, 1] → Σ); 3–cells are smooth homo-

topies of homotopies of paths in Σ (i.e., param-

eterized smooth maps j : [0, 1] × [0, 1] × [0, 1] →

Σ). Categorical composition is defined by past-

ing paths and homotopies. In this way, the fol-

lowing recursive homotopy dynamics emerges

on the LSF–manifold Σ (**):

12

0 − cell : x0 • x0 ∈ M ; in the higher cells below: t, s ∈ [0, 1];

f -

1 − cell : x0 • • x1 f : x0 ≃ x1 ∈ M,

f : [0, 1] → M, f : x0 7→ x1 , x1 = f (x0 ), f (0) = x0 , f (1) = x1 ;

e.g., linear path: f (t) = (1 − t) x0 + t x1 ; or

Euler–Lagrangian f − dynamics with endpoint conditions (x0 , x1 ) :

d

f i = fxi , with x(0) = x0 , x(1) = x1 , (i = 1, ..., n);

dt ẋ

f

R

2 − cell : x0 • h • x1 h : f ≃ g ∈ M,

∨

g

h : [0, 1] × [0, 1] → M, h : f 7→ g, g = h(f (x0 )),

h(x0 , 0) = f (x0 ), h(x0 , 1) = g(x0 ), h(0, t) = x0 , h(1, t) = x1

e.g., linear homotopy: h(x0 , t) = (1 − t) f (x0 ) + t g(x0 ); or

homotopy between two Euler–Lagrangian (f, g) − dynamics

with the same endpoint conditions (x0 , x1 ) :

d d

fẋi = fxi , and g i = gxi with x(0) = x0 , x(1) = x1 ;

dt dt ẋ

f

j R

3 − cell : x0 • h > i • x1 j : h ≃ i ∈ M,

y x

g

j : [0, 1] × [0, 1] × [0, 1] → M, j : h 7→ i, i = j(h(f (x0 )))

j(x0 , t, 0) = h(f (x0 )), j(x0 , t, 1) = i(f (x0 )),

j(x0 , 0, s) = f (x0 ), j(x0 , 1, s) = g(x0 ),

j(0, t, s) = x0 , j(1, t, s) = x1

e.g., linear composite homotopy: j(x0 , t, s) = (1 − t) h(f (x0 )) + t i(f (x0 ));

or, homotopy between two homotopies between above two Euler-

Lagrangian (f, g) − dynamics with the same endpoint conditions (x0 , x1 ).

13

On the micro–LSF level, instead of a single information flow during the transition Intention

path defined by the Newtonian–like equation −→ Action. In the connectionist language, (9)

of motion (8), we have an ensemble of fluctuat- represents activation dynamics, to which our

ing and crossing paths with weighted probabil- M onitor process gives a kind of backpropaga-

ities (of the unit total sum). This ensemble of tion feedback, a common type of supervised

micro–paths is defined by the simplest instance learning1

of our adaptive path integral (5), similar to the

Feynman’s original sum over histories, ws (t + 1) = ws (t) − η∇J(t), (12)

R

hAction|Intentionipaths = Σ D[wx] eiS[x] , (9)

or the learning rate, and ∇J(n) denotes the

where D[wx] is a functional measure on the gradient of the ‘performance hyper–surface’ at

space of all weighted paths, and the exponential the t-th iteration.

depends on the action S[x] given by (7). This Now, the basic question about our local de-

procedure can be redefined in a mathematically cision making process, occurring under uncer-

cleaner way if we Wick–rotate the time variable tainty at the intention formation faze F, is:

t to imaginary values, t 7→ τ = it, thereby mak- Which alternative to choose? In our path–

ing all integrals real: integral language this reads: Which path (al-

ternative) should be given the highest proba-

Σ D[wx] eiS[x] W ick- Σ D[wx] e−S[x] . (10) bility weight w? This problem can be either

R R

iteratively solved by the learning process (12),

Discretization of (10) gives the standard thermo- controlled by the M ON IT OR feedback, which

dynamic–like partition function we term algorithmic approach, or by the lo-

cal decision making process under uncertainty,

−wj E j /T

X

Z= e , (11) which we term heuristic approach (Ivancevic

j and Aidman, 2007). This qualitative analy-

sis is based on the micro–level interpretation

where E j is the motion energy eigenvalue (re- of the Newtonian–like action S[x], given by (7)

flecting each possible motivational energetic state),

and figuring both processing speed ẋ and LTM

T is the temperature–like environmental con- (i.e., the force–field ϕ(x), see next subsection).

trol parameter, and the sum runs over all mo- Here we consider three different cases:

tion energy eigenstates (labelled by the index

j). From (11), we can further calculate all 1. If the potential ϕ(x) is not very depen-

thermodynamic–like and statistical properties dent upon position x(t), then the more di-

of MD and CD, as for example, transition en- rect paths contribute the most, as longer

tropy, S = kB ln Z, etc. 1

Note that we could also use a reward–based, rein-

forcement learning rule (Suttton and Barto, 1998), in

3.1 Noisy Decision Making in the LSF which system learns its optimal policy:

From CD–perspective, our adaptive path inte-

gral (9) calculates all (alternative) pathways of

14

paths, with higher mean square velocities reach the brain at any time. In this theory,

[ẋ(t)]2 make the exponent more negative the importance of conscious, directed attention

(after Wick rotation (10)). is minimized. The type of attention involving

low level filtering corresponds to the concept of

2. On the other hand, suppose that ϕ(x) early selection.

does indeed depend on position x. For Although we termed this ‘heuristic approach’

simplicity, let the potential increase for in the sense that we can instantly feel both the

the larger values of x. Then a direct path processing speed ẋ and the LTM field ϕ(x) in-

does not necessarily give the largest con- volved, there is clearly a psycho–physical rule

tribution to the overall transition prob- in the background, namely the averaging min-

ability, because the integrated value of imum relation (13).

