nl226ithmm.dvi
|
|
- ひろと えいさか
- 5 years ago
- Views:
Transcription
1 Markov 1,a),b) Markov (HMM),, Stick-breaking (Adams+ 010), Markov (ithmm),,, Markov, PCFG Stick-breaking,, The Infinite Tree Hidden Markov Model for Unsupervised Hierarchical Part-of-speech Induction Daichi Mochihashi 1,a) Hiroshi Noji,b) Abstract: Hidden Markov models (HMM) is widely used in statistics and machine learning However, it cannot learn latent states where these states are actually structured Extending the tree-structured stickbreaking processes (Adams+ 010) hierarchically as from DP to HDP, this paper proposes an Infinite Tree Hidden Markov models (ithmm) whose states constitute a latent hierarchy Experimental results on natural language texts show the validity of the proposed algorithm Keywords: Tree-structured stick-breaking process, Hidden Markov models, Nonparametric Bayes, Unsupervised learning 1 Markov (HMM)[1] HMM K ( HMM K ),,,,,,, [],, HMM, HMM,, ( ) HMM, 90 [3][4], 000 [5],, Markov [6][7] s k {1K}, 1M k, HMM, M 1 M K, K k=1 M k, 1, The Institute of Statistical Mathematics Nara Institute of Science and Technology a) daichi@ismacjp b) noji@isnaistjp [8] 010 [9],,,,, [10][11] c 016 Information Processing Society of Japan 1
2 情報処理学会研究報告 ることで, より高精度な学習を可能にしている しかし, こ の場合でも細分化は上で述べた問題から 1 段階に限られて おり, また既知の品詞体系を必要とする 未知の言語を解 析する場合や, たとえば動詞句と形容詞句がより上の階層 で統合されるような可能性も考えると, 計算言語学の立場 からは, こうした品詞階層自体を言語データから学習でき る統計的枠組が求められているといえる そこで本論文では, ノンパラメトリックベイズ法の立場か = 基底測度 G0 G DP(α, G0 ) 図 ディリクレ過程による基底測度 G0 からの G の生成 ルであり, HMM は初期の形態素解析 (茶筌) に使われたほ ら上の問題をすべて解決し, 隠れ状態が無限の分岐と無限の か, 現在でも半教師あり学習に用いられている [8] モデル (ithmm) および, それに基づいた階層的な品詞の ム) によっており性能が低いとみなされていたが [3], Gold- 深さをもつ木構造上で定義される無限木構造隠れ Markov 教師なし学習法を提案する 提案法はディリクレ過程が木 の縦方向の深さおよび横方向のそれぞれの分岐に存在する 木構造 Stick-breaking 過程 [1] をそれ自体無限木構造上で 階層化したものであり, こうして得られる無限木構造上の 状態遷移確率と, この上で拡散過程として生成される出力 確率分布によって観測系列が生成される この ithmm は 自然言語処理に限らず, 情報科学一般に適用できる HMM の本質的な拡張であり, 多くの分野での適用が期待できる 以下, 章で提案法の基礎となる無限隠れ Markov モデ ルおよびディリクレ過程, その具体的実現である Stickbreaking 過程について説明する 3 章では木構造 Stickbreaking 過程 (TSSB) とそのポリアの壷表現について説明 し, 4 章で TSSB を階層化した階層的木構造 Stick-breaking 過程 (HTSSB) とそれに基づいた無限木構造隠れ Markov モデルと特別な MCMC 法による学習について述べる 5 章 で HHMM などの関連研究との違いについて述べた後, 6 章 で日本語や英語のテキストに対して実験を行って優位性を 示し, 特に半教師あり学習に用いることも可能であること を示す 7 章で展望を示し, 全体をまとめる 無限隠れ Markov モデルと Stick-breaking 過程 HMM は図 1 のように, 観測列 w = w1 w wt の背後 に隠れ状態列 s = s1 s st があり, s から w が生成され たとする確率モデルである 1 次の HMM では時刻 t での 状態 st は一つ前の状態 st 1 のみに依存すると考え, w と s が生成される同時確率は p(w, s) = T Y p(wt st )p(st st 1 ) (1) t=1 で表される ただし, s0 は初期状態である 隠れ状態を名 詞や動詞のような品詞とみなすと, これは品詞学習のモデ 図 1 隠れ Markov モデルの構造 は観測値を, は未知の確率変 数を表す c 016 Information Processing Society of Japan 品詞の教師なし学習は最初は最尤推定 (EM アルゴリズ water ら [5] はこれを MCMC 法によりベイズ推定するこ とで, 局所解を避け, 高精度な解が得られることを示した これらの研究では状態数 品詞数 K は既知であるとして いるが, この K も学習できるのが無限隠れ Markov モデル (Infinite HMM, ihmm)[6][13] である 1 ihmm と HDP まず, HMM では生成モデルから, 状態は状態遷移確率 p(st st 1 ) によって生成されることに注意しよう 通常の HMM では, これは決まった K 個の状態への確率分布とな るが, ihmm では, これが可算無限個の要素を持つディリ クレ過程から生成されたと考える ディリクレ過程とは, 図 のように基底測度とよばれる親の分布 G0 に似た無限 次元の離散的な測度を生成する確率過程であり, G DP(α, G0 ) () と書かれる 集中度パラメータ α > 0 が大きいほど G は G0 に似たものとなるが, 期待値は常に E[G] = G0 である ただし, 各状態 k で別々にこの遷移確率 Gk を G0 から サンプルすると, 他の状態との重なりが 0 になってしまい, HMM の状態が共有されなくなってしまう そこで, ihmm ではまず全体の離散的な G DP(η, H) をサンプルし, こ れを基底測度として各 Gk DP(α, G) (k = 1 ) を生 成する階層ディリクレ過程 (HDP) によって, 遷移する状態 を共有し, その事前分布を G で与える このとき α によっ て, Gk が事前分布 G と平均的にどれほど似ているかが制 御されることになる Stick-breaking 過程と CDP 表現 上ではディリクレ過程およびそれに基づく ihmm の構 成を測度論的に述べた よく知られているように, ディリ クレ過程に基づく G DP(α, G0 ) からのサンプルは図 3 のような CRP(中国料理店過程) で表すことができる [14] ここでは, G からのサンプル x1, x,, xn が与えられた とき, 次の xn+1 のとる値の確率は G を積分消去すること により, Z p(xn+1 x1 xn ) = p(xn+1 G) p(g x1 xn )dg (3) n /(n+α) (k = 1,, K) k = α/(n+α) G0 (xn+1 ) (k = K +1) (4)
