…p…^†[…fiflF”¯ Pattern Recognition

Size: px
Start display at page:

Download "…p…^†[…fiflF”¯ Pattern Recognition"

Transcription

1 Pattern Recognition Shin ichi Satoh National Institute of Informatics June 11, 2019

2 (Support Vector Machines) (Support Vector Machines: SVM) SVM Vladimir N. Vapnik and Alexey Ya. Chervonenkis 1963 SVM (Structural Risk) : ( : empirical risk) (VC ) (overfitting) SVM Bernhard E. Boser, Isabelle M. Guyon and Vladimir N. Vapnik 1992 Corinna Cortes and Vapnik 1993

3 SVM SVM (Linear Support Vector Machines) margin

4 SVM X = {x i R d } (i = 1,, n) y i { 1, 1} ( ) w x b = 0 w x b = 1 and w x b = 1 2 w

5 SVM w w x i b 1 for x i with y i = 1 w x i b 1 for x i with y i = 1 y i (w x i b) 1 for all i = 1,, n Minimize w subject to y i (w x i b) 1 for all i = 1,, n.

6 (Primal Form) (quadratic programming optimization problem) 1 arg min (w,b) 2 w 2 subject to (for any i = 1,, n) y i (w x i b) 1

7 (Dual Form) (Lagrange multipliers) α i { } 1 n L = 2 w 2 α i [y i (w x i b) 1] i=1 arg min max w,b α Karush-Kuhn-Tucker (KKT Karush-Kuhn-Tucker conditions) L L = 0, w b = 0 α i 0, α i [y i (w x i b) 1] = 0 L

8 Maximize f (x) subject to g i (x) 0, h j (x) = 0 Karush-Kuhn-Tucker Karush-Kuhn-Tucker f (x) α i g i (x) λ j h j (x) = 0 g i (x) 0, h j (x) = 0 α i 0 α i g i (x) = 0

9 Maximize f (x) subject to h j (x) = 0 : (Lagrange Multipliers Method) (Lagrangian): L = f (x) λ j h j (x) : L = f (x) λ j h j (x) = 0 h j (x) = 0

10 L w = 0 w = α i y i x i L b = 0 α i y i = 0 α i [y i (w x i b) 1] = 0 if y i (w x i b) 1 > 0 then α i = 0, otherwise (y i (w x i b) 1 = 0) α i 0 α i > 0 x i b = 1 {α i > 0} (w x i y i ) α i >0

11 Maximize: n L(α) = α i 1 α i α j y i y j xi T x j 2 subject to: i=1 α i 0 and i,j n α i y i = 0 i=1

12 Minimize subject to (Quadratic Programming Optimization) 1 2 x T Qx + p T x Cx b C eq x = b eq LB x UB x = α, Q i,j = y i y j x T i x j, p = [1 1 1], LB = 0, C eq = y, b eq = 0

13 SVM (Python) h=x*l qpp = cvxopt.matrix(h.t.dot(h)) qpq = cvxopt.matrix(-np.ones(n), (n, 1)) qpg = cvxopt.matrix(-np.eye(n)) qph = cvxopt.matrix(np.zeros(n), (n, 1)) qpa = cvxopt.matrix(l.astype(float), (1, n)) qpb = cvxopt.matrix(0.) cvxopt.solvers.options[ abstol ] = 1e-5 cvxopt.solvers.options[ reltol ] = 1e-10 cvxopt.solvers.options[ show_progress ] = False res=cvxopt.solvers.qp(qpp, qpq, qpg, qph, qpa, qpb) alpha = np.reshape(np.array(res[ x ]),-1) w=np.sum(x*(np.ones(n)*(l*alpha)),axis=1) sv=alpha>1e-5 isv=np.where(sv)[-1] b=np.sum(w.t.dot(x[:,isv])-l[isv])/np.sum(sv)

14 SVM (Matlab) h=x; h(:,l<0)=-h(:,l<0); options=optimset( Algorithm, interior-point-convex ); alpha=quadprog(h *h,-ones(1,size(x,2)),[],[],l,0,... zeros(1,size(x,2)),[],[],options) ; w=sum(x.*(ones(size(x,1),1)*(l.*alpha)),2); sv=alpha>1e-5; isv=find(sv); b=sum(w *x(:,isv)-l(isv))/sum(sv);

15 SVM (Scilab) h=x; h(:,l<0)=-h(:,l<0); alpha=quapro(h *h,-ones(size(x,2),1),l,0,... zeros(size(x,2),1),[],1) ; w=sum(x.*(ones(size(x,1),1)*(l.*alpha)),2); sv=alpha>1e-5; isv=find(sv); b=sum(w *x(:,isv)-l(isv))/sum(sv);

16 (linear)

17 (slinear)

18 (slack variables) ξ i y i (w x i b) 1 ξ i { 1 arg min w,ξ,b 2 w 2 + C subject to } n ξ i i y i (w x i b) 1 ξ i, ξ 0

19 arg min w,b { 1 2 w 2 + C } n max(1 y i (w x i + b), 0) max(1 y i (w x i + b), 0) (hinge loss) i hinge loss 0-1 loss

20 KKT Maximize subject to L(α) = n α i 1 2 i=1 0 α i C, i,j α i α j y i y j xi T x j n α i y i = 0 : x i with 0 < α i < C (x i with α i = C ). i=1

21 SVM

22 (slinear)

23 (qlinear)

24 x ϕ(x) : x = [x 1 x 2 ] T, ϕ(x) = [x 1 x 2 x 2 1 x 1x 2 x 2 2 ]T. ϕ(x) x

25 SVM k(x, y) = ϕ(x) ϕ(y) : (Polynomial Kernel) k(x, y) = (x y + 1) p, k(x, y) = (x y) p (Gaussian Kernel, Radial Basis Function (RBF) Kernel) x y 2 k(x, y) = exp( ) 2σ 2

26 SVM Maximize n L(α) = α i 1 α i α j y i y j xi T x j 2 subject to i=1 0 α i C, i,j n α i y i = 0 i=1 x i x j

27 Maximize L(α) = = n α i 1 α i α j y i y j ϕ(x i ) T ϕ(x j ) 2 i=1 i,j n α i 1 α i α j y i y j k(x i, x j ) 2 i=1 i,j subject to 0 α i C, n α i y i = 0 i=1