the potential is higher than over another From the decision making point of view, all

paths. possible paths (alternatives) represent the con-

3. Finally, consider a path that deviates wide- sequences of decision making. They are, by

ly from the direct path. Then ϕ(x) de- default, short–term consequences, as they are

creases over that path, but at the same modelled in the micro–time–level. However,

time the velocity ẋ increases. In this case, the path integral formalism allows calculation

we expect that the increased velocity ẋ of the long–term consequences, just by extend-

would more than compensate for the de- ing the integration time, tf in − → ∞. Besides,

creased potential over the path. this averaging decision mechanics – choosing

the optimal path – actually performs the ‘aver-

Therefore, the most important path (i.e., the aging lift’ in the LSF: from the micro–level to

path with the highest weight w) would be the the macro–level.

one for which any smaller integrated value of For example, one of the simplest types of

the surrounding field potential ϕ(x) is more performance–degrading disturbances in the LSF

than compensated for by an increase in kinetic– is what we term motivational fatigue – a moti-

like energy m 2

2 ẋ . In principle, this is neither vational drag factor that slows the actors’ prog-

the most direct path, nor the longest path, but ress towards their goal. There are two funda-

rather a middle way between the two. For- mentally different sources of this motivational

mally, it is the path along which the average drag, both leading to apparently the same re-

Lagrangian is minimal, duction in performance: (a) tiredness / exhaus-

m tion and (b) satiation (e.g., boredom). Both in-

< ẋ2 + ϕ(x) > −→ min, (13) volve the same underlying mechanism (the rais-

2

ing valence of the alternatives to continuing the

i.e., the path that requires minimal memory (both action) but the alternatives will differ consider-

LTM and WM) and processing speed. This me- ably, depending on the properties of the task,

chanical result is consistent with the ‘cognitive from self–preservation / recuperation in the ex-

filter theory’ of selective attention (Broadbent, haustion case through to competing goals in the

1958), which postulates a low level filter that satiation case.

allows only a limited number of percepts to The spatial representation of this motiva-

15

tional drag is relatively simple: uni–dimensional nomenon of phase transitions (Caiani et al, 1997;

LSF–coordinates may be sufficient for most pur- Pettini, 2007). The geometrical formulation

poses, which makes it attractive for the initial of the dynamics of conservative systems (see

validation of our predictive model. Similarly Ivancevic, 2006b, 2008a) was first used by Krylov

uncomplicated spatial representations can be (1979) in his studies on the dynamical founda-

achieved for what we term motivational boost tions of statistical mechanics and subsequently

derived from the proximity to the goal (includ- became a standard tool to study abstract sys-

ing the well–known phenomenon of ‘the home tems in ergodic theory.

stretch’): the closer the goal (e.g., a finishing The simplest, mechanical–like LSF–action

line) is perceived to be, the stronger its ‘pulling in the individual’s LSF–manifold Σ has a Rie-

power’ (Lewin 1951, 1997). Combinations of mannian locomotion form (summation conven-

motivational drag and motivational boost ef- tion is always assumed)

fects may be of particular interest in a range of

tf in

1

Z

applications. These combinations can be mod- S[q] = [aij q̇ i q̇ j − V (q)] dt, (14)

elled within relatively simple uni–dimensional 2 tini

LSF–coordinate systems.

where aij is the ‘material’ metric tensor that

generates the total ‘kinetic energy’ of cognitive

4 Geometric Chaos and Topo- (loco)motions defined by their configuration co-

logical Phase Transitions ordinates q i and velocities q̇ i , with the motiva-

tional potential energy V (q) and the standard

In this section we extend the LSF–formalism to Hamiltonian

incorporate geometrical chaos (Ivancevic et al, N

2008; Ivancevic, 2006c, 2008a) and associated

X 1 2

H(p, q) = p + V (q), (15)

topological phase transitions. 2 i

i=1

It is well–known that on the basis of the er-

godic hypothesis, statistical mechanics describes where pi are the canonical (loco)motion mo-

the physics of many–degrees of freedom sys- menta.

tems by replacing time averages of the relevant Dynamics of N DOF mechanical–like sys-

observables with ensemble averages. Therefore, tems with action (14) and Hamiltonian (15) are

instead of using statistical ensembles, we can commonly given by the set of geodesic equations

investigate the Hamiltonian (microscopic) dy- (Ivancevic, 2006b, 2007a)

namics of a system undergoing a phase tran-

sition. The reason for tackling dynamics is d2 q i j

i dq dq

k

+ Γ jk = 0, (16)

twofold. First, there are observables, like Lya- ds2 ds ds

punov exponents, that are intrinsically dynam- where Γi are the Christoffel symbols of the

jk

ical. Second, the geometrization of Hamilto- affine Levi–Civita connection of the Rieman-

nian dynamics in terms of Riemannian geom- nian LSF–manifold Σ.

etry provides new observables and, in general, Alternatively, a description of the extrema

an interesting framework to investigate the phe- of the Hamilton’s action (14) can be obtained

16

using the Eisenhart metric (see Eisenhart, 1929) dynamical systems given by the Riemannian

on an enlarged LSF space-time manifold (given action (14) and Hamiltonian (15), using the for-

by {q 0 ≡ t, q 1 , . . . , q N } plus one real coordinate mula (Casetti et al, 2000)

q N +1 ), whose arc–length is

λ1 = lim 1/2t log(ΣN 2

i=1 [Ji (t) (20)

2 0 2 i j 0 N +1 t→∞

ds = −2V (q)(dq ) + aij dq dq + 2dq dq .

(17) + Ji2 (t)]/ΣN 2 2

i=1 [Ji (0) + Ji (0)]).