3 K x 1 x n, n k k, G 3, (4) k π 1 γ 3 CRP π 3 CRP G,, 4 Stick-breaking π = (π 1,π, ) 1 Stick-breaking( ) γ k Be(1,α), [15] π γ k Be(1,α) (5) k 1 π k = γ k (1 γ j ) (6) G = π k δ(θ k ), θ k G 0 (7) k=1 5 Chinese District Process (CDP) [16] Be(α,β), δ(x) x 1, 4, 1, SBP γ k (k = 1,,3, ), π k G 0, k θ k G, Chinese District Process (CDP) ( ) {θ k } k=1, G [16], SBP CRP π = (π 1,π, ), GEM, 3 SBP(α) SBP G, HMM CDP SBP γ k, [16], π D = {x 1,x, } CDP HMM, Gibbs, π γ k, Gael [17],(7),(6) k π k, x n 1 k 1, Markov [7] 1, k, γ k ihmm (1431, 6689 ) D k n 0 (k),, n 1 (k), Be(1+n 0 (k),α+n 1 (k)), E[γ k D] = 1+n 0 (k) 1+α+n 0 (k)+n 1 (k), π k E[π k D] = (8) k 1 1+n 0 (k) α+n 1 (j) 1+α+n 0 (k)+n 1 (k) 1+α+n 0 (j)+n 1 (j) (9) 3 CRP( ) π 1 γ 1 γ γ3 k=1 k= k=3 k=4 1 γ 1 1 γ, K 7, 1=, =, 3=,,, HMM,,,,, * 1 π k, 5 1,, 3,, (, ) (8),,,,, Stick-breaking [1] *1,, [18], AR c 016 Information Processing Society of Japan 3
4 1 she 43 to 387 i 34 it 65 you 18 alice 166 and 147 they 76 there 61 he 55 that 39 who 37 4 and 466 of 343 in 6 said 174 to 163 as 163 that 15 for 13 at 1 but 11 with 114 on 83 the 106 a 473 her 116 very 84 its 50 my 46 no 44 his 44 this 39 an 37 your 36 as 31 5 way 45 mouse 41 thing 39 queen 37 head 36 cat 35 hatter 34 duchess 34 well 31 time 31 tone 8 rabbit 8 3 was 77 had 16 said 113 be 77 is 73 went 58 were 56 see 5 could 5 know 50 thought 44 herself 4 6 little 9 great 3 very long large right 0 same 17 good 17 white 11 other 11 poor 10 first 10 1 ihmm 31 TSSB TSSB, 3 Stick-breaking ν s Be(1,α) ν [] 1 ν [], [], [1],[],[3], [4], [4] SBP π k SBP, ν [4],, [4 1], [4 ], [4 ] ν Polya [0] [4,, [4 ], TSSB, π s TSSB π s Nested CRP [1][] Nested HDP [3], s s s s DP,,, π s = ν s φ s φ s (1 ν s ) (11), = ν s (1 ν s ) φ s (1), TSSB,, ν, 1 s Be(1,α), ψ sk Be(1,γ) (13) k 1 ν Be(1,α) (10) φ sk = ψ sk (1 ψ sj ) (14) π [] φ [1] = ψ [1] π [1] SBP(α) φ [] = ψ [] (1 ψ [1] ) 6 ν s,ψ s TSSB π,, (1 ν) SBP(γ), ( 7) TSSB, s = s 1 s s 3, s = [] ( ), s = [ 4 1] 4,, 3,, s ψ s Be(1,γ) [], ν Stick-breaking (Tree-structured Stick-brea-, king process, TSSB) [1], SBP(α),, CDP [1],[],[3],, ψ [k], TSSB, [k],, ν, Markov [19] [k], s s s s s s c 016 Information Processing Society of Japan 4
5 p(d n s) ρ TSSB(1, 1, 1) TSSB(5, 0, 10) 0 r 1 r 9 Slice sampling+retrospective sampling TSSB MCMC r TSSB s, p(d n s) CDP, TSSB(, 0, ) TSSB(1, 05, 5) 7 TSSB(α 0,λ,γ), CDP s Stick-breaking n 1 (s), n 0 (s), CDP, 1 s m 1 (s), m 0 (1) SBP (6), ν 1+n 0 (s) E[ν SBP, ψ s D] = (16) 1+α+n 0 (s)+n 1 (s) 1+m 0 (s) E[ψ sk D] = (17) 7, TSSB 1+γ+m 0 (s)+m 1 (s) (1),, π s, (13) [1] SBP CRP α [1] α(s) = α 0 λ s (15), CDP 33, 0 < λ 1 TSSB, N, s s d 1 d N, (15), TSSB s n Gibbs, 10, TSSB (α 0,λ,γ), s CDP s 1,,s n, s CDP s, 8,, ν s, ψ sk p(s n d n ) p(d n s)π s s n 7, TSSB, 3 TSSB CDP s, TSSB (1) SBP,TSSB, TSSB CDP 7, s π s [0,1), s = s 1 s s n, π s s, r = Unif[0,1), TSSB r, Retrospective sampling [4] 9 p(d n s)π s s n, s n p(d n s n )π sn 0 ρ, p(d n s)π s > ρ s 8 TSSB CDP CDP, Retrospective sampling CDP / s, d n c 016 Information Processing Society of Japan 5
6 [0,1),, HDP Stick-breaking, π 1 [1] γ k (k = 1,, ) 4 Markov Stick-breaking, [13], s ν,ψ (, ν s Be (αν s Markov,α 1 )) u u sν, (0) ( k )) ψ sk Be (αψ sk,α 1 ψ sj (1) 41, HMM, ν s, ψ sk TSSB ν s,, ψ sk,(13) K HMM, K (16) (17), K K αν s, E[ν s D] = +n 0(s) α(1 u s ν u )+n () 0(s)+n 1 (s), αψ sk E[ψ sk D] = +m 0(sk) α(1 k 1 TSSB,, ψ sj )+m (3) 0(sk)+m 1 (sk) TSSB s, TSSB ν s, ψ sk, π s ( 11) TSSB,, (16)(17) π s,, s = [ 3], n = n 0 +n 1, βk l = l j=k β j, E[γ [ 3] π [ k n 0,n 1 ] = αβ k+n 0 [], αβk +n (4) π [], = αβ k π [] (, αβk +n αβ k n n0 + (5) +n n ) = µ ˆp+(1 µ) p (6) 4 TSSB, π s ˆp = n 0, n, p = β k β (7) k TSSB π s n µ = 31, TSSB αβk +n (8) Stick-breaking,, Bernoulli, π s DP ˆp TSSB p µ, π s DP n µ, π = SBP(γ), (6) SBP β = (β 1,β, ), 1: for iter = 1 iters do : for n in randperm(1 N) do 3: p(d n s n ) d n, π s n 4: Draw s n p(d n s n )π sn 5: p(d n s n ) d n, π s n 6: end for 7: end for π DP(α,β) (18) ( ( γ k Be αβ k,α 1 k )) β j * [5] TSSB 10 TSSB Gibbs, HDP (19) ν s, ψ sk, TSSB ν s, ψ sk,, HDP (19) αβ k αβ k, α,, π π, π (16)(17), Stick-breaking (HTSSB) [5] *, c 016 Information Processing Society of Japan 6