28 α i w = α i y i ϕ(x i ) ϕ(x i ) w b = 1 #sv = 1 #sv = 1 #sv (w ϕ(x i ) y i ) i sv i sv j i sv j α j y j ϕ(x j ) T ϕ(x i ) y i α j y j k(x j, x i ) y i

29 x f (x) = w ϕ(x) b = α i y i ϕ(x i ) T ϕ(x) b = α i y i k(x i, x) b f (x) x

30 (qlinear, Polynomial kernel)

31 (nonlinear, C=1, RBF kernel)

32 (nonlinear, C=1000, RBF kernel)

33 1 SVM : SVM SVM Python: cvxopt ( pip install cvxopt ) Matlab: quadprog (optimization toolbox ) Scilab: quapro (quapro toolbox ) linear, slinear, qlinear, nonlinear

34 w x b = 1 w x b = 1 2 w 2

35 SVM : arg min w,ξ,b { 1 2 w 2 + C } n ξ i i 2 subject to : Maximize subject to y i (w x i b) 1 ξ i, ξ 0 L(α) = n α i 1 α i α j y i y j xi T x j 2 i=1 0 α i C, i,j n α i y i = 0 i=1

TD(0) Q AC (Reward): () Pr(r t+1 s t+1 = s,s t = s, a t = a) t R a ss = E(r t+1 s t+1 = s,s t = s, a t = a) R t = r t+1 + γr t γ T r t+t +1 = T

TD(0) Q AC (Reward): () Pr(r t+1 s t+1 = s,s t = s, a t = a) t R a ss = E(r t+1 s t+1 = s,s t = s, a t = a) R t = r t+1 + γr t γ T r t+t +1 = T () 2009 TD(0) Q AC 2009 1/42 2009 2/42 TD(0) Q AC (Renforcement Learnng) : (polcy) Acton: a t Agent (= Controller) Envronment (= Controlled object) State: s t Reward: r t TD(0) Q AC (Envronment) (Markov

More information

( ) () () ( ) () () () ()

( ) () () ( ) () () () () 5 1! (Linear Programming, LP) LP OR LP 1.1 1.1.1 1. 2. 3. 4. 4 5. 1000 4 1.1? 1.2 1 1 http://allrecipes.com/ 6 1 1.1 ( ) () 1 0.5 1 0.75 200 () 1.5 1 0.5 1 50 ( ) 2 2 1 30 () 2.25 0.5 2 2.25 30 () 2 100

More information

nakata/nakata.html p.1/20

nakata/nakata.html p.1/20 http://www.me.titech.ac.jp/ nakata/nakata.html p.1/20 1-(a). Faybusovich(1997) Linear systems in Jordan algebras and primal-dual interior-point algorithms,, Euclid Jordan p.2/20 Euclid Jordan V Euclid

More information

x T = (x 1,, x M ) x T x M K C 1,, C K 22 x w y 1: 2 2

x T = (x 1,, x M ) x T x M K C 1,, C K 22 x w y 1: 2 2 Takio Kurita Neurosceince Research Institute, National Institute of Advanced Indastrial Science and Technology takio-kurita@aistgojp (Support Vector Machine, SVM) 1 (Support Vector Machine, SVM) ( ) 2

More information

Natural Language Processing Series 1 WWW WWW 1. ii Foundations of Statistical NLPMIT Press 1999 2. a. b. c. 25 3. a. b. Web WWW iii 2. 3. 2009 6 v 2010 6 1. 1.1... 1 1.2... 4 1.2.1... 6 1.2.2... 12 1.2.3...

More information

第3章 非線形計画法の基礎

第3章 非線形計画法の基礎 3 February 25, 2009 1 Armijo Wolfe Newton 2 Newton Lagrange Newton 2 SQP 2 1 2.1 ( ) S R n (n N) f (x) : R n x f R x S f (x ) = min x S R n f (x) (nonlinear programming) x 0 S k = 0, 1, 2, h k R n ɛ k

More information

Support Vector Machine (SVM) 4 SVM SVM 2 80% 100% SVM SVM SVM 4 SVM 2 2 SVM 4

Support Vector Machine (SVM) 4 SVM SVM 2 80% 100% SVM SVM SVM 4 SVM 2 2 SVM 4 Analysis of Groove Feelings of Drums Plays 47 56340 19 1 31 Support Vector Machine (SVM) 4 SVM SVM 2 80% 100% SVM SVM SVM 4 SVM 2 2 SVM 4 1 1 1.1........................................ 1 1.1.1.............................

More information

1 n 1 1 2 2 3 3 3.1............................ 3 3.2............................. 6 3.2.1.............. 6 3.2.2................. 7 3.2.3........................... 10 4 11 4.1..........................

More information

untitled

untitled ,, 2 2.,, A, PC/AT, MB, 5GB,,,, ( ) MB, GB 2,5,, 8MB, A, MB, GB 2 A,,,? x MB, y GB, A (), x + 2y () 4 (,, ) (hanba@eee.u-ryukyu.ac.jp), A, x + 2y() x y, A, MB ( ) 8 MB ( ) 5GB ( ) ( ), x x x 8 (2) y y

More information

ERATO100913

ERATO100913 ERATO September 13, 2010, DC2 1/25 1. 2 2. 2/25 3/25 3/25 2 3/25 2 3/25 1 1 0.5 0.5 0 0 0.5 1 0 0 0.5 1 4/25 1 1 0.5 0.5 0 0 0.5 1 (0, 0) 0 0 0.5 1 4/25 1 1 0.5 0.5 0 0 0.5 1 (0, 0) ( 1, 0) 0 0 0.5 1 4/25

More information

( ) ? () 1.1 ( 3 ) j x j 10 j 1 10 j = 1,..., 10 x 1 + x x 10 =

( ) ? () 1.1 ( 3 ) j x j 10 j 1 10 j = 1,..., 10 x 1 + x x 10 = 5 1! (Linear Programming, LP) LP OR LP 1.1 1.1.1 1. 2. 3. 4. 5. ( ) ( ) 1.1 6 1 1.1 ( ) 1 110 2 98 3 85 4 90 5 73 6 62 7 92 8 88 9 79 10 75 1.1.2 4? 900 40 80 120 () 1.1 ( 3 ) j x j 10 j 1 10 j = 1,...,

More information

inkiso.dvi

inkiso.dvi Ken Urai May 19, 2004 5 27 date-event uncertainty risk 51 ordering preordering X X X (preordering) reflexivity x X x x transitivity x, y, z X x y y z x z asymmetric x y y x x = y X (ordering) completeness

More information

1 1 2 3 2.1.................. 3 2.2.......... 6 3 7 3.1......................... 7 3.1.1 ALAGIN................ 7 3.1.2 (SVM).........................

