2. Modeling approach based on free energyfree energy and
distortion energydistortion energy
Near linear model - aims for the simplest
explanations
Estimation of clustering dynamical parameters
by statistical inference
Multi-spectral decomposition, in hierarchy of
scales
Application: scale analysis of complex systemsApplication: scale analysis of complex systems
Multi-scale decomposition via
dynamical cascades
3. Introduction (1 of 2)
Clustering parameters:
• Selected window of computation: Wr
• Computed cluster vector within Wr
Statistical inference defines PDF, with the associated
distortion energies, F and V.
Energy functions are generally multi-dimensional and
non-convex/concave
Non-linear map defines dynamical scale-space
clustering.
Clustering is a model-free approach to signal decomp.Clustering is a model-free approach to signal decomp.
c
4. Introduction (2 of 2)
Data binding – no ordering relation assumed (model-free), although a
priori neighboring information used - speeds up numerical computation.
Brain waves: 2 nucleons decomp.
5. Model of signal distortion:
- definitions
Distortion measure:
1. d = z2
= (Cx-X)2
+ (Cy-Y)2
e.g. in still images
2. d = z2
= e.g. in motion images
2
][ vIIt
⋅∇+
Partition functions: ,
2
∑
−
=
rW
z
rZ
β
Distortion energies -
free energy, and variance:
( )∑=
rW
PvxdV
,
PDF: ,
2
Z
r
P
zβ−
=
( ) ,log
1
, r ZvF
β
β −=
6. Scale-space computing
Series of convex/concave min/max of free energy F
brings in eq. up-scale melting & down-scale cooling:
( ) ( ) ,
1
0
∫=
β
ββ
β
β dVF ( )
( )∫
=
−
β
ββ
β 0
dV
rZ
( )
)1(
.
,2
β
δβ
∂
∂
±=
−=
∂
∂
−= ∑
F
PIc
c
F
c
rW
Evolution scheme – path integrals:
Way to move through the scale-space ?
7. Motion through the scale-space:
wave information propagation
Mass-energy-information conservation principle
Coupling +/- mass-energy possible
12
ββ >
( )β,vF
Vv grad=
(1)
(2)
)2(
v
V
v
∂
∂
±=δ
)3(
β
δβ
∂
∂
±=
F
∫ ∫ =
∂
∂
+
∂
∂
=
S S
d
v
V
vd
F
dU 0β
β
V
F 2
2
2
∇=
∂
∂
β
8. Most singular manifolds (MSM), and a data nucleons
MSM (2 colors), and a nucleon (4 colors)
10. Scalable coding
Coupled data structure of the hierarchy of
binary decompsitions.
Efficient coding, control, data transfer.
Parallelization: computing and control by
parallel computing architectures.
(v4, W4)(v3, W3)
(v2, W2)(v1, W1)(v0, W0)
βc
3
(v3
, W3
)
βc
2
(v2
, W2
)
βc
0
(v0
, W0
)
βc
1
(v1
, W1
)
11. Focus on computability and complexity –
relationship to statistical physics
o Computing paradigm assumes:
o Motion via scale-space wave information propagation, and
o Uncertainty relation wrt the information content of a cluster
o What makes it polynomial in complexity (ref. 2)?
o Unique statistical description, although chaotic motion possible
o No strange attractors due to the conservative motion
Within this description: multi-scale decomposition of the information
content into clusters
Coupling of the energy exchange – synergetics
Coupled manifolds spanning the content of the information clusters
12. Summary presentation of current work
Scalable data decomposition:
• genotype information, encription and coding,
• progressive transmission
• segmented control
Multidimension scaling:
• Dynamical cascades via space-time synergism.
Images: multi-spectral decomposition and clusters
couplings, spectral signature recognition
Movements:
• trajectory analysis
• learning
Bio/chemical informatics:
• data-mining and knowledge discovery
Synchronous computing scheme: upscale melting &
downscale cooling
Parallel computing implementation
13. Perspective and future directions
Scale-space approach to computing, analysis,
and signal control.
Bioinformatics, computational physics.
Model of signal distortion analogous to that of the
networked physical systems.
Dynamical data modeling: multidimensional
scaling via spatio-temporal synergism.
Segmented control of multispectral components.
Parallel computer implementation.
14. BIBLIOGRAPHY
Jovovic, M., [2013], Genotype Information and the Space-Time Generatio
Jovovic, M., [2012], Stochastic Resonance Synergetics (SRS)
Hypothesis: A Road to Attention, Memory, and Behavioral Data-
driven Study
Jovovic, M., [2011], Brain wave synergies, analysis and coding
Jovovic, M., G. Fox, [2007], Multi-dimensional data scaling – dynamical
cascade approach, Technical Report - Indiana University, USA.
Jovovic, M., H. Yahia, I. Herlin [2003], Hierarchical scale
decomposition of images – singular features analysis, Technical report,
INRIA, AIR Lab, France.
Jovovic, M., S. Jonic, D. Popovic [1999], Automatic synthesis of
synergies for control of reaching – hierarchical clustering. Medical
Engineering and Physics.
Editor's Notes
- Conservative information propagation – no dissipation, just decomposition across the scales