7 TSSB HCDP( CDP) 11 HTSSB TSSB,, TSSB,, TSSB Bernoulli n() 0 π HTSSB(α,π 0 ) (9), HTSSB, Markov HTSSB-HMM, ithmm (Infinite, Tree HMM, Markov ), TSSB ithmm ithmm, HTSSB, π [] TSSB(α 0,λ,γ) n(s) n(s)+α, ] α n(s)+α ν s (30), TSSB ψ SBP CRP, s, m(s) = m 0 (s)+m 1 (s) [ m(s) m(s)+α, ] α m(s)+α ψ s Bernoulli *3,, *3,, +1 (31) Bernoulli, 44 ithmm,,, π [1] HTSSB(α,π [] ), ithmm π [], π [1 1] s t [4 3], HMM HTSSB(α,π [1] ), π [1 ] 1, Gibbs s 0, (1) HMM, ( ) w t, s 1,s,s 3,, s t w 1,w,w 3, p(s t w t,w t,s t ) p(w t s t )p(s t+1 s t )p(s t s t 1 ) (3), HTSSB π, [5] 1 TSSB s t w t, HDP Pitman-Yor, 3 HTSSB, TSSB s, w t w t, s t, TSSB s t,, ithmm s t [s 1 s s 3 ], HMM 1 K, ihmm 43 ithmm HCDP TSSB 3 CDP, s t 7 [0,1) ithmm, HTSSB, (3) HTSSB, TSSB, 33 Retrospective sampling Slice sampling s t [0,1) CRP, r Unif[0,1), ν SBP k, TSSB s, (3),, p(s t s t 1 ) (4), s SBP s t, p(w t s t )p(s t+1 s t ) n(s) = n 0 (s)+n 1 (s), 33, [ p(w t s t ), p(w t s t )p(s t+1 s t ) c 016 Information Processing Society of Japan 7
8 1: function draw state (s t 1,s t,s t+1,w t ) : slice = p(w t s t )p(s t+1 s t ) Unif[0,1) 3: st := 0;ed := 1 4: while true do 5: u := Unif[st,ed) 6: s := s t 1 TSSB find node(u) 7: p := p(w t s)p(s t+1 s) 8: if p > slice then 9: return s 10: else 11: if s < s t then 1: st := u 13: else 14: ed := u 15: end if 16: end if 17: end while (1) s t, ρ = p(w t s t )p(s t+1 s t ) Unif[0,1) st = 0,ed = 1 () r Unif[st,ed), p(s t s t 1 ) TSSB s t [8], HCDP (3) p(w t s t )p(s t+1 s t ) > ρ,, ±1 MH st,ed () p(s t s t 1 ) s t HCDP p(s t+1 s t ), 1 s, MH 9999% p( s), s, p( s ) p( s), EOS TSSB, s EOS 1, s EOS q s = p(eos s) q s q s Be(τ 0,τ 1 ) (35), s EOS c 0 (s), c 1 (s), q s E[q s c 0 (s),c 1 (s)] = τ 0 +c 0 (s) τ 0 +τ 1 +c 0 (s)+c 1 (s) [1], s HMM (Hierarchical, µ s N(µ s,σ HMM, HHMM) [9], ) [30], HMM, HMM,, G s = {p( s)} Pitman-Yor [6] 1: for iter = 1 iters do G s HPY(G s,d s,θ s ) (33) : for n in randperm(1 N) do *4 3: remove (w, t, s t 1, s t, s t+1 ) 4: Draw s t = draw state(w t,s t 1,s t,s t+1 ) 5: if MH-accept(s t, s t) then EOS, 6: s t = s t EOS, 7: end if 8: add (w HMM t, s t 1, s t, s t+1 ) 9: end for 0 EOS, 1 10: end for, ithmm [] 11: function add (w t, s t 1, s t, s t+1 ) 1: s t add customer (w t ) 13: s, EOS t 1 add customer(s t ) 14: s t add customer(s t+1 ) *4, κ 15: function remove (w t, s t 1, s t, s t+1 ) 16: s t remove customer (w t ) G s HPY(κG 0+(1 κ)g s,d s,θ s ) (34) 17: s t remove customer(s t+1 ) 18: s t 1 remove customer(s t ) [1], d,θ, G 0 1/V 13 ithmm Gibbs (36), (τ 0,τ 1 ) = (1,100) (1 q s ) TSSB,, [7] 1 ithmm s t u < s u s,ithmm 13 TSSB find node(u) [0,1) u (3) p(s TSSB, [1] t s t 1 ) p(s t+1 s t ),, s t+1 =s t p(s t s t 1 ) 1 p(s t+1 s), Metropolis-Hastings 5 c 016 Information Processing Society of Japan 8
9 ihmm [] M 1 ithmm λ, PPL γ= ihmm γ= γ= γ= M= ithmm λ = λ= wish ( ), wonder , HMM tell see take do talk , HMM HHMM, 6 next one that mind two indeed round bill [ 3] know think say [0 0] don t could are can would must might should [ 7] be have go remember do get [4] [4 0] mock voice , Jordan Markov queen way (HMDT) [31], HMDT HHMM gryphon tone hatter thing , mouse side , HMDT duchess bit , caterpillar face ,, cat cat MCMC 3 1, [], C , Xeon 37GHz 1, 6, , , 31, 4,, 6,, ithmm, Viterbi ( ) 63 1 HMM,, ithmm 5, *5, 4, 3 0 8, 4 1 9, , , 9,, *5 Klingon Hamlet c 016 Information Processing Society of Japan 9
10 [ 0 0] el mej Ha joh nadev wa Hegh [3] Merialdo, B: Tagging English Text with a Probabilis- [1] tugh [1 1] DaH *Hamlet* vaj ta reh not tugh jihvad jihvad *polonyus* chich eh yo tic Model, Computational linguistics, Vol 0, No, pp (1994) [4] Kupiec, J: Robust part-of-speech tagging using a hidden Markov model, Computer Speech & Language, Vol 6, No 3, pp 5 4 (199) [5] Goldwater, S and Griffiths, T: A Fully Bayesian Approach to Unsupervised Part-of-Speech Tagging, Proceedings of ACL 007, pp (007) [6] Beal, M J, Ghahramani, Z and Rasmussen, C E: The [ 1] Infinite Hidden Markov Model, NIPS 001, pp vaj (001) je [7] Van Gael, J, Saatci, Y, Teh, Y W and Ghahramani, po Z: Beam sampling for the infinite hidden Markov model, pol ICML 008, pp (008) vida [8] Suzuki, J and Isozaki, H: Semi-Supervised Sequential ta be nal Labeling and Segmentation Using Giga-Word Scale Unlabeled Data, ACL:HLT 008, pp (008) jabbi ID [9] Christodoulopoulos, C, Goldwater, S and Steedman, 3,733, 19,97 M: Two decades of unsupervised POS induction: How, DaH(=now), vaj(=then) far have we come?, EMNLP 010, pp (010), el(=go), mej(=leave) [10] Matsuzaki, T, Miyao, Y and Tsujii, J: Probabilistic CFG with latent annotations, ACL 005, pp 75 8, (005) [11] Shindo, H, Miyao, Y, Fujino, A and Nagata, M: Bayesian Symbol-Refined Tree Substitution Grammars 7 for Syntactic Parsing, ACL 01, pp (01) [1] Adams, R P, Ghahramani, Z and Jordan, M I: Tree-, Structured Stick Breaking for Hierarchical Data, NIPS, 010, pp 19 7 (010) [13] Teh, Y W, Jordan, M I, Beal, M J and Blei, Stick-breaking [1] Stickbreaking, No 476, pp (006) D M: Hierarchical Dirichlet Processes, JASA, Vol 101, Pitman-Yor, [14] Blackwell, D and MacQueen, J B: Ferguson Distributions via Pólya Urn Schemes, Annals of Statistics, Vol 1, HMM, No, pp (1973) Markov (ithmm) [15] Sethuraman, J: A Constructive Definition of Dirichlet, TSSB Priors, Statistica Sinica, Vol 4, pp (1994) [16] Paisley, J and Carin, L: Hidden Markov models with stick-breaking priors, IEEE Transactions on Signal Processing, Vol 57, pp (009) Gibbs, HMM Beam, [17] Neal, R M: Slice sampling, Annals of Statistics, pp, (003), [18] Hjort, N L, Holmes, C, Müller, P and Walker, S G: Bayesian Nonparametrics, Cambridge University Press [0,1), (010) [19] Mochihashi, D and Sumita, E: The Infinite Markov Neal Embedded HMM [3] Model, Advances in Neural Information Processing Systems 0 (NIPS 007), pp (008) [0] Mauldin, R D, Sudderth, W D and Williams, S C: Polya Trees and Random Distributions, Annals of Statistics, Vol 0, No 3, pp (199), MCMC [1] Blei, D M, Griffiths, T L and Jordan, M I: The Nested Chinese Restaurant Process and Bayesian, Nonparametric Inference of Topic Hierarchies, JACM,, Vol 57, No, pp 1 30 (010) [1] Rabiner, L R: A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition, Proceedings of the IEEE, Vol 77, No, pp 57 86(1989) [] Bishop, C M: Pattern Recognition and Machine Learning, Information Science and Statistics, Springer (007) [] Ahmed, A, Hong, L and Smola, A: Nested Chinese Restaurant Franchise Processes: Applications to User Tracking and Document Modeling, ICML 013, pp (013) [3] Paisley, J, Wang, C, Blei, D and Jordan, M I: Nested hierarchical Dirichlet processes, IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol 37, No, pp (015) [4] Papaspiliopoulos, O and Roberts, G O: Retrospective c 016 Information Processing Society of Japan 10
11 Markov chain Monte Carlo methods for Dirichlet process hierarchical models, Biometrika, Vol 95, No 1, pp (008) [5] Noji, H, Mochihashi, D and Miyao, Y: Hierarchical Tree-Structured Stick-Breaking Priors, NIPS 013 workshop: Modern Nonparametric Methods in Machine Learning (013) [6] Teh, Y W: A Bayesian Interpretation of Interpolated Kneser-Ney, Technical Report TRA/06, School of Computing, NUS (006) [7] Minka, T: The Dirichlet-tree distribution (1999) minka/papers/ dirichlet/minka-dirtreepdf [8] Johnson, M, Griffiths, T L and Goldwater, S: Bayesian Inference for PCFGs via Markov Chain Monte Carlo, Proceedings of HLT/NAACL 007, pp (007) [9] Fine, S, Singer, Y and Tishby, N: The Hierarchical Hidden Markov Model: Analysis and Applications, Machine Learning, Vol 3, No 1, pp 41 6 (1998) [30] Heller, K A, Teh, Y W and Görür, D: Infinite Hierarchical Hidden Markov Models, AISTATS 009, pp 4 31 (009) [31] Jordan, M I, Ghahramani, Z and Saul, L K: Hidden Markov decision trees, Advances in Neural Information Processing Systems (1997), pp (1997) [3] Neal, R M, Beal, M J and Roweis, S T: Inferring state sequences for non-linear systems with embedded hidden Markov models, Advances in Neural Information Processing Systems 16 (004), pp (004) 6 [] OOV [0 0] [0 1] OOV [3 1] OOV [5] OOV [5 3] [11] OOV [0] OOV [0 0 0] [0 1 ] OOV [3 1 6] OOV [5 0] OOV [5 5] OOV [11 0 1] c 016 Information Processing Society of Japan 11
21 Pitman-Yor Pitman- Yor [7] n -gram W w n-gram G Pitman-Yor P Y (d, θ, G 0 ) (1) G P Y (d, θ, G 0 ) (1) Pitman-Yor d, θ, G 0 d 0 d 1 θ Pitman-Yor G
ol2013-nl-214 No6 1,a) 2,b) n-gram 1 M [1] (TG: Tree ubstitution Grammar) [2], [3] TG TG 1 2 a) ohno@ilabdoshishaacjp b) khatano@maildoshishaacjp [4], [5] [6] 2 Pitman-Yor 3 Pitman-Yor 1 21 Pitman-Yor
More informationDirichlet process mixture Dirichlet process mixture 2 /40 MIRU2008 :