1 1 2 3 2.1.................. 3 2.2.......... 6 3 7 3.1......................... 7 3.1.1 ALAGIN................ 7 3.1.2 (SVM)......................... [5] Yahoo! Yahoo! (SVM) 3 F 7 7 (SVM) 3 F 6 0 1 1 2 3 2.1.................. 3 2.2.......... 6 3 7 3.1......................... 7 3.1.1 ALAGIN................ 7 3.1.2 (SVM)........................... 8

More information

2 1/2 1/4 x 1 x 2 x 1, x 2 9 3x 1 + 2x 2 9 (1.1) 1/3 RDA 1 15 x /4 RDA 1 6 x /6 1 x 1 3 x 2 15 x (1.2) (1.3) (1.4) 1 2 (1.5) x 1

2 1/2 1/4 x 1 x 2 x 1, x 2 9 3x 1 + 2x 2 9 (1.1) 1/3 RDA 1 15 x /4 RDA 1 6 x /6 1 x 1 3 x 2 15 x (1.2) (1.3) (1.4) 1 2 (1.5) x 1 1 1 [1] 1.1 1.1. TS 9 1/3 RDA 1/4 RDA 1 1/2 1/4 50 65 3 2 1/15 RDA 2/15 RDA 1/6 RDA 1 1/6 1 1960 2 1/2 1/4 x 1 x 2 x 1, x 2 9 3x 1 + 2x 2 9 (1.1) 1/3 RDA 1 15 x 1 + 2 1/4 RDA 1 6 x 1 1 4 1 1/6 1 x 1 3

More information

kut-paper-template.dvi

kut-paper-template.dvi 26 Discrimination of abnormal breath sound by using the features of breath sound 1150313 ,,,,,,,,,,,,, i Abstract Discrimination of abnormal breath sound by using the features of breath sound SATO Ryo

More information

数理計画法入門 サンプルページ この本の定価 判型などは, 以下の URL からご覧いただけます. このサンプルページの内容は, 初版 1 刷発行時のものです.

数理計画法入門 サンプルページ この本の定価 判型などは, 以下の URL からご覧いただけます.   このサンプルページの内容は, 初版 1 刷発行時のものです. 数理計画法入門 サンプルページ この本の定価 判型などは, 以下の URL からご覧いただけます. http://www.morikita.co.jp/books/mid/092181 このサンプルページの内容は, 初版 1 刷発行時のものです. i 1947 G.B. Dantzig 3 (1) (2) (3) (4) Microsoft Excel Web 4 1 2 2 2 2 3 ii 4

More information

(note-02) Rademacher 1/57

(note-02) Rademacher 1/57 (note-02) Rademacher 1/57 (x 1, y 1 ),..., (x n, y n ) X Y f : X Y Y = R f Y = {+1, 1}, {1, 2,..., G} f x y 1. (x 1, y 1 ),..., (x n, y n ) f(x i ) ( ) 2. x f(x) Y 2/57 (x, y) f(x) f(x) y (, loss) l(f(x),

More information

Dimitrov [2] MPC Dimitrov MPC MPC MPC [14] MPC [7] KKT MPC [14] Rao LDL MPC MCP [5] 2 MPC u k

Dimitrov [2] MPC Dimitrov MPC MPC MPC [14] MPC [7] KKT MPC [14] Rao LDL MPC MCP [5] 2 MPC u k 1 2 In model predictive control (MPC), an optimal control problem is solved at each time steps to determine control input. To realize on-line control of MPC, reducing computational time is requisite. In

More information

106 4 4.1 1 25.1 25.4 20.4 17.9 21.2 23.1 26.2 1 24 12 14 18 36 42 24 10 5 15 120 30 15 20 10 25 35 20 18 30 12 4.1 7 min. z = 602.5x 1 + 305.0x 2 + 2

106 4 4.1 1 25.1 25.4 20.4 17.9 21.2 23.1 26.2 1 24 12 14 18 36 42 24 10 5 15 120 30 15 20 10 25 35 20 18 30 12 4.1 7 min. z = 602.5x 1 + 305.0x 2 + 2 105 4 0 1? 1 LP 0 1 4.1 4.1.1 (intger programming problem) 1 0.5 x 1 = 447.7 448 / / 2 1.1.2 1. 2. 1000 3. 40 4. 20 106 4 4.1 1 25.1 25.4 20.4 17.9 21.2 23.1 26.2 1 24 12 14 18 36 42 24 10 5 15 120 30

More information

CVaR

CVaR CVaR 20 4 24 3 24 1 31 ,.,.,. Markowitz,., (Value-at-Risk, VaR) (Conditional Value-at-Risk, CVaR). VaR, CVaR VaR. CVaR, CVaR. CVaR,,.,.,,,.,,. 1 5 2 VaR CVaR 6 2.1................................................

More information

鉄鋼協会プレゼン

鉄鋼協会プレゼン NN :~:, 8 Nov., Adaptive H Control for Linear Slider with Friction Compensation positioning mechanism moving table stand manipulator Point to Point Control [G] Continuous Path Control ground Fig. Positoining

More information

Overview (Gaussian Process) GPLVM GPDM 2 / 59

Overview (Gaussian Process) GPLVM GPDM 2 / 59 daichi@ism.ac.jp 2015-3-3( ) 1 / 59 Overview (Gaussian Process) GPLVM GPDM 2 / 59 (Gaussian Process) y 2 1 0 1 2 3 8 6 4 2 0 2 4 6 8 x x y (regressor) D = { (x (n), y (n) ) } N, n=1 x (n+1) y (n+1), (

More information

第10章 アイソパラメトリック要素

第10章 アイソパラメトリック要素 June 5, 2019 1 / 26 10.1 ( ) 2 / 26 10.2 8 2 3 4 3 4 6 10.1 4 2 3 4 3 (a) 4 (b) 2 3 (c) 2 4 10.1: 3 / 26 8.3 3 5.1 4 10.4 Gauss 10.1 Ω i 2 3 4 Ξ 3 4 6 Ξ ( ) Ξ 5.1 Gauss ˆx : Ξ Ω i ˆx h u 4 / 26 10.2.1

More information

Research on decision making in multi-player games with imperfect information

Research on decision making in multi-player games with imperfect information Research on decision making in multi-player games with imperfect information 37-086521 22 2 9 UCT UCT 46 % 60000 9 % 1 1 1.1........................................ 1 1.2.....................................