The manifold has a Lorentzian structure (Pet-

Lyapunov exponents measure the strength of

tini, 2007) and the dynamical trajectories are

dynamical chaos.

those geode-sics satisfying the condition ds2 =

2 Now, to relate these results to topological

Cdt , where C is a positive constant. In this

phase transitions within the LSF–manifold Σ,

geometrical framework, the instability of the

recall that any two high–dimensional manifolds

trajectories is the instability of the geodesics,

Σv and Σv′ have the same topology if they can

and it is completely determined by the curva-

be continuously and differentiably deformed into

ture properties of the LSF–manifold Σ accord-

one another, that is if they are diffeomorphic.

ing to the Jacobi equation of geodesic deviation

Thus by topology change the ‘loss of diffeomor-

(see Ivancevic, 2006b, 2007a)

phicity’ is meant (Pettini, 2007). In this re-

D2 J i dq j dq m spect, the so–called topological theorem (Fran-

2

+ Ri jkm Jk = 0, (18) zosi and Pettinni, 2004) says that non–analyti-

ds ds ds

city is the ‘shadow’ of a more fundamental phe-

whose solution J, usually called Jacobi vari-

nomenon occurring in the system’s configura-

ation field, locally measures the distance be-

tion manifold (in our case the LSF–manifold):

tween nearby geodesics; D/ds stands for the

i a topology change within the family of equipo-

covariant derivative along a geodesic and R jkm

tential hypersurfaces

are the components of the Riemann curvature

tensor of the LSF–manifold Σ. Σv = {(q 1 , . . . , q N ) ∈ RN | V (q 1 , . . . , q N ) = v},

Using the Eisenhart metric (17), the rele-

vant part of the Jacobi equation (18) is given where V and q i are the microscopic interaction

by the tangent dynamics equation (Casetti et potential and coordinates respectively. This

al, 1996; Caiani et al, 1997) topological approach to PTs stems from the nu-

merical study of the dynamical counterpart of

d2 J i i k

+ R 0k0 J = 0, (i = 1, . . . , N ), (19) phase transitions, and precisely from the obser-

dt2

vation of discontinuous or cuspy patterns dis-

where the only non-vanishing components of played by the largest Lyapunov exponent λ1 at

the curvature tensor of the LSF–manifold Σ the transition energy (Casetti et al, 2000). Lya-

are punov exponents cannot be measured in labo-

ratory experiments, at variance with thermo-

Ri 0k0 = ∂ 2 V /∂q i ∂q j .

dynamic observables, thus, being genuine dy-

The tangent dynamics equation (19) is com- namical observables they are only be estimated

monly used to define Lyapunov exponents in in numerical simulations of the microscopic dy-

17

namics. If there are critical points of V in con- let v̄ = v/N be the potential energy per degree

figuration space, that is points qc = [q 1 , . . . , q N ] of freedom. If there exists N0 , and if for any

such that ∇V (q)|q=qc = 0, according to the pair of values v̄ and v̄ ′ belonging to a given in-

Morse Lemma (see e.g., Hirsch, 1976), in the terval Iv̄ = [v̄0 , v̄1 ] and for any N > N0 then

neighborhood of any critical point qc there al- the sequence of the Helmoltz free energies

ways exists a coordinate system {FN (β)}N ∈N – where β = 1/T (T is the tem-

q̃(t) = [q̃ 1 (t), .., q̃ N (t)] for which perature) and β ∈ Iβ = (β(v̄0 ), β(v̄1 )) – is

uniformly convergent at least in C 2 (Iβ ) [the

2

V (q̃) = V (qc ) − q̃12 − · · · − q̃k2 + q̃k+1 2

+ · · · + q̃N , space of twice differentiable functions in the in-

(21) terval Iβ ], so that limN →∞ FN ∈ C 2 (Iβ ) and

where k is the index of the critical point, i.e., neither first nor second order phase transitions

the number of negative eigenvalues of the Hes- can occur in the (inverse) temperature interval

sian of the potential energy V . In the neigh- (β(v̄0 ), β(v̄1 )), where the inverse temperature is

borhood of a critical point of the LSF–manifold defined as (Pettini, 2007)

Σ, (21) yields

(−)

β(v̄) = ∂SN (v̄)/∂v̄, while

2 i j

∂ V /∂q ∂q = ±δij , Z

(−)

SN (v̄) = N −1 log dN q

which gives k unstable directions which con- V (q)≤v̄N

tribute to the exponential growth of the norm is one of the possible definitions of the micro-

of the tangent vector J (Casetti et all, 2000). canonical configurational entropy. The inten-

This means that the strength of dynami- sive variable v̄ has been introduced to ease the

cal chaos within the individual’s LSF–manifold comparison between quantities computed at dif-

Σ, measured by the largest Lyapunov exponent ferent N -values.

λ1 given by (20), is affected by the existence of This theorem means that a topology change

critical points qc of the potential energy V (q). of the {Σv }v∈R at some vc is a necessary con-

However, as V (q) is bounded below, it is a good dition for a phase transition to take place at

Morse function, with no vanishing eigenvalues the corresponding energy value. The topol-

of its Hessian matrix. According to Morse the- ogy changes implied here are those described

ory, the existence of critical points of V is asso- within the framework of Morse theory through

ciated with topology changes of the hypersur- ‘attachment of handles’ to the LSF–manifold Σ

faces {Σv }v∈R . (Hirsch, 1976).

More precisely, let VN (q1 , . . . , qN ) : RN → In the LSF path–integral language, we can

R, be a smooth, bounded from below, finite- say that suitable topology changes of equipo-

range and confining potential2 . Denote by Σv = tential submanifolds of the individual’s LSF–

V −1 (v), v ∈ R, its level sets, or equipotential manifold Σ can entail thermodynamic–like phase

hypersurfaces, in the LSF–manifold Σ. Then transitions, according to the general formula:

2

These requirements for V are fulfilled by standard

interatomic and intermolecular interaction potentials,

hphase out | phase ini :=

R

as well as by classical spin potentials. Σ D[wΦ] eiS[Φ] .

topology−change

18

The statistical behavior of the LSF–(loco)moti- integrals,

on system (14) with the standard Hamiltonian Z

(15) is encompassed, in the canonical ensemble, dσ /k∇V k,

by its partition function, given by the phase– Σv

space path integral (Ivancevic, 2007a, 2008a) defined on the {Σv }v∈R . Once the microscopic