Dirichlet Process : joint work with: Max Welling (UC Irvine), Yee Whye Teh (UCL, Gatsby) http://kenichi.kurihara.googlepages.com/miru_workshop.pdf 1 /40 MIRU2008 : Dirichlet process mixture Dirichlet process
More information? (EM),, EM? (, 2004/ 2002) von Mises-Fisher ( 2004) HMM (MacKay 1997) LDA (Blei et al. 2001) PCFG ( 2004)... Variational Bayesian methods for Natural
SLC Internal tutorial Daichi Mochihashi daichi.mochihashi@atr.jp ATR SLC 2005.6.21 (Tue) 13:15 15:00@Meeting Room 1 Variational Bayesian methods for Natural Language Processing p.1/30 ? (EM),, EM? (, 2004/
More information& 3 3 ' ' (., (Pixel), (Light Intensity) (Random Variable). (Joint Probability). V., V = {,,, V }. i x i x = (x, x,, x V ) T. x i i (State Variable),
.... Deeping and Expansion of Large-Scale Random Fields and Probabilistic Image Processing Kazuyuki Tanaka The mathematical frameworks of probabilistic image processing are formulated by means of Markov
More informationPowerPoint プレゼンテーション
ノンパラメトリックベイズによるメディア処理 202.. 5 AI チャレンジ研究会 NTT コミュニケーション科学基礎研究所中野允裕 nakano.masahiro@lab.ntt.co.jp 5,6 年前であれば 教科書に載っているような各種ツールのノンパラベイズ化が話題の中心になっていたが 主成分分析 非負値行列分解 確率文脈自由文法 独立成分分析 隠れマルコフモデル n-gram ダイナミックベイジアンネット
More informationIPSJ SIG Technical Report Pitman-Yor 1 1 Pitman-Yor n-gram A proposal of the melody generation method using hierarchical pitman-yor language model Aki
Pitman-Yor Pitman-Yor n-gram A proposal of the melody generation method using hierarchical pitman-yor language model Akira Shirai and Tadahiro Taniguchi Although a lot of melody generation method has been
More informationNo. 3 Oct The person to the left of the stool carried the traffic-cone towards the trash-can. α α β α α β α α β α Track2 Track3 Track1 Track0 1
ACL2013 TACL 1 ACL2013 Grounded Language Learning from Video Described with Sentences (Yu and Siskind 2013) TACL Transactions of the Association for Computational Linguistics What Makes Writing Great?
More informationGraham Neubig ノンパラメトリックベイズ法 ノンパラメトリックベイズ法 Graham Neubig 2011 年 5 月 10 1
ノンパラメトリックベイズ法 Graham Neubig 2011 年 5 月 10 日 @NAIST 1 概要 ノンパラメトリックベイズ法について ベイズ法の基礎理論 サンプリングによる推論 サンプリングを利用した HMM の学習 有限 HMM から無限 HMM へ 近年の展開 ( サンプリング法 モデル化法 音声処理 言語処理のおける応用 基本は離散分布の教師なし学習 2 Non-parametric
More information0
Speedy & Simple Kenji, Yoshio, and Goro are good at English. They have their ways of learning. Kenji often listens to English songs and tries to remember all the words. Yoshio reads one English book every
More informationMicrosoft PowerPoint - Ishiguro_IBIS_presentation.pptx
2011/11/10 第 14 回情報論的学習理論ワークショップ NTT コミュニケーション科学基礎研究所石黒勝彦 SNSにおける友達コミュニティ抽出 オンラインショッピング履歴に基づくレコメンド 論文と著者の組み合わせによる研究トピック解析 SNS 上の友達関係は簡単に変化します CMが当たると突然商品が売れ出します 研究プロジェクトや異動で共著者は変わります フォ0 1 0 1 0 チローし0
More informationL1 What Can You Blood Type Tell Us? Part 1 Can you guess/ my blood type? Well,/ you re very serious person/ so/ I think/ your blood type is A. Wow!/ G
L1 What Can You Blood Type Tell Us? Part 1 Can you guess/ my blood type? 当ててみて / 私の血液型を Well,/ you re very serious person/ so/ I think/ your blood type is A. えーと / あなたはとっても真面目な人 / だから / 私は ~ と思います / あなたの血液型は
More information1 IDC Wo rldwide Business Analytics Technology and Services 2013-2017 Forecast 2 24 http://www.soumu.go.jp/johotsusintokei/whitepaper/ja/h24/pdf/n2010000.pdf 3 Manyika, J., Chui, M., Brown, B., Bughin,
More informationIPSJ SIG Technical Report Vol.2014-CG-155 No /6/28 1,a) 1,2,3 1 3,4 CG An Interpolation Method of Different Flow Fields using Polar Inter
,a),2,3 3,4 CG 2 2 2 An Interpolation Method of Different Flow Fields using Polar Interpolation Syuhei Sato,a) Yoshinori Dobashi,2,3 Tsuyoshi Yamamoto Tomoyuki Nishita 3,4 Abstract: Recently, realistic
More information< D8291BA2E706466>
A 20 1 26 20 10 10 16 4 4! 20 6 11 2 2 3 3 10 2 A. L. T. Assistant Language Teacher DVD AV 3 A. E. T.Assistant English Teacher A. L. T. 40 3 A 4 B A. E. T. A. E. T. 6 C 2 CD 4 4 4 4 4 8 10 30 5 20 3 5
More information3 4 26 1980 1 WWW 26! 3, ii 4 7!! 4 2010 8 1. 1.1... 1 1.2... 2 1.3... 3 1.4... 7 1.5... 9... 9 2. 2.1... 10 2.2... 13 2.3... 16 2.4... 18... 21 3. 3.1... 22 3.2... 24 3.3... 33... 38 iv 4. 4.1... 39 4.2...