More information

IS1-09 第 回画像センシングシンポジウム, 横浜,14 年 6 月 2 Hough Forest Hough Forest[6] Random Forest( [5]) Random Forest Hough Forest Hough Forest 2.1 Hough Forest 1 2.2

IS1-09 第 回画像センシングシンポジウム, 横浜,14 年 6 月 2 Hough Forest Hough Forest[6] Random Forest( [5]) Random Forest Hough Forest Hough Forest 2.1 Hough Forest 1 2.2 IS1-09 第 回画像センシングシンポジウム, 横浜,14 年 6 月 MI-Hough Forest () E-mail: ym@vision.cs.chubu.ac.jphf@cs.chubu.ac.jp Abstract Hough Forest Random Forest MI-Hough Forest Multiple Instance Learning Bag Hough Forest

More information

fiš„v3.dvi

fiš„v3.dvi (2001) 49 1 23 42 2000 10 16 2001 4 23 NTT * 1. 1.1 1998 * 104 0033 1 21 2 7F 24 49 1 2001 1999 70 91 MIT M. Turk Recognition Using Eigenface (Turk and Pentland (1991)). 1998 IC 1 CPU (Jain and Waller

More information

s t 1, 2,..., 10 s t a, b,..., k t s 1, 2,..., 10 1 a, b,..., k 1 s t ts 1 0 ( 2.25) ½ ¾ ½¼ x 1j = 1 x 2c = 1 x 3e = 1

s t 1, 2,..., 10 s t a, b,..., k t s 1, 2,..., 10 1 a, b,..., k 1 s t ts 1 0 ( 2.25) ½ ¾ ½¼ x 1j = 1 x 2c = 1 x 3e = 1 72 2 2 2 2.24 2 s t, 2,..., 0 s t a, b,..., k t s, 2,..., 0 a, b,..., k s t 0 ts 0 ( 2.25) 2.24 2 ½ ¾ ½¼ x j = x 2c = x 3e = x 4s = x 5g = x 6i = x 7d = x 8h = x 9f = x 0k = x ta = x tb = x ts = 9 2.26

More information

…X…p†[…X’³‚¥›»‡¨‡æ‡Ñ…}…‰…`…J†[…l…‰−w‘K‡Ì‡½‡ß‡Ì“ÅfiK›»…A…‰…S…−…Y…•‡ÆCV†EPR‡Ö‡Ì›žŠp

…X…p†[…X’³‚¥›»‡¨‡æ‡Ñ…}…‰…`…J†[…l…‰−w‘K‡Ì‡½‡ß‡Ì“ÅfiK›»…A…‰…S…−…Y…•‡ÆCV†EPR‡Ö‡Ì›žŠp CV PR 1, 1, 2 1 2 2009-08-31 @ PRMU/CVIM http://www.ibis.t.u-tokyo.ac.jp/ryotat/prmu09/ UT / Tokyo Tech DAL PRMU/CVIM 1 / 58 Outline 1-2 Dual Augmented Lagrangian Proximal minimization Legendre 3 4 UT

More information

../dvi98/me98enve.dvi

../dvi98/me98enve.dvi Chapter 5 1 2 CHAPTER 5. 5.1 8 (x ) 2 +(y + ) 2 1=j 2 R9 x + y = x + y = p 2 x + y = p 2 ff(x; y; ) =j 2 Rg ( f(x; y; ) = @ @ f(x; y; ) = (5:1) f 3 (x; y) = f 3 (x; y) ==)9 : f(x; y; ) = @f @x @f @y 1

More information

ohpmain.dvi

ohpmain.dvi fujisawa@ism.ac.jp 1 Contents 1. 2. 3. 4. γ- 2 1. 3 10 5.6, 5.7, 5.4, 5.5, 5.8, 5.5, 5.3, 5.6, 5.4, 5.2. 5.5 5.6 +5.7 +5.4 +5.5 +5.8 +5.5 +5.3 +5.6 +5.4 +5.2 =5.5. 10 outlier 5 5.6, 5.7, 5.4, 5.5, 5.8,

More information

50. (km) A B C C 7 B A 0

50. (km) A B C C 7 B A 0 49... 5 A B C. (. )?.. A A B C. A 4 0 50. (km) A B C..9 7. 4.5.9. 5. 7.5.0 4..4 7. 5.5 5.0 4. 4.. 8. 7 8.8 9.8. 8 5. 5.7.7 9.4 4. 4.7 0 4. 7. 8.0 4.. 5.8.4.8 8.5. 8 9 5 C 7 B 5 8 7 4 4 A 0 0 0 4 5 7 8

More information

*2.5mm ”ŒŠá‡ÆfiÁ™¥‡Ì…Z†[…t…X…N…−†[…j…fi…O

*2.5mm ”ŒŠá‡ÆfiÁ™¥‡Ì…Z†[…t…X…N…−†[…j…fi…O I. Takeuchi, Nagoya Institute of Technology 1/38 f(x) = w 1 x 1 + w 2 x 2 +... + w d x d f(x) = α 1 K(x, x 1 ) + α 2 K(x, x 2 ) +... + α n K(x, x n ) {wj } d j=1 f {αi } n i=1 f I. Takeuchi, Nagoya Institute

More information

‚åŁÎ“·„´Šš‡ðŠp‡¢‡½‹âfi`fiI…A…‰…S…−…Y…•‡ÌMarkovŸA“½fiI›ð’Í

‚åŁÎ“·„´Šš‡ðŠp‡¢‡½‹âfi`fiI…A…‰…S…−…Y…•‡ÌMarkovŸA“½fiI›ð’Í Markov 2009 10 2 Markov 2009 10 2 1 / 25 1 (GA) 2 GA 3 4 Markov 2009 10 2 2 / 25 (GA) (GA) L ( 1) I := {0, 1} L f : I (0, ) M( 2) S := I M GA (GA) f (i) i I Markov 2009 10 2 3 / 25 (GA) ρ(i, j), i, j I

More information

untitled

untitled K-Means 1 5 2 K-Means 7 2.1 K-Means.............................. 7 2.2 K-Means.......................... 8 2.3................... 9 3 K-Means 11 3.1.................................. 11 3.2..................................

