Classification Tool
First Claim
1. A training system for classifying a physical condition, the system comprising:
- a) an input module configured to receive n labeled data points {(a1, y1), . . . ,(an, yn)}, at least one of said n labeled data points describing at least one physical parameter for a physical phenomenon, said n labeled data points including vectors ai ε
and markers yi ε
{−
1,1}, each of said n labeled data points placed in one of two sets I+={i;
(ai,1)} and I−
={i;
(ai,−
1)};
b) a mathematical formulation module configured to use said n labeled data points to model said physical phenomenon as a mathematical formulation, said mathematical formulation including;
i) variables x=(w,b,Δ
), where w ε
, b ε
, Δ
ε
;
ii) an objective function ƒ
(x)=Δ
; and
iii) a plurality of constraints, said plurality of constraints including;
ci(x)≡
ci(w,b,Δ
)=(w,ai)−
b−
Δ
≧
0,iε
I+;
(1)
ci(x)≡
ci(w,b,Δ
)=−
(w,ai)+b−
Δ
≧
0,iε
I−
; and
(2)
∥
w∥
2=1;
(3)c) a transformer module configured to build a specific function L(w,b,Δ
,λ
,γ
,τ
) using;
i) said mathematical formulation;
ii) Lagrange multipliers λ
=(λ
1, . . . , λ
n);
iii) scaling parameters γ
, τ
, k;
iv) a class of transformation functions with a predefined set of properties; and
v) said n labeled data points, said n labeled data points further including a nonlinear resealing part;
d) a Lagrange multipliers updater module configured to calculate updated Lagrange multipliers {circumflex over (λ
)}=({circumflex over (λ
)}1, . . . , {circumflex over (λ
)}n);
e) a scaling parameters updater module configured to calculate updated scaling parameters {circumflex over (γ
)}, {circumflex over (τ
)}, {circumflex over (k)}; and
f) an iterative solver module configured to generate a classification rule using;
i) said specific function;
ii) said Lagrange multipliers updater module;
iii) said scaling parameters updater module;
iv) a general stopping criteria verifier module; and
v) an accuracy parameter (ε
).
3 Assignments
0 Petitions
Accused Products
Abstract
A classification system that includes a first input module, a marker module, unclassified Lagrange multiplier calculation module, and a classification module. The unclassified vector describes a physical condition. The marker module assigns a marker to the unclassified vector. The marker should place the unclassified vector into one of two classes. The unclassified Lagrange multiplier calculation module calculates a classifying Lagrange multiplier for the unclassified vector using the marker and a classification rule. The classification rule may be generated using a training system for classifying the physical condition. The classification module is configured to determine that the marker places the unclassified vector in the proper class if the classifying Lagrange multiplier is small.
11 Citations
25 Claims
-
1. A training system for classifying a physical condition, the system comprising:
-
a) an input module configured to receive n labeled data points {(a1, y1), . . . ,(an, yn)}, at least one of said n labeled data points describing at least one physical parameter for a physical phenomenon, said n labeled data points including vectors ai ε
and markers yi ε
{−
1,1}, each of said n labeled data points placed in one of two sets I+={i;
(ai,1)} and I−
={i;
(ai,−
1)};b) a mathematical formulation module configured to use said n labeled data points to model said physical phenomenon as a mathematical formulation, said mathematical formulation including; i) variables x=(w,b,Δ
), where w ε
, b ε
, Δ
ε
;ii) an objective function ƒ
(x)=Δ
; andiii) a plurality of constraints, said plurality of constraints including;
ci(x)≡
ci(w,b,Δ
)=(w,ai)−
b−
Δ
≧
0,iε
I+;
(1)
ci(x)≡
ci(w,b,Δ
)=−
(w,ai)+b−
Δ
≧
0,iε
I−
; and
(2)
∥
w∥
2=1;
(3)c) a transformer module configured to build a specific function L(w,b,Δ