More information28 Horizontal angle correction using straight line detection in an equirectangular image
28 Horizontal angle correction using straight line detection in an equirectangular image 1170283 2017 3 1 2 i Abstract Horizontal angle correction using straight line detection in an equirectangular image
More information<4D6963726F736F667420506F776572506F696E74202D2089708CEA8D758DC0814091E396BC8E8C8145914F92758E8C81458C6097658E8C81458F9593AE8E8C>
英 語 特 別 講 座 代 名 詞 前 置 詞 形 容 詞 助 動 詞 #1 英 語 特 別 講 座 2010 代 名 詞 前 置 詞 形 容 詞 助 動 詞 英 語 特 別 講 座 代 名 詞 前 置 詞 形 容 詞 助 動 詞 #2 代 名 詞 日 本 語 私 あなた 彼 のうしろに は の を に のもの をつけて 使 う どこに 置 くかは 比 較 的 自 由 私 はジャスコに 行 った ジャスコに
More informationIPSJ SIG Technical Report Vol.2010-CVIM-170 No /1/ Visual Recognition of Wire Harnesses for Automated Wiring Masaki Yoneda, 1 Ta
1 1 1 1 2 1. Visual Recognition of Wire Harnesses for Automated Wiring Masaki Yoneda, 1 Takayuki Okatani 1 and Koichiro Deguchi 1 This paper presents a method for recognizing the pose of a wire harness
More information鈴木(最終版)
2 4 3 Richards and Sampson (1974) 3 (language transfer) 3 Richards 1971 Overgeneralization 5 5 (2000) 6 6 7 8 9 10 10 11 14 Corder 1967 Snow(1998 Developmental error Interference error SOV SVO (1) a:
More information10 11 12 33.4 1 open / window / I / shall / the? 79.3 2 something / want / drink / I / to. 43.5 3 the way / you / tell / the library / would / to / me
-1- 10 11 12 33.4 1 open / window / I / shall / the? 79.3 2 something / want / drink / I / to. 43.5 3 the way / you / tell / the library / would / to / me? 28.7 4 Miyazaki / you / will / in / long / stay
More information1 Fig. 1 Extraction of motion,.,,, 4,,, 3., 1, 2. 2.,. CHLAC,. 2.1,. (256 ).,., CHLAC. CHLAC, HLAC. 2.3 (HLAC ) r,.,. HLAC. N. 2 HLAC Fig. 2
CHLAC 1 2 3 3,. (CHLAC), 1).,.,, CHLAC,.,. Suspicious Behavior Detection based on CHLAC Method Hideaki Imanishi, 1 Toyohiro Hayashi, 2 Shuichi Enokida 3 and Toshiaki Ejima 3 We have proposed a method for
More informationIPSJ SIG Technical Report 1, Instrument Separation in Reverberant Environments Using Crystal Microphone Arrays Nobutaka ITO, 1, 2 Yu KITANO, 1
1, 2 1 1 1 Instrument Separation in Reverberant Environments Using Crystal Microphone Arrays Nobutaka ITO, 1, 2 Yu KITANO, 1 Nobutaka ONO 1 and Shigeki SAGAYAMA 1 This paper deals with instrument separation
More information平成29年度英語力調査結果(中学3年生)の概要
1 2 3 1 そう思う 2 どちらかといえば そう思う 3 どちらかといえば そう思わない 4 そう思わない 4 5 楽しめるようになりたい 6 1 そう思う 2 どちらかといえば そう思う 3 どちらかといえば そう思わない 4 そう思わない 7 1 そう思う 2 どちらかといえば そう思う 3 どちらかといえば そう思わない 4 そう思わない 8 1 そう思う 2 どちらかといえば そう思う
More information浜松医科大学紀要
On the Statistical Bias Found in the Horse Racing Data (1) Akio NODA Mathematics Abstract: The purpose of the present paper is to report what type of statistical bias the author has found in the horse
More information,,.,.,,.,.,.,.,,.,..,,,, i
22 A person recognition using color information 1110372 2011 2 13 ,,.,.,,.,.,.,.,,.,..,,,, i Abstract A person recognition using color information Tatsumo HOJI Recently, for the purpose of collection of
More informationA Japanese Word Dependency Corpus ÆüËܸì¤Îñ¸ì·¸¤ê¼õ¤±¥³¡¼¥Ñ¥¹
A Japanese Word Dependency Corpus 2015 3 18 Special thanks to NTT CS, 1 /27 Bunsetsu? What is it? ( ) Cf. CoNLL Multilingual Dependency Parsing [Buchholz+ 2006] (, Penn Treebank [Marcus 93]) 2 /27 1. 2.