More information

<4D F736F F D B B83578B6594BB2D834A836F815B82D082C88C60202E646F63>

<4D F736F F D B B83578B6594BB2D834A836F815B82D082C88C60202E646F63> MATLAB/Simulink による現代制御入門 サンプルページ この本の定価 判型などは, 以下の URL からご覧いただけます. http://www.morikita.co.jp/books/mid/9241 このサンプルページの内容は, 初版 1 刷発行当時のものです. i MATLAB/Simulink MATLAB/Simulink 1. 1 2. 3. MATLAB/Simulink

More information

Real AdaBoost HOG 2009 3 A Graduation Thesis of College of Engineering, Chubu University Efficient Reducing Method of HOG Features for Human Detection based on Real AdaBoost Chika Matsushima ITS Graphics

More information

kato-kuriki-2012-jjas-41-1.pdf

kato-kuriki-2012-jjas-41-1.pdf Vol. 41, No. 1 (2012), 1 14 2 / JST CREST T 2 T 2 2 K K K K 2,,,,,. 1. t i y i 2 2 y i = f (t i ; c) + ε i, f (t; c) = c h t h = c ψ(t), i = 1,...,N (1) h=0 c = (c 0, c 1, c 2 ), ψ(t) = (1, t, t 2 ) 3

More information

II

II II 16 16.0 2 1 15 x α 16 x n 1 17 (x α) 2 16.1 16.1.1 2 x P (x) P (x) = 3x 3 4x + 4 369 Q(x) = x 4 ax + b ( ) 1 P (x) x Q(x) x P (x) x P (x) x = a P (a) P (x) = x 3 7x + 4 P (2) = 2 3 7 2 + 4 = 8 14 +

More information

f(x) x S (optimal solution) f(x ) (optimal value) f(x) (1) 3 GLPK glpsol -m -d -m glpsol -h -m -d -o -y --simplex ( ) --interior --min --max --check -

f(x) x S (optimal solution) f(x ) (optimal value) f(x) (1) 3 GLPK glpsol -m -d -m glpsol -h -m -d -o -y --simplex ( ) --interior --min --max --check - GLPK by GLPK http://mukun mmg.at.infoseek.co.jp/mmg/glpk/ 17 7 5 : update 1 GLPK GNU Linear Programming Kit GNU LP/MIP ILOG AMPL(A Mathematical Programming Language) 1. 2. 3. 2 (optimization problem) X

More information

2016 DRDoS F037-1

2016 DRDoS F037-1 2016 DRDoS 2017 1 30 5115F037-1 1 4 1.1..................................... 4 1.2..................................... 5 1.3.................................... 6 2 DRDoS 7 2.1 DRDoS.....................................

More information

例題ではじめる部分空間法 - パターン認識へのいざない -

例題ではじめる部分空間法  - パターン認識へのいざない - - - ( ) 69 2012 5 22 (1) ( ) MATLAB/Octave 3 download http://www.tuat.ac.jp/ s-hotta/rsj2012 (2) ( ) [1] 対応付け 0 1 2 3 4 未知パターン ( クラスが未知 ) 利用 5 6 7 8 クラス ( 概念 ) 9 訓練パターン ( クラスが既知 ) (3) [1] 識別演算部 未知パターン

More information

( ) ( 40 )+( 60 ) Schrödinger 3. (a) (b) (c) yoshioka/education-09.html pdf 1

( ) ( 40 )+( 60 ) Schrödinger 3. (a) (b) (c)   yoshioka/education-09.html pdf 1 2009 1 ( ) ( 40 )+( 60 ) 1 1. 2. Schrödinger 3. (a) (b) (c) http://goofy.phys.nara-wu.ac.jp/ yoshioka/education-09.html pdf 1 1. ( photon) ν λ = c ν (c = 3.0 108 /m : ) ɛ = hν (1) p = hν/c = h/λ (2) h

More information

1. A0 A B A0 A : A1,...,A5 B : B1,...,B12 2. 5 3. 4. 5. A0 (1) A, B A B f K K A ϕ 1, ϕ 2 f ϕ 1 = f ϕ 2 ϕ 1 = ϕ 2 (2) N A 1, A 2, A 3,... N A n X N n X N, A n N n=1 1 A1 d (d 2) A (, k A k = O), A O. f

More information

(Compton Scattering) Beaming 1 exp [i (k x ωt)] k λ k = 2π/λ ω = 2πν k = ω/c k x ωt ( ω ) k α c, k k x ωt η αβ k α x β diag( + ++) x β = (ct, x) O O x

(Compton Scattering) Beaming 1 exp [i (k x ωt)] k λ k = 2π/λ ω = 2πν k = ω/c k x ωt ( ω ) k α c, k k x ωt η αβ k α x β diag( + ++) x β = (ct, x) O O x Compton Scattering Beaming exp [i k x ωt] k λ k π/λ ω πν k ω/c k x ωt ω k α c, k k x ωt η αβ k α x β diag + ++ x β ct, x O O x O O v k α k α β, γ k γ k βk, k γ k + βk k γ k k, k γ k + βk 3 k k 4 k 3 k

More information

015 015 1 1 1 4 1.1.................................... 4 1.... 5 1.3... 6 1.4... 8 1.5... 8 10.1...................................... 10.... 10.3.................................... 10.4 1... 11.5...

More information

2

2 1 2 3 4 5 6 7 2007 30,870 2008 32,426 2009 34,971 13,000 8 9 http://www.hokkaido-marathon.com/volunteer/ http://www.hokkaido-marathon.com/volunteer/leader.html 2009 / http://www.shonan-kokusai.jp/archives/volunteer/

More information

記号と準備

記号と準備 tbasic.org * 1 [2017 6 ] 1 2 1.1................................................ 2 1.2................................................ 2 1.3.............................................. 3 2 5 2.1............................................