R Z t′ interaction potential V (q) is given, the configu-

ZN = Σ D[p]D[q] exp{i [pq̇−H(p, q)] dτ },ration space of the system is automatically fo-

top−ch

t

(22) liated into the family {Σv }v∈R of these equipo-

where we have used the shorthand notation tential hypersurfaces. Now, from standard sta-

tistical mechanical arguments we know that, at

dq(τ )dp(τ )

R Z Y

Σ D[p]D[q] ≡ . any given value of the inverse temperature β,

top−ch

τ

2π the larger the number N , the closer to Σv ≡

Σuβ are the microstates that significantly con-

The phase–space path integral (22) can be cal- tribute to the averages, computed through Z (β),

N

culated as the partition function (Franzosi et of thermodynamic observables. The hypersur-

al, 2000) face Σuβ is the one associated with

Z YN Z Y

ZN (β) = dpi dq i e−βH(p,q) C −1

uβ = (ZN ) dq i V (q)e−βV (q) ,

i=1

NZ Y N

π 2 the average potential energy computed at a given

= dq i e−βV (q)

β β. Thus, at any β, if N is very large the effec-

i=1

N Z ∞ tive support of the canonical measure shrinks

π 2 dσ

Z

very close to a single Σv = Σuβ . Hence, the

= dv e−βv , , (23)

β 0 Σv k∇V k basic origin of a phase transition lies in a suit-

able topology change of the {Σv }, occurring

where the last term is written using the so–

at some vc (Franzosi et al, 2000). This topol-

called co–area formula (Federer, 1969), and v

ogy change induces the singular behavior of the

labels the equipotential hypersurfaces Σv of the

thermodynamic observables at a phase transi-

LSF–manifold Σ,

tion. It is conjectured that the counterpart of a

Σv = {(q 1 , . . . , q N ) ∈ RN |V (q 1 , . . . , q N ) = v}. phase transition is a breaking of diffeomorphic-

ity among the surfaces Σv , it is appropriate to

Equation (23) shows that the relevant statis- choose a diffeomorphism invariant to probe if

tical information is contained in the canonical and how the topology of the Σv changes as a

configurational partition function function of v. Fortunately, such a topological

Z Y invariant exists, the Euler characteristic of the

C

ZN = dq i V (q)e−βV (q) . LSF–manifold Σ, defined by (Ivancevic, 2007a)

N

Note that ZN C is decomposed, in the last term X

χ(Σ) = (−1)k bk (Σ), (24)

of (23), into an infinite summation of geometric k=0

19

where the Betti numbers bk (Σ) are diffeomor- as a macroscopic system which is a quantum

phism invariants.3 This homological formula system not in the trivial sense that it is made,

can be simplified by the use of the Gauss–Bonnet–like all existing matter, by quantum compo-

Hopf theorem, that relates χ(Σ) with the total nents such as atoms and molecules, but in the

Gauss–Kronecker curvature KG of the LSF– sense that some of its macroscopic properties

manifold Σ can best be described with recourse to quan-

Z tum dynamics (see Freeman and Vitiello, 2006

χ(Σ) = KG dσ, (25) and references therein).

Σ Phase transitions can also be associated with

where autonomous robot competence levels, as infor-

mal specifications of desired classes of behav-

dσ = det(a)dx1 dx2 · · · dxn

p

iors for robots over all environments they will

encounter, as described by Brooks’ subsump-

is the invariant volume measure of the LSF– tion architecture approach. The distributed

manifold Σ and a is the determinant of the network of augmented finite–state machines can

LSF metric tensor aij (Ivancevic, 2008a). exist in different phases or modalities of their

The domain of validity of the ‘quantum’ is state–space variables, which determine the sys-

not restricted to the microscopic world (Umezawa,tems intrinsic behavior. The phase transition

1993). There are macroscopic features of clas- represented by this approach is triggered by ei-

sically behaving systems, which cannot be ex- ther internal (a set–point) or external (a com-

plained without recourse to the quantum dy- mand) control stimuli, such as a command to

namics. This field theoretic model leads to the transition from a sleep mode to awake mode,

view of the phase transition as a condensation or walking to running.

that is comparable to the formation of fog and

rain drops from water vapor, and that might

serve to model both the gamma and beta phase 5 Joint Action of Several

transitions. According to such a model, the Agents

production of activity with long-range correla-

tion in the brain takes place through the mech- In this section we propose an LSF–based model

anism of spontaneous breakdown of symmetry of the joint action of two or more actors, where

(SBS), which has for decades been shown to de- actors can be both humans and robots. This

scribe long-range correlation in condensed mat- joint action takes place in the joint LSF mani-

ter physics. The adoption of such a field theo- fold ΣJ , composed of individual LSF manifolds

retic approach enables modelling of the whole Σα , Σβ , .... It has a sophisticated geometrical

cerebral hemisphere and its hierarchy of com- and dynamical structure as follows.

ponents down to the atomic level as a fully in- To model the dynamics of the two–actor co–

tegrated macroscopic quantum system, namely action, we propose to associate each of the ac-

3

tors with a set of their own time dependent tra-

The Betti numbers bk are the dimensions of the de

Rham’s cohomology vector spaces H k (Σ; R) (therefore

jectories, which constitutes an n−dimensional

the bk are integers). Riemannian LSF–manifold, Σα = {xi (ti )} and

20

Σβ = {y j (tj )}, respectively. Their associated agents, vanishes. Note that the cognitive (loco)

tangent bundles contain their individual nD motions of the two agents αi [xi (ti )] and β j [y j (tj )],

(loco)motion velocities, T Σα = {ẋi (ti ) = dxi /dti }generally occur at different times ti and tj un-

and T Σβ = {ẏ j (tj ) = dy j /dtj }. Further, fol- less ti = tj , when cognitive synchronization oc-

lowing the general LSF–formalism, outlined in curs.

the introduction, we use the modelling machin- The second term in (26) represents kinetic

ery consisting of: (i) Adaptive joint action at energy of the physical interaction. Namely, when

the top–master level, describing the externally– the cognitive synchronization in the first term

appearing deterministic, continuous and smooth takes place, the second term of physical kinetic

dynamics, and (ii) Corresponding adaptive path energy is activated in the common manifold,

integral (30) at the bottom–slave level, describ- which is one of the agents’ Life Spaces, say

ing a wildly fluctuating dynamics including both Σα = {xi (ti )}.