More informationWII-D 2017 (1) (2) (1) (2) [Tanaka 07] [ 04] [ 10] [ 13, 13], [ 08] [ 13] (1) (2) 2 2 e.g., Wikipedia [ 14] Wikipedia [ 14] Linked Open
Web 2017 Original Paper Supporting Exploratory Information Access Based on Comic Content Information 1 Ryo Yamashita Byeongseon Park Mitsunori Matsushita Nomura Research Institute, LTD. r-yamashita@nri.co.jp
More informationAERA_English_CP_Sample_org.pdf
W e l c o m e t o J A P A N 254 Singer-songwriter Kyrie Kristmanson I am isolating myself, when I am writing songs. Q: I have heard that you have been writing songs in the middle of nature. Why? A: The
More informationPage 1 of 6 B (The World of Mathematics) November 20, 2006 Final Exam 2006 Division: ID#: Name: 1. p, q, r (Let p, q, r are propositions. ) (10pts) (a
Page 1 of 6 B (The World of Mathematics) November 0, 006 Final Exam 006 Division: ID#: Name: 1. p, q, r (Let p, q, r are propositions. ) (a) (Decide whether the following holds by completing the truth
More informationC. S2 X D. E.. (1) X S1 10 S2 X+S1 3 X+S S1S2 X+S1+S2 X S1 X+S S X+S2 X A. S1 2 a. b. c. d. e. 2
I. 200 2 II. ( 2001) 30 1992 Do X for S2 because S1(is not desirable) XS S2 A. S1 S2 B. S S2 S2 X 1 C. S2 X D. E.. (1) X 12 15 S1 10 S2 X+S1 3 X+S2 4 13 S1S2 X+S1+S2 X S1 X+S2. 2. 3.. S X+S2 X A. S1 2
More information[1] 2 キトラ古墳天文図に関する従来の研究とその問題点 mm 3 9 mm cm 40.3 cm 60.6 cm 40.5 cm [2] 9 mm [3,4,5] [5] 1998
18 1 12 2016 キトラ古墳天文図の観測年代と観測地の推定 2015 5 15 2015 10 7 Estimating the Year and Place of Observations for the Celestial Map in the Kitora Tumulus Mitsuru SÔMA Abstract Kitora Tumulus, located in Asuka, Nara
More informationIPSJ SIG Technical Report Vol.2012-MUS-96 No /8/10 MIDI Modeling Performance Indeterminacies for Polyphonic Midi Score Following and
MIDI 1 2 3 2 1 Modeling Performance Indeterminacies for Polyphonic Midi Score Following and Its Application to Automatic Accompaniment Nakamura Eita 1 Yamamoto Ryuichi 2 Saito Yasuyuki 3 Sako Shinji 2
More information10 2000 11 11 48 ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) CU-SeeMe NetMeeting Phoenix mini SeeMe Integrated Services Digital Network 64kbps 16kbps 128kbps 384kbps
More informationkubostat2017b p.1 agenda I 2017 (b) probability distribution and maximum likelihood estimation :
kubostat2017b p.1 agenda I 2017 (b) probabilit distribution and maimum likelihood estimation kubo@ees.hokudai.ac.jp http://goo.gl/76c4i 2017 11 14 : 2017 11 07 15:43 1 : 2 3? 4 kubostat2017b (http://goo.gl/76c4i)
More information,,,,., C Java,,.,,.,., ,,.,, i
24 Development of the programming s learning tool for children be derived from maze 1130353 2013 3 1 ,,,,., C Java,,.,,.,., 1 6 1 2.,,.,, i Abstract Development of the programming s learning tool for children
More informationTHE INSTITUTE OF ELECTRONICS, INFORMATION AND COMMUNICATION ENGINEERS TECHNICAL REPORT OF IEICE.
THE INSTITUTE OF ELECTRONICS, INFORMATION AND COMMUNICATION ENGINEERS TECHNICAL REPORT OF IEICE. E-mail: {ytamura,takai,tkato,tm}@vision.kuee.kyoto-u.ac.jp Abstract Current Wave Pattern Analysis for Anomaly
More information3 2 2 (1) (2) (3) (4) 4 4 AdaBoost 2. [11] Onishi&Yoda [8] Iwashita&Stoica [5] 4 [3] 3. 3 (1) (2) (3)
(MIRU2012) 2012 8 820-8502 680-4 E-mail: {d kouno,shimada,endo}@pluto.ai.kyutech.ac.jp (1) (2) (3) (4) 4 AdaBoost 1. Kanade [6] CLAFIC [12] EigenFace [10] 1 1 2 1 [7] 3 2 2 (1) (2) (3) (4) 4 4 AdaBoost
More information-2-
Unit Children of the World NEW HORIZON English Course 'Have you been to?' 'What have you done as a housework?' -1- -2- Study Tour to Bangladesh p26 P26-3- Example: I am going to Bangladesh this spring.