More information

20 9 19 1 3 11 1 3 111 3 112 1 4 12 6 121 6 122 7 13 7 131 8 132 10 133 10 134 12 14 13 141 13 142 13 143 15 144 16 145 17 15 19 151 1 19 152 20 2 21 21 21 211 21 212 1 23 213 1 23 214 25 215 31 22 33

More information

bag-of-words bag-of-keypoints Web bagof-keypoints Nearest Neighbor SVM Nearest Neighbor SIFT Nearest Neighbor bag-of-keypoints Nearest Neighbor SVM 84

bag-of-words bag-of-keypoints Web bagof-keypoints Nearest Neighbor SVM Nearest Neighbor SIFT Nearest Neighbor bag-of-keypoints Nearest Neighbor SVM 84 Bag-of-Keypoints Web G.Csurka bag-of-keypoints Web Bag-of-keypoints SVM 5.% Web Image Classification with Bag-of-Keypoints Taichi joutou and Keiji yanai Recently, need for generic image recognition is

More information

main.dvi

main.dvi SGC - 70 2, 3 23 ɛ-δ 2.12.8 3 2.92.13 4 2 3 1 2.1 2.102.12 [8][14] [1],[2] [4][7] 2 [4] 1 2009 8 1 1 1.1... 1 1.2... 4 1.3 1... 8 1.4 2... 9 1.5... 12 1.6 1... 16 1.7... 18 1.8... 21 1.9... 23 2 27 2.1

More information

20 6 4 1 4 1.1 1.................................... 4 1.1.1.................................... 4 1.1.2 1................................ 5 1.2................................... 7 1.2.1....................................

More information

Date Wed, 20 Jun (JST) From Kuroki Gen Message-Id Subject Part 4

Date Wed, 20 Jun (JST) From Kuroki Gen Message-Id Subject Part 4 Part 4 2001 6 20 1 2 2 generator 3 3 L 7 4 Manin triple 8 5 KP Hamiltonian 10 6 n-component KP 12 7 nonlinear Schrödinger Hamiltonian 13 http//wwwmathtohokuacjp/ kuroki/hyogen/soliton-4txt TEX 2002 1 17

More information

1, 2, 2, 2, 2 Recovery Motion Learning for Single-Armed Mobile Robot in Drive System s Fault Tauku ITO 1, Hitoshi KONO 2, Yusuke TAMURA 2, Atsushi YAM

1, 2, 2, 2, 2 Recovery Motion Learning for Single-Armed Mobile Robot in Drive System s Fault Tauku ITO 1, Hitoshi KONO 2, Yusuke TAMURA 2, Atsushi YAM 1, 2, 2, 2, 2 Recovery Motion Learning for Single-Armed Mobile Robot in Drive System s Fault Tauku ITO 1, Hitoshi KONO 2, Yusuke TAMURA 2, Atsushi YAMASHITA 2 and Hajime ASAMA 2 1 Department of Precision

More information

syuu_2_10_3.dvi

syuu_2_10_3.dvi [1] [1, 2, 3] [1, 4, 5] 6 7 3 (0.66) (0.65) 1 [6] 0 1 1 2 3 2.1................................ 3 2.1.1.................................. 3 2.1.2.................................. 3 2.2...........................

More information

双対問題

双対問題 一次式とノルムで構成された最適化問題とその双対問題 第一部ゲージ最適化問題とその応用問題 2018 年 7 月 25 日 ( 本資料は 7 月 30 日に改訂 ) 京都大学大学院情報学研究科 山下信雄 本講演の概要 第 1 部一般化ゲージ最適化問題とその応用問題 第 2 部ゲージ関数とその諸性質 第 3 部一般化ゲージ最適化問題の双対問題 I. ラグランジュ双対問題とFenchel 双対問題 II.

More information

untitled

untitled . x2.0 0.5 0 0.5.0 x 2 t= 0: : x α ij β j O x2 u I = α x j ij i i= 0 y j = + exp( u ) j v J = β y j= 0 j j o = + exp( v ) 0 0 e x p e x p J j I j ij i i o x β α = = = + +.. 2 3 8 x 75 58 28 36 x2 3 3 4

More information

2.2 h h l L h L = l cot h (1) (1) L l L l l = L tan h (2) (2) L l 2 l 3 h 2.3 a h a h (a, h)

2.2 h h l L h L = l cot h (1) (1) L l L l l = L tan h (2) (2) L l 2 l 3 h 2.3 a h a h (a, h) 1 16 10 5 1 2 2.1 a a a 1 1 1 2.2 h h l L h L = l cot h (1) (1) L l L l l = L tan h (2) (2) L l 2 l 3 h 2.3 a h a h (a, h) 4 2 3 4 2 5 2.4 x y (x,y) l a x = l cot h cos a, (3) y = l cot h sin a (4) h a

More information

Kalman ( ) 1) (Kalman filter) ( ) t y 0,, y t x ˆx 3) 10) t x Y [y 0,, y ] ) x ( > ) ˆx (prediction) ) x ( ) ˆx (filtering) )

Kalman ( ) 1) (Kalman filter) ( ) t y 0,, y t x ˆx 3) 10) t x Y [y 0,, y ] ) x ( > ) ˆx (prediction) ) x ( ) ˆx (filtering) ) 1 -- 5 6 2009 3 R.E. Kalman ( ) H 6-1 6-2 6-3 H Rudolf Emil Kalman IBM IEEE Medal of Honor(1974) (1985) c 2011 1/(23) 1 -- 5 -- 6 6--1 2009 3 Kalman ( ) 1) (Kalman filter) ( ) t y 0,, y t x ˆx 3) 10) t

More information

dy + P (x)y = Q(x) (1) dx dy dx = P (x)y + Q(x) P (x), Q(x) dy y dx Q(x) 0 homogeneous dy dx = P (x)y 1 y dy = P (x) dx log y = P (x) dx + C y = C exp

dy + P (x)y = Q(x) (1) dx dy dx = P (x)y + Q(x) P (x), Q(x) dy y dx Q(x) 0 homogeneous dy dx = P (x)y 1 y dy = P (x) dx log y = P (x) dx + C y = C exp + P (x)y = Q(x) (1) = P (x)y + Q(x) P (x), Q(x) y Q(x) 0 homogeneous = P (x)y 1 y = P (x) log y = P (x) + C y = C exp{ P (x) } = C e R P (x) 5.1 + P (x)y = 0 (2) y = C exp{ P (x) } = Ce R P (x) (3) αy