,λ
,γ
,τ
) using;i) said mathematical formulation; ii) Lagrange multipliers λ
=(λ
1, . . . , λ
n);iii) scaling parameters γ
, τ
, k;iv) a class of transformation functions with a predefined set of properties; and v) said n labeled data points, said n labeled data points further including a nonlinear resealing part; d) a Lagrange multipliers updater module configured to calculate updated Lagrange multipliers {circumflex over (λ
)}=({circumflex over (λ
)}1, . . . , {circumflex over (λ
)}n);e) a scaling parameters updater module configured to calculate updated scaling parameters {circumflex over (γ
)}, {circumflex over (τ
)}, {circumflex over (k)}; andf) an iterative solver module configured to generate a classification rule using; i) said specific function; ii) said Lagrange multipliers updater module; iii) said scaling parameters updater module; iv) a general stopping criteria verifier module; and v) an accuracy parameter (ε
). - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22)
-
-
23. A classification system, said classification system comprising:
-
a) an first input module configured to receive an unclassified vector that describes a physical condition; b) a marker module configured to assign a marker to said unclassified vector, said marker placing said unclassified vector in one of two classes; c) an unclassified Lagrange multiplier calculation module configured to calculate an classifying Lagrange multiplier for said unclassified vector using; i) said marker; and ii) a classification rule, said classification rule configured to distinguish between said two classes, said classification rule generated using a training system for classifying said physical condition, said training system comprising; (1) a second input module configured to receive n labeled data points {(a1, y1), . . . ,(an, yn)}, at least one of said n labeled data points describing at least one physical parameter for a physical phenomenon, said n labeled data points including vectors ai ε
and markers yi ε
{−
1,1}, each of said n labeled data points placed in one of two sets I+={i;
(ai,1)} and I−
={i;
(ai,−
1)};(2) a mathematical formulation module configured to use said n labeled data points to model said physical phenomenon as a mathematical formulation, said mathematical formulation including; (a) variables x=(w,b,Δ
), where w ε
, b ε
, Δ
ε
;(b) an objective function ƒ
(x)=Δ
; and(c) a plurality of constraints, said plurality of constraints including;
ci(x)≡
ci(w,b,Δ
)=(w,ai)−
b−
Δ
≧
0,iε
I+;
(i)
ci(x)≡
ci(w,b,Δ
)=−
(w,ai)+b−
Δ
≧
0,iε
I−
; and
(ii)
∥
w∥
2=1;
(iii)(3) a transformer module configured to build a specific function L(w,b,Δ
,λ
,γ
,τ
) using;(a) said mathematical formulation; (b) Lagrange multipliers λ
=(λ
1, . . . , λ
n);(c) scaling parameters γ
, τ
, k;(d) a class of transformation functions with a predefined set of properties; and (e) said n labeled data points, said n labeled data points further including a nonlinear resealing part; (4) a Lagrange multipliers updater module configured to calculate updated Lagrange multipliers {circumflex over (λ
)}=({circumflex over (λ
)}1, . . . , {circumflex over (λ
)}n);(5) a scaling parameters updater module configured to calculate updated scaling parameters {circumflex over (γ
)}, {circumflex over (τ
)}, {circumflex over (k)}; and(6) an iterative solver module configured to generate a classification rule using; (a) said specific function; (b) said Lagrange multipliers updater module; (c) said scaling parameters updater module; (d) a general stopping criteria verifier module; and (e) an accuracy parameter (ε
); andd) a classification module configured determine that said marker places said unclassified vector in the proper class if said classifying Lagrange multiplier is small.