continuous trajectories and Markov chains. This Conversely, if we have a need to represent

lower–level joint dynamics can be further dis- coaction of three actors, say αi , β j and γ k (e.g.,

cretized into a partition function of the corre- αi in charge of acceleration, β j – deceleration

sponding statistical dynamics. and γ k − steering), we can associate each of

The smooth joint action with two terms, them with an nD Riemannian Life–Space man-

representing cognitive/motivational potential en- ifold, Σα = {xi (ti )}, Σβ = {y j (tj )}, and Σγ =

ergy and physical kinetic energy, is formally {z k (tk )}, respectively, with the corresponding

given by: tangent bundles containing their individual (loco)

motion velocities, T Σα = {ẋi (ti ) = dxi /dti },

A[x, y; ti , tj ] =

T Σβ = {ẏ j (tj ) = dy j /dtj } and T Σγ = {ż k (tk )

1

Z Z

2 i j

αi β j δ(Iij ) ẋ (ti ) ẏ (tj ) dti dtj = dz k /dtk }. Then, instead of (26) we have

2 ti tj

1

Z A[ti , tj , tk ; t] =

+ gij ẋi (t)ẋj (t) dt, (26) 1

Z Z Z

2 t αi (ti )β j (tj ) γ k (tk ) ×

2

i j

2 2 ti tj tk

with Iij = x (ti ) − y (tj ) ,

2

where IN ≤ ti , tj , t ≤ OU T. δ(Iijk ) ẋi (ti ) ẏ j (tj ) ż k (tk ) dti dtj dtk

1

Z

The first term in (26) represents potential en- + W M (t, q, q̇) q̇ r q̇ s dt, (27)

2 t rs

ergy of the cognitive/motivational interaction

between the two agents αi and β j .4 It is a dou-

ble integral over a delta function of the square where IN ≤ ti , tj , tk , t ≤ OU T, with

of interval I 2 between two points on the paths Iijk = [x (ti ) − y (tj )] + [y (tj ) − z (tk )]2

2 i j 2 j k

only when this interval, representing the mo-

tivational cognitive distance between the two Due to an intrinsic chaotic coupling, the

4

Although, formally, this term contains cognitive ve-

three–actor (or, n−actor, n > 3) joint action

locities, it still represents ‘potential energy’ from the (27) has a considerably more complicated geo-

physical point of view. metrical structure then the bilateral co–action

21

(26).5 It actually happens in the common 3nD integral (30) represents an adaptive path mea-

Finsler manifold ΣJ = Σα ∪ Σβ ∪ Σγ , param- sure, defined as a weighted product

eterized by the local joint coordinates depen- N

dent on the common time t. That is, ΣJ = D[w, x, y] = lim

Y

s

wij dxi dy j , (31)

{q r (t), r = 1, ..., 3n}. Geometry of the joint N →∞

s=1

manifold ΣJ is defined by the Finsler metric (i, j = 1, ..., n).

function ds = F (q r , dq r ), defined by

Similarly, in case of the triple joint action,

F 2 (q, q̇) = grs (q, q̇)q̇ r q̇ s , (28) the adaptive path integral reads,

R

and the Finsler tensor Crst (q, q̇), defined by hOU T |IN i := Σ D[w; x, y, z; q] eiA[ti ,tj ,tk ;t] ,

(Ivancevic, 2007a) (32)

1 ∂ 3 F 2 (q, q̇) 1 ∂grs with the adaptive path measure defined by

Crst (q, q̇) = r s t

= . (29)

4 ∂ q̇ ∂ q̇ ∂ q̇ 2 ∂ q̇ r ∂ q̇ s N

Y

S

D[w; x, y, z; q] = lim wijkr dxi dy j dz k dq r ,

From the Finsler definitions (28)–(29), it fol- N →∞

S=1

lows that the partial interaction manifolds, Σα ∪ (i, j, k = 1, ..., n; r = 1, ..., 3n). (33)

Σβ , Σβ ∪ Σy and Σα ∪ Σy , have Riemannian

structures with the corresponding interaction The proposed path integral approach repre-

kinetic energies, sents a new family of more expansive function-

representation methods, which is now capable

1 i j 1 i k of representing input/output behavior of more

Tαβ = gij ẋ ẏ , Tαγ = gik ẋ ż ,

2 2 than one actor. However, as we add the sec-

1 j k

Tβγ = gjk ẏ ż . ond and subsequent actors to the model, the

2 requirements for the rigorous geometrical rep-

At the slave LSF–level, the adaptive path resentations of their respective LSFs become

integral, representing an ∞−dimensional neu- nontrivial. For a single actor or a two–actor

ral network, corresponding to the adaptive bi- co–action the Riemannian geometry was suffi-

lateral joint action (26), reads cient, but it becomes insufficient for modelling

R the n–actor (with n ≥ 3) joint action, due to

iA[x,y;ti ,tj ]

Σ

hOU T |IN i := D[w, x, y] e , (30) an intrinsic chaotic coupling between the in-

dividual actors’ LSFs. To model an n–actor

where the Lebesgue integration is performed joint LSF, we have to use the Finsler geome-

over all continuous paths xi = xi (ti ) and y j = try, which is a generalization of the Riemannian

y j (tj ), while summation is performed over all one. This progression may seem trivial, both

associated discrete Markov fluctuations and from standard psychological point of view, and

jumps. The symbolic differential in the path from computational point of view, but it is not

5

Recall that the necessary condition for chaos in con-

trivial from the geometrical perspective.

tinuous temporal or spatio-temporal systems is to have Our extended LSF formalism is closely re-

three variables with nonlinear couplings between them. lated to the Haken-Kelso-Bunz (HKB) model

22

of self-organization in the human motor sys- are coupling parameters (from which the criti-

tem (Haken et al, 1985; Kelso, 1995), including: cal frequency where the phase transition occurs

multi-stability, phase transitions and hystere- can be calculated).