More information

°ÌÁê¿ô³ØII

°ÌÁê¿ô³ØII July 14, 2007 Brouwer f f(x) = x x f(z) = 0 2 f : S 2 R 2 f(x) = f( x) x S 2 3 3 2 - - - 1. X x X U(x) U(x) x U = {U(x) x X} X 1. U(x) A U(x) x 2. A U(x), A B B U(x) 3. A, B U(x) A B U(x) 4. A U(x),

More information

icml10yomikai.key

icml10yomikai.key A Fast Augmented Lagrangian Algorithm for Learning Low- Rank Matrices Ryota Tomioka, Taiji Suzuki, Masashi Sugiyama, Hisashi Kashima 2010/8/18 ICML2010 1 Matrix completion [Srebro et al. 05; Abernethy

More information

第1章 微分方程式と近似解法

第1章 微分方程式と近似解法 April 12, 2018 1 / 52 1.1 ( ) 2 / 52 1.2 1.1 1.1: 3 / 52 1.3 Poisson Poisson Poisson 1 d {2, 3} 4 / 52 1 1.3.1 1 u,b b(t,x) u(t,x) x=0 1.1: 1 a x=l 1.1 1 (0, t T ) (0, l) 1 a b : (0, t T ) (0, l) R, u

More information

量子力学 問題

量子力学 問題 3 : 203 : 0. H = 0 0 2 6 0 () = 6, 2 = 2, 3 = 3 3 H 6 2 3 ϵ,2,3 (2) ψ = (, 2, 3 ) ψ Hψ H (3) P i = i i P P 2 = P 2 P 3 = P 3 P = O, P 2 i = P i (4) P + P 2 + P 3 = E 3 (5) i ϵ ip i H 0 0 (6) R = 0 0 [H,

More information

2 (March 13, 2010) N Λ a = i,j=1 x i ( d (a) i,j x j ), Λ h = N i,j=1 x i ( d (h) i,j x j ) B a B h B a = N i,j=1 ν i d (a) i,j, B h = x j N i,j=1 ν i

2 (March 13, 2010) N Λ a = i,j=1 x i ( d (a) i,j x j ), Λ h = N i,j=1 x i ( d (h) i,j x j ) B a B h B a = N i,j=1 ν i d (a) i,j, B h = x j N i,j=1 ν i 1. A. M. Turing [18] 60 Turing A. Gierer H. Meinhardt [1] : (GM) ) a t = D a a xx µa + ρ (c a2 h + ρ 0 (0 < x < l, t > 0) h t = D h h xx νh + c ρ a 2 (0 < x < l, t > 0) a x = h x = 0 (x = 0, l) a = a(x,

More information

k 0 given, k t 0. 1 β t U (Af (k t ) k t+1 ) ( 1)+β t+1 U (Af (k t+1 ) k t+2 ) Af (k t+1 ) = 0 (4) t=1,2,3,...,t-1 t=t terminal point k T +1 = 0 2 T k

k 0 given, k t 0. 1 β t U (Af (k t ) k t+1 ) ( 1)+β t+1 U (Af (k t+1 ) k t+2 ) Af (k t+1 ) = 0 (4) t=1,2,3,...,t-1 t=t terminal point k T +1 = 0 2 T k 2012 : DP(1) 24 5 6 1 (Dynamic Programming) (Dynamic Programming) Bellman Stokey and Lucas with Prescott(1987) 1.1 max {c t,k t+1 } T o T β t U (c t ) (1) subject to c t + k t+1 = Af (k t ), (2) k 0 given,

More information

Lecture note 10: II Osaka Institute of Technology

Lecture note 10: II Osaka Institute of Technology Lecture note 10: II Osaka Institute of Technology , 2002. MATLAB, 1998. 1991. G. Goodwin, et al., Control System Design, Prentice Hall, New Jersey, 2001. Osaka Institute of Technology II 2 ẋ = Ax Bu y

More information

I A A441 : April 15, 2013 Version : 1.1 I Kawahira, Tomoki TA (Shigehiro, Yoshida )

I A A441 : April 15, 2013 Version : 1.1 I   Kawahira, Tomoki TA (Shigehiro, Yoshida ) I013 00-1 : April 15, 013 Version : 1.1 I Kawahira, Tomoki TA (Shigehiro, Yoshida) http://www.math.nagoya-u.ac.jp/~kawahira/courses/13s-tenbou.html pdf * 4 15 4 5 13 e πi = 1 5 0 5 7 3 4 6 3 6 10 6 17

More information

Introduction of Self-Organizing Map * 1 Ver. 1.00.00 (2017 6 3 ) *1 E-mail: furukawa@brain.kyutech.ac.jp i 1 1 1.1................................ 2 1.2...................................... 4 1.3.......................

More information

newmain.dvi

newmain.dvi 数論 サンプルページ この本の定価 判型などは, 以下の URL からご覧いただけます. http://www.morikita.co.jp/books/mid/008142 このサンプルページの内容は, 第 2 版 1 刷発行当時のものです. Daniel DUVERNEY: THÉORIE DES NOMBRES c Dunod, Paris, 1998, This book is published

More information

: 1g99p038-8

: 1g99p038-8 16 17 : 1g99p038-8 1 3 1.1....................................... 4 1................................... 5 1.3.................................. 5 6.1..................................... 7....................................