-
-
24. A method for generating a classification rule comprising:
-
a) receiving n labeled data points {(a1, y1), . . . ,(an, yn)}, at least one of said n labeled data points describing at least one physical parameter for a physical phenomenon, said n labeled data points including vectors ai ε
and markers yi ε
{−
1,1}, each of said n labeled data points placed in one of two sets I+={i;
(ai,1)} and I−
={i;
(ai,−
1)};b) modeling said physical phenomenon as a mathematical formulation using said n labeled data points, said mathematical formulation including; i) variables x=(w,b,Δ
), where w ε
b ε
Δ
ε
;ii) an objective function ƒ
(x)=Δ
; andiii) a plurality of constraints, said plurality of constraints including;
ci(x)≡
ci(w,b,Δ
)=(w,ai)−
b−
Δ
≧
0,iε
I+;
(1)
ci(x)≡
ci(w,b,Δ
)=−
(w,ai)+b−
Δ
≧
0,iε
I−
; and
(2)
∥
w∥
2=1;
(3)c) building a specific function L(w,b,Δ
,λ
,γ
,τ
) using;i) said mathematical formulation; ii) Lagrange multipliers λ
=(λ
1, . . . , λ
n);iii) scaling parameters γ
, τ
, k;iv) a class of transformation functions with a predefined set of properties; and v) said n labeled data points, said n labeled data points further including a nonlinear resealing part; d) calculating updated Lagrange multipliers {circumflex over (λ
)}=({circumflex over (λ
)}1, . . . , {circumflex over (λ
)}n);e) calculating updated scaling parameters {circumflex over (γ
)}, {circumflex over (τ
)}, {circumflex over (k)}; andf) iteratively generating a classification rule using; i) said specific function; ii) said Lagrange multipliers updater module; iii) said scaling parameters updater module; iv) a general stopping criteria verifier module; and v) an accuracy parameter (ε
).
-
-
25. A method classifying an unclassified vector, the method comprising:
-
a) assigning a marker to said unclassified vector, said marker placing said unclassified vector in one of two classes; b) calculating a classifying Lagrange multiplier for said unclassified vector using; i) said marker; and ii) a classification rule, said classification rule configured to distinguish between said two classes, said classification rule generated using a training method, said training method comprising; (1) receiving n labeled data points {(a1, y1), . . . ,(an, yn)}, at least one of said n labeled data points describing a physical parameter for a physical phenomenon, said n labeled data points including vectors ai ε
and markers yi ε
{−
1,1}, each of said n labeled data points placed in one of two sets I+={i;
(ai,1)} and I−
={i;
(ai,−
1)};(2) modeling said physical phenomenon as a mathematical formulation using said n labeled data points, said mathematical formulation including; (a) variables x=(w,b,Δ
), where w ε
, b ε
, Δ
ε
;(b) an objective function ƒ
(x)=Δ
; and(c) a plurality of constraints, said plurality of constraints including;
ci(x)≡
ci(w,b,Δ
)=(w,ai)−
b−
Δ
≧
0,iε
I+;
(i)
ci(x)≡
ci(w,b,Δ
)=−
(w,ai)+b−
Δ
≧
0,iε
I−
; and
(ii)
∥
w∥
2=1;
(iii)(3) Building a specific function L(w,b,Δ
,λ
,γ
,τ
) using;(a) said mathematical formulation; (b) Lagrange multipliers λ
=(λ
1, . . . , λ
n);(c) scaling parameters γ
, τ
, k;(d) a class of transformation functions with a predefined set of properties; and (e) said n labeled data points, said n labeled data points further including a nonlinear resealing part; (4) calculating updated Lagrange multipliers {circumflex over (λ
)}=({circumflex over (λ
)}1, . . . , {circumflex over (λ
)}n);(5) calculating updated scaling parameters {circumflex over (γ
)}, {circumflex over (τ
)}, {circumflex over (k)}; and(6) iteratively generating a classification rule using; (a) said specific function; (b) said Lagrange multipliers updater module; (c) said scaling parameters updater module; (d) a general stopping criteria verifier module; and (e) an accuracy parameter (ε
); andc) determining that said marker places said unclassified vector in the proper class if said classifying Lagrange multiplier is small.
-
Specification