sis effects, presenting a contrary view to the From a quantum perspective, closely related

purely feedback driven neural systems. HKB to the LSF model are the recent developments

uses the concepts of synergetics (order param- of Hong and Newell (2008a, 2008b) in motor

eters, control parameters, instability, etc) and control that deal with feedback information and

the mathematical tools of nonlinearly coupled environmental uncertainty. The probabilistic

(nonlinear) dynamical systems to account for nature of human action can be characterized

self-organized behavior both at the coopera- by entropies at the level of the organism, task,

tive, coordinative level and at the level of the and environment. Systematic changes in motor

individual coordinating elements. The HKB adaptation are characterized as task–organism

model stands as a building block upon which and environment–organism tradeoffs in entropy.

numerous extensions and elaborations have been Such compensatory adaptations lead to a view

constructed. In particular, it has been possible of goal–directed motor control as the product

to derive it from a realistic model of the cortical of an underlying conservation of entropy across

sheet in which neural areas undergo a reorga- the task-organism-environment system. The

nization that is mediated by intra- and inter- conservation of entropy supports the view that

cortical connections. Also, the HKB model de- context dependent adaptations in human goal–

scribes phase transitions (‘switches’) in coordi- directed action are guided fundamentally by

nated human movement as follows: (i) when natural law and provides a novel means of ex-

the agent begins in the anti-phase mode and amining human motor behavior. This is funda-

speed of movement is increased, a spontaneous mentally related to the Heisenberg uncertainty

switch to symmetrical, in-phase movement oc- principle and further support the argument for

curs; (ii) this transition happens swiftly at a the primacy of a probabilistic approach toward

certain critical frequency; (iii) after the switch the study of bio-psychological systems.

has occurred and the movement rate is now de-

creased the subject remains in the symmetrical

mode, i.e. she does not switch back; and (iv) no

such transitions occur if the subject begins with

symmetrical, in-phase movements. The HKB

dynamics of the order parameter relative phase

as is given by a nonlinear first-order ODE:

φ̇ = (α + 2βr 2 ) sin φ − βr 2 sin 2φ,

where φ is the phase relation (that character-

izes the observed patterns of behavior, changes

abruptly at the transition and is only weakly

dependent on parameters outside the phase tran-

sition), r is the oscillator amplitude, while α, β

23

Figure 2: Diagram of the R

life space foam: classical representation of Lewinian life space, with

an adaptive path integral Σ (denoting integration over continuous paths and summation over

discrete Markov jumps) acting inside it and generating microscopic fluctuation dynamics.

24

Figure 3: Transition–propagator corresponding to each of the motivational phases {F, I, M, T },

consisting of an ensemble of feedforward paths propagating through the ‘wood of obstacles’.

The paths affected by driving and restraining force–fields, as well as by the local LSF–geometry.

Transition goes from Intention, occurring at a sample time instant t0 , to Action, occurring at

some later time t1 . Each propagator is controlled by its own M onitor feedback.

25

Figure 4: Feynman action for modelling human joint action, including potential energy (motivational

cognition) in two timescales, physical energy in a single timescale (after synchronization has already

occurred), and the distance between two agents in the motivational cognition space.

26

6 Conclusion 1996.

Bernstein NA. Some emergent problems of

General stochastic and quantum dynamics, de- the regulation of motor acts. In: Whiting H

veloped in a framework of Feynman path inte- (ed) Human Motor Actions: Bernstein Reassessed

grals, have recently been applied by Ivancevic 343–358. North Holland, Amsterdam 1982.

and Aidman (2007) to Lewinian field–theoretic Bernstein NA. The Coordination and Regu-

psychodynamics, resulting in the development lation of Movements. Pergamon London 1967.

of a new concept of Life–Space Foam (LSF) as Broadbent DE. Perception and communi-

a natural medium for motivational (MD) and cations. Pergamon Press, London 1958.

cognitive (CD) psychodynamics. According to Brooks RA. A Robust Layered Control Sys-

the LSF–formalism, the classic Lewinian life tem for a Mobile Robot. IEEE Trans Rob Aut

space can be macroscopically represented as a 1986;(2)1:14–23.

smooth manifold with steady force–fields and Brooks RA. A robot that walks: Emergent

behavioral paths, while at the microscopic level behavior form a carefully evolved network. Neu

it is more realistically represented as a collec- Comp 1989;12:253–262.

tion of wildly fluctuating force–fields, (loco)mo- Brooks RA. Elephants Don’t Play Chess.

tion paths and local geometries (and topolo- Rob Aut Sys 1990;6:3–15.

gies with holes). This paper extends the LSF– Busemeyer JR, Wang Z and Townsend JT.

model to incorporate the notion of phase tran- Quantum dynamics of human decision-making.

sitions and complements it with embedded geo- J Math Psych 2006;50:220–241.

metrical chaos. As a result, the extended LSF– Caiani L, Casetti L, Clementi C and Pet-

model is able to rigorously represent co–action tini M. Geometry of Dynamics Lyapunov Ex-

by two or more human–like agents in the com- ponents and Phase Transitions. Phys Rev Lett

mon LSF–manifold. The extended LSF–model 1997;79:4361-4364.

is also related to the HKB–model of self-organi- Casetti L, Clementi C and Pettini M. Rie-

zation in the human motor system, presenting mannian theory of Hamiltonian chaos and Lya-

a contrary view to the purely feedback driven punov exponents. Phys Rev E 1996;54:5969.

neural systems, as well as Hong–Newell entropy– Casetti L, Pettini M and Cohen EGD. Geo-

approach to adaptation in human goal–directed metric Approach to Hamiltonian Dynamics and

motor control. Statistical Mechanics. Phys Rep 2000;337:237-

341.

Droll, JA, Hayhoe, MM, Triesch, J and Sul-

References livan, BT. Task demands control acquisition

and storage of visual information. J Exp Psych:

Amari S. Dynamics of pattern formation in Hum Perc Perf 2005;31:14161438.

lateral-inhibition type neural fields. Biol Cy- Eisenhart LP. Dynamical trajectories and

bern 1977;27:77–87. geodesics. Math Ann 1929;30:591–606.