More information

数値計算:常微分方程式

数値計算:常微分方程式 ( ) 1 / 82 1 2 3 4 5 6 ( ) 2 / 82 ( ) 3 / 82 C θ l y m O x mg λ ( ) 4 / 82 θ t C J = ml 2 C mgl sin θ θ C J θ = mgl sin θ = θ ( ) 5 / 82 ω = θ J ω = mgl sin θ ω J = ml 2 θ = ω, ω = g l sin θ = θ ω ( )

More information

離散最適化基礎論 第 11回 組合せ最適化と半正定値計画法

離散最適化基礎論 第 11回  組合せ最適化と半正定値計画法 11 okamotoy@uec.ac.jp 2019 1 25 2019 1 25 10:59 ( ) (11) 2019 1 25 1 / 38 1 (10/5) 2 (1) (10/12) 3 (2) (10/19) 4 (3) (10/26) (11/2) 5 (1) (11/9) 6 (11/16) 7 (11/23) (11/30) (12/7) ( ) (11) 2019 1 25 2

More information

ad bc A A A = ad bc ( d ) b c a n A n A n A A det A A ( ) a b A = c d det A = ad bc σ {,,,, n} {,,, } {,,, } {,,, } ( ) σ = σ() = σ() = n sign σ sign(

ad bc A A A = ad bc ( d ) b c a n A n A n A A det A A ( ) a b A = c d det A = ad bc σ {,,,, n} {,,, } {,,, } {,,, } ( ) σ = σ() = σ() = n sign σ sign( I n n A AX = I, YA = I () n XY A () X = IX = (YA)X = Y(AX) = YI = Y X Y () XY A A AB AB BA (AB)(B A ) = A(BB )A = AA = I (BA)(A B ) = B(AA )B = BB = I (AB) = B A (BA) = A B A B A = B = 5 5 A B AB BA A

More information

数学の基礎訓練I

数学の基礎訓練I I 9 6 13 1 1 1.1............... 1 1................ 1 1.3.................... 1.4............... 1.4.1.............. 1.4................. 3 1.4.3........... 3 1.4.4.. 3 1.5.......... 3 1.5.1..............

More information

基礎数学I

基礎数学I I & II ii ii........... 22................. 25 12............... 28.................. 28.................... 31............. 32.................. 34 3 1 9.................... 1....................... 1............

More information

THE INSTITUTE OF ELECTRONICS, INFORMATION AND COMMUNICATION ENGINEERS TECHNICAL REPORT OF IEICE. TV A310

THE INSTITUTE OF ELECTRONICS, INFORMATION AND COMMUNICATION ENGINEERS TECHNICAL REPORT OF IEICE. TV A310 THE INSTITUTE OF ELECTRONICS, INFORMATION AND COMMUNICATION ENGINEERS TECHNICAL REPORT OF IEICE. TV 367 0035 1011 A310 E-mail kawamura@suou.waseda.jp Total Variation Total Variation Total Variation Abstract

More information

/ 2 n n n n x 1,..., x n 1 n 2 n R n n ndimensional Euclidean space R n vector point R n set space R n R n x = x 1 x n y = y 1 y n distance dx,

/ 2 n n n n x 1,..., x n 1 n 2 n R n n ndimensional Euclidean space R n vector point R n set space R n R n x = x 1 x n y = y 1 y n distance dx, 1 1.1 R n 1.1.1 3 xyz xyz 3 x, y, z R 3 := x y : x, y, z R z 1 3. n n x 1,..., x n x 1. x n x 1 x n 1 / 2 n n n n x 1,..., x n 1 n 2 n R n n ndimensional Euclidean space R n vector point 1.1.2 R n set

More information

BIS CDO CDO CDO CDO Cifuentes and O Connor[1] Finger[6] Li[8] Duffie and Garleânu[4] CDO Merton[9] CDO 1 CDO CDO CDS CDO three jump model Longstaff an

BIS CDO CDO CDO CDO Cifuentes and O Connor[1] Finger[6] Li[8] Duffie and Garleânu[4] CDO Merton[9] CDO 1 CDO CDO CDS CDO three jump model Longstaff an CDO 2010 5 18 CDO(Collateralized Debt Obligation) Duffie and Garleânu[4] CDO CDS(Credit Default Swap) Duffie and Garleânu[4] 4 CDO CDS CDO CDS CDO 2007 CDO CDO CDS 1 1.1 2007 2008 9 15 ( ) CDO CDO 80 E-mail:taiji.ohka@gmail.com

More information

1 ( ) Einstein, Robertson-Walker metric, R µν R 2 g µν + Λg µν = 8πG c 4 T µν, (1) ( ds 2 = c 2 dt 2 + a(t) 2 dr 2 ) + 1 Kr 2 r2 dω 2, (2) (ȧ ) 2 H 2

1 ( ) Einstein, Robertson-Walker metric, R µν R 2 g µν + Λg µν = 8πG c 4 T µν, (1) ( ds 2 = c 2 dt 2 + a(t) 2 dr 2 ) + 1 Kr 2 r2 dω 2, (2) (ȧ ) 2 H 2 1 ( ) Einstein, Robertson-Walker metric, R µν R 2 g µν + Λg µν = 8πG c 4 T µν, (1) ( ds 2 = c 2 dt 2 + a(t) 2 dr 2 ) + 1 Kr 2 r2 dω 2, (2) (ȧ ) 2 H 2 = 8πG a 3c 2 ρ Kc2 a 2 + Λc2 3 (3), ä a = 4πG Λc2 (ρ

More information

T rank A max{rank Q[R Q, J] t-rank T [R T, C \ J] J C} 2 ([1, p.138, Theorem 4.2.5]) A = ( ) Q rank A = min{ρ(j) γ(j) J J C} C, (5) ρ(j) = rank Q[R Q,

T rank A max{rank Q[R Q, J] t-rank T [R T, C \ J] J C} 2 ([1, p.138, Theorem 4.2.5]) A = ( ) Q rank A = min{ρ(j) γ(j) J J C} C, (5) ρ(j) = rank Q[R Q, (ver. 4:. 2005-07-27) 1 1.1 (mixed matrix) (layered mixed matrix, LM-matrix) m n A = Q T (2m) (m n) ( ) ( ) Q I m Q à = = (1) T diag [t 1,, t m ] T rank à = m rank A (2) 1.2 [ ] B rank [B C] rank B rank

More information

ばらつき抑制のための確率最適制御

ばらつき抑制のための確率最適制御 ( ) http://wwwhayanuemnagoya-uacjp/ fujimoto/ 2011 3 9 11 ( ) 2011/03/09-11 1 / 46 Outline 1 2 3 4 5 ( ) 2011/03/09-11 2 / 46 Outline 1 2 3 4 5 ( ) 2011/03/09-11 3 / 46 (1/2) r + Controller - u Plant y

More information