Bernstein NA, Latash ML and Turvey MT Federer H. Geometric Measure Theory.

(Eds). Dexterity and its development. Hills- Springer New York 1969.

dale NJ England: Lawrence Erlbaum Associates

27

Fogassi L, Ferrari PF, Gesierich B, Rozzi S, Heermann DW. Computer Simulation Meth-

Chersi F and Rizzolatti G. Parietal lobe: From ods in Theoretical Physics (2nd ed). Springer,

action organization to intention understanding. Berlin 1990.

Science 2005;29:662–667. Hirsch MW. Differential Topology. Springer,

Franzosi R and Pettini M. Theorem on the New York 1976.

origin of Phase Transitions. Phys Rev Lett Hong SL, Newell KM. Entropy compensa-

2004;92:060601. tion in human motor adaptation. Chaos

Franzosi R, Pettini M, Spinelli L. Topol- 2008b;18(1):013108.

ogy and phase transitions: a paradigmatic evi- Hong SL, Newell KM. Entropy conserva-

dence. Phys Rev Lett 2000;84:2774–2777. tion in the control of human action. Nonl Dyn

Freeman WJ and Vitiello G. Nonlinear brain Psych Life Sci 2008a;12(2):163-190.

dynamics as macroscopic manifestation of un- Ivancevic T, Jain L, Pattison J and Hariz

derlying many-body field dynamics. Phys Life A. Nonlinear Dynamics and Chaos Methods in

Rev 2006;(3)2:93–118. Neurodynamics and Complex Data Analysis.

Freeman WJ. Mass Action in the Nervous Nonl Dyn 2008(in press, on line first, Springer).

System. Acad Press, New York 1975/2004. Ivancevic V and Aidman E. Life-space foam:

Freeman WJ. Neurodynamics. An Explo- A medium for motivational and cognitive dy-

ration of Mesoscopic Brain Dynamics. Springer, namics. Physica A 2007;382:616–630.

London 2000. Ivancevic V and Ivancevic T. Applied Dif-

Gardiner C.W. Handbook of Stochastic Meth-ferential Geometry: A Modern Introduction.

ods for Physics Chemistry and Natural Sciences World Scientific, Singapore 2007a.

2nd ed. Springer, New York 1985. Ivancevic V and Ivancevic T. Complex Non-

Glimcher PW. Indeterminacy in brain and linearity: Chaos, Phase Transitions, Topology

behaviour. Ann Rev Psych 2005;56:25–56. Change and Path Integrals. Springer, Berlin

Gold M. A Kurt Lewin Reader the Com- 2008a.

plete Social Scientist. Am Psych Assoc Wash- Ivancevic V and Ivancevic T. Computational

ington 1999. Mind: A Complex Dynamics Perspective.

Haken H. Advanced Synergetics: Instabil- Springer, Berlin 2007c.

ity Hierarchies of Self-Organizing Systems and Ivancevic V and Ivancevic T. Geometrical

Devices (3nd ed). Springer, Berlin 1993. Dynamics of Complex Systems: A Unified Mod-

Haken H. Principles of Brain Functioning: elling Approach to Physics Control Biomechan-

A Synergetic Approach to Brain Activity Be- ics Neurodynamics and Psycho-Socio-Economical

havior and Cognition. Springer, Berlin 1996. Dynamics. Springer, Dordrecht 2006b.

Haken H. Synergetics: An Introduction (3rd Ivancevic V and Ivancevic T. High–Dimen-

ed). Springer, Berlin 1983. sional Chaotic and Attractor Systems. Springer,

Haken, H, Kelso, JAS and Bunz H. A the- Berlin 2006c.

oretical model of phase transitions in human Ivancevic V and Ivancevic T. Natural Bio-

hand movements. Biol Cybern 1985;51:347– dynamics. World Scientific, Singapore 2006a.

356. Ivancevic V and Ivancevic T. Neuro–Fuzzy

Associative Machinery for Comprehensive Brain

28

and Cognition Modelling. Springer, Berlin 2007b. Sebanz N, Bekkering H and Knoblich G.

Ivancevic V and Ivancevic T. Quantum Leap: Joint action: bodies and minds moving together.

From Dirac and Feynman Across the Universe Tr Cog Sci 2006;(10)2:70–76.

to Human Body and Mind. World Scientific, Sutton RS and Barto AG. Reinforcement

Singapore 2008b. Learning: An Introduction. MIT Press, Cam-

Izhikevich EM and Edelman GM. Large- bridge 1998.

Scale Model of Mammalian Thalamocortical Sys- Todorov E, Jordan MI. Optimal feedback

tems. PNAS 2008;105:3593–3598. control as a theory of motor coordination. Nat

Kelso, JAS. Dynamic Patterns: The Self Neurosci 2002;(5)11:1226-1235.

Organization of Brain and Behavior. MIT Press, Tognoli, E, Lagarde, J, DeGuzman, GC and

Cambridge 1995. Kelso JAS The phi complex as a neuromarker

Knill DC, Maloney LT and Trommershauser of human social coordination. PNAS

J. Sensorimotor Processing and Goal-Directed 2007;104(19):81908195.

Movement. J Vis 2007;7(5)i:1-2. Umezawa H. Advanced field theory: micro

Knoblich G and Jordan S. Action coordina- macro and thermal concepts. Am Inst Phys,

tion in individuals and groups: Learning antic- New York 1993.

ipatory control. J Exp Psych: Learn Mem Cog Wiener N. Cybernetics. Wiley, New York

2003;29:1006–1016. 1961.

Krylov NS. Works on the foundations of

statistical mechanics. Princeton Univ Press 1979.

Land, MF and Hayhoe, M. In what ways

do eye movements contribute to everyday ac-

tivities? Vis Res 2001;41:35593565.

Lewin K. Field Theory in Social Science.

Univ Chicago Press 1951.

Lewin K. Resolving Social Conflicts and Field

Theory in Social Science. Am Psych Assoc

Washington 1997.

Newman-Norlund RD, Noordzij ML, Meu-

lenbroek RGJ and Bekkering H. Exploring the

brain basis of joint action: Co-ordination of ac-

tions goals and intentions. Soc Neurosci

2007;(2)1:48–65.

Pettini M. Geometry and Topology in Hamil-

tonian Dynamics and Statistical Mechanics.

Springer, New York 2007.

Schöner G. Dynamical Systems Approaches

to Cognition. In: Cambridge Handbook of Com-

putational Cognitive Modelling. Cambridge Univ

Press 2007.

29

- PRP Anna University QATransféré parNaresh Konduru
- kkTransféré parRajat Kamra
- Exponential&PossionTransféré parChandler Jiacheng Liao
- poger_pem1Transféré parhamadani
- 2.MARKOV CHAAINS.pptxTransféré parsandy
- Rr411501 Performance Evaluation of Computer SystemsTransféré parSrinivasa Rao G
- Markov Chains Chap 5Transféré parraghu ram
- Chapter_2.pdfTransféré parHerbertTorres
- Stochastic ProcessesTransféré parmelanocitos
- StochastTransféré parLuis Alberto Fuentes
- psapphaTransféré parpili catena
- dsTransféré pardavid smith
- FinalTransféré parThinh
- revision.pdfTransféré parsdsds-54
- Atcs Exercises8Transféré parzukchu
- stochastic calculusTransféré parB.I
- Nata TextbTransféré parDeepak Bhardwaj
- 10.1.1.116Transféré parsandeep rawal
- chap2Transféré parRuhul Amin Mukut
- Locally Associative Planes of Combinatorially Liouville Systems and Convergence MethodsTransféré parHenry Pablo
- g 05313347Transféré parIOSRJEN : hard copy, certificates, Call for Papers 2013, publishing of journal
- Non StationaryTransféré parAshraf Ul Haque
- Page rank.pdfTransféré parShruti Bansal
- An integrated Facial Expression RecognitionTransféré parFelice Chew
- MATH3804 SyllabusTransféré parChakShanique
- BOOK - N. Balakrishnan - Anatoly Lisnianski, Ilia Frenkel Eds. Recent Advances in System Reliability Signatures - Reliability Engineering - SpringerTransféré parFreeLatinBird
- Chain Graph Models and Their Causal InterpretationsTransféré partradag
- 16891_TESE_SILVIA.pdfTransféré parMarcel Costa
- Statistical Process Control & Software Reliability Trend an Analysis Based on Inter Failure Time DataTransféré parNavneet
- Lec19Transféré parspitzersglare

- Redes Neuronales Cc COgTransféré parnemismd
- ¿EL ÉXITO O EL FRACASO DEPORTIVO PUEDE ESTAR DETERMINADO POR LOS ASPECTOS PSICOLÓGICOS ANTES DE COMPETIRTransféré parnemismd
- Ogbe KaTransféré parnemismd
- Cantos a OrisasTransféré parnemismd
- Control Stroop Interf, Evidence Bilingual Task Notas CopyTransféré parnemismd
- Writing and “Reference” Ifá DivinationTransféré parBabalawo Fagbenusola Aworeni
- Mundo Real e imagen percibidaTransféré parnemismd
- reneTransféré parnemismd
- CP47-1-DIMENTransféré parnemismd
- reneTransféré parnemismd
- cognitivo2Transféré parnemismd
- cognitivo2Transféré parnemismd
- stdy_ooriTransféré parnemismd
- Carta_espTransféré parnemismd
- Praising ObatalaTransféré parValentin Cabrera
- ORÍKÌ ELATransféré parnemismd
- zulia 2008Transféré parnemismd
- stdy_esuTransféré parnemismd

- Symbol Error Rate for QAMTransféré parĐinh Nhâm
- paranoid-reading-and-reparative-reading copy.pdfTransféré parMatías Moscardi
- ch2Transféré parapi-3777076
- On Discourse AnalysisTransféré parArmandouaa
- 2160710-DOS-Unit-3Transféré parmanjunath
- Simphiwe Mayisela Getuienis Oor OBK 5 Maart 2019Transféré parElvira
- Gender Identity Essay PDFTransféré parConnie Diep
- 168880-tkt-module-1-learner-needs.pdfTransféré parRachel Maria Ribeiro
- Review of Policy ResearchTransféré parintemperante
- Mbwela e bantuísmos no BrasilTransféré parDiego Marques
- Forms of InheritanceTransféré parshyambtech@
- MOODLE FMA1 Lecture Notes 2 in Math 1[1]Transféré parbea
- Rph Listening and Speaking My Cousin My Neighbour Week 1 and 2Transféré parFadzilah Khatim
- Anubias-Sheeja 3Transféré parnguyennam29
- proofofefficacydocumentfireaway-nihalnazeemrevisedTransféré parapi-233066115
- target pricingTransféré parRuby Refuerzo Buccat
- Case StudyTransféré parrasonawane
- Hassan a. Yousef-Power System Load Frequency Control_ Classical and Adaptive Fuzzy Approaches-Taylor & Francis_ CRC Press (2017)Transféré parChetan Kotwal
- Using PEST to Calibrate ModelsTransféré parKaren Hanson
- Paper-Análisis Comparativo de La Evolución de La Norma Para Diseño Sísmico de Edificios Nch433Transféré parSergio Meza
- paper- art and geometry final draftTransféré parapi-246220297
- 110040_1Transféré parAlcy Monteiro Jr
- PRESENTATION: Evolution of health situation in Papua and East IndonesiaTransféré parADB Health Sector Group
- common core math standards jigsaw 3rd gradeTransféré parapi-28847298
- Print ScienceTransféré parKassandra Ebol
- RequestedFile[1]Transféré parRAS2ROGER
- Contoh Soalan RAE Dan Jawapan 2-1Transféré parNorakmal Hisyam Akmal
- Chap18-Concurrency Control TechniquesTransféré parnomaddarcy
- One thousand but not twoTransféré parjprat
- ch5Transféré parMarcos Soriano