More information4. C i k = 2 k-means C 1 i, C 2 i 5. C i x i p [ f(θ i ; x) = (2π) p 2 Vi 1 2 exp (x µ ] i) t V 1 i (x µ i ) 2 BIC BIC = 2 log L( ˆθ i ; x i C i ) + q
x-means 1 2 2 x-means, x-means k-means Bayesian Information Criterion BIC Watershed x-means Moving Object Extraction Using the Number of Clusters Determined by X-means Clustering Naoki Kubo, 1 Kousuke
More information16_.....E...._.I.v2006
55 1 18 Bull. Nara Univ. Educ., Vol. 55, No.1 (Cult. & Soc.), 2006 165 2002 * 18 Collaboration Between a School Athletic Club and a Community Sports Club A Case Study of SOLESTRELLA NARA 2002 Rie TAKAMURA
More informationVol.8 No (July 2015) 2/ [3] stratification / *1 2 J-REIT *2 *1 *2 J-REIT % J-REIT J-REIT 6 J-REIT J-REIT 10 J-REIT *3 J-
Vol.8 No.2 1 9 (July 2015) 1,a) 2 3 2012 1 5 2012 3 24, 2013 12 12 2 1 2 A Factor Model for Measuring Market Risk in Real Estate Investment Hiroshi Ishijima 1,a) Akira Maeda 2 Tomohiko Taniyama 3 Received:
More informationuntitled
A Consideration on Studies of English Literature in Japan This paper attempts to formulate the significance of English literary studies in present-day Japan, and to carve out new horizons of them. First,
More informationALT : Hello. May I help you? Student : Yes, please. I m looking for a white T-shirt. ALT : How about this one? Student : Well, this size is good. But do you have a cheaper one? ALT : All right. How about
More information1 Web [2] Web [3] [4] [5], [6] [7] [8] S.W. [9] 3. MeetingShelf Web MeetingShelf MeetingShelf (1) (2) (3) (4) (5) Web MeetingShelf
1,a) 2,b) 4,c) 3,d) 4,e) Web A Review Supporting System for Whiteboard Logging Movies Based on Notes Timeline Taniguchi Yoshihide 1,a) Horiguchi Satoshi 2,b) Inoue Akifumi 4,c) Igaki Hiroshi 3,d) Hoshi
More informationmain.dvi
1 1 1 2 3 LDA Estimating and Analyzing a Domain Topic Model of Entries Kensaku Makita 1 Hiroko Suzuki 1 Daichi Koike 1 Takehito Utsuro 2 Yasuhide Kawada 3 Abstract: In order to address the issue of quickly
More informationIPSJ SIG Technical Report Vol.2016-CE-137 No /12/ e β /α α β β / α A judgment method of difficulty of task for a learner using simple
1 2 3 4 5 e β /α α β β / α A judgment method of difficulty of task for a learner using simple electroencephalograph Katsuyuki Umezawa 1 Takashi Ishida 2 Tomohiko Saito 3 Makoto Nakazawa 4 Shigeichi Hirasawa
More informationOptical Flow t t + δt 1 Motion Field 3 3 1) 2) 3) Lucas-Kanade 4) 1 t (x, y) I(x, y, t)
http://wwwieice-hbkborg/ 2 2 4 2 -- 2 4 2010 9 3 3 4-1 Lucas-Kanade 4-2 Mean Shift 3 4-3 2 c 2013 1/(18) http://wwwieice-hbkborg/ 2 2 4 2 -- 2 -- 4 4--1 2010 9 4--1--1 Optical Flow t t + δt 1 Motion Field
More informationtikeya[at]shoin.ac.jp The Function of Quotation Form -tte as Sentence-final Particle Tomoko IKEYA Kobe Shoin Women s University Institute of Linguisti
tikeya[at]shoin.ac.jp The Function of Quotation Form -tte as Sentence-final Particle Tomoko IKEYA Kobe Shoin Women s University Institute of Linguistic Sciences Abstract 1. emphasis 2. Speaker s impressions
More information258 5) GPS 1 GPS 6) GPS DP 7) 8) 10) GPS GPS 2 3 4 5 2. 2.1 3 1) GPS Global Positioning System
Vol. 52 No. 1 257 268 (Jan. 2011) 1 2, 1 1 measurement. In this paper, a dynamic road map making system is proposed. The proposition system uses probe-cars which has an in-vehicle camera and a GPS receiver.
More information17 Proposal of an Algorithm of Image Extraction and Research on Improvement of a Man-machine Interface of Food Intake Measuring System
1. (1) ( MMI ) 2. 3. MMI Personal Computer(PC) MMI PC 1 1 2 (%) (%) 100.0 95.2 100.0 80.1 2 % 31.3% 2 PC (3 ) (2) MMI 2 ( ),,,, 49,,p531-532,2005 ( ),,,,,2005,p66-p67,2005 17 Proposal of an Algorithm of
More informationuntitled
Show & Tell Presentation - 170 - Presentation 1) Choose 1 topic 2) Write the reasons why you chose the topic. 3) Think about 3 points for the topic. Class No Name What would you like to do after graduation?
More information国際恋愛で避けるべき7つの失敗と解決策
7 http://lovecoachirene.com 1 7! 7! 1 NOT KNOWING WHAT YOU WANT 2 BEING A SUBMISSIVE WOMAN 3 NOT ALLOWING THE MAN TO BE YOUR HERO 4 WAITING FOR HIM TO LEAD 5 NOT SPEAKING YOUR MIND 6 PUTTING HIM ON A PEDESTAL
More informationRead the following text messages. Study the names carefully. 次のメッセージを読みましょう 名前をしっかり覚えましょう Dear Jenny, Iʼm Kim Garcia. Iʼm your new classmate. These ar
LESSON GOAL: Can read a message. メッセージを読めるようになろう Complete the conversation using your own information. あなた自身のことを考えて 会話を完成させましょう 1. A: Whatʼs your name? B:. 2. A: Whatʼs your phone number, (tutor says studentʼs
More information01.Œk’ì/“²fi¡*
AIC AIC y n r n = logy n = logy n logy n ARCHEngle r n = σ n w n logσ n 2 = α + β w n 2 () r n = σ n w n logσ n 2 = α + β logσ n 2 + v n (2) w n r n logr n 2 = logσ n 2 + logw n 2 logσ n 2 = α +β logσ
More informationgengo.dvi
4 97.52% tri-gram 92.76% 98.49% : Japanese word segmentation by Adaboost using the decision list as the weak learner Hiroyuki Shinnou In this paper, we propose the new method of Japanese word segmentation
More informationS1Šû‘KŒâ‚è
are you? I m thirteen years old. do you study at home every day? I study after dinner. is your cat? It s under the table. I leave for school at seven in Monday. I leave for school at seven on Monday. I
More informationJOURNAL OF THE JAPANESE ASSOCIATION FOR PETROLEUM TECHNOLOGY VOL. 66, NO. 6 (Nov., 2001) (Received August 10, 2001; accepted November 9, 2001) Alterna
JOURNAL OF THE JAPANESE ASSOCIATION FOR PETROLEUM TECHNOLOGY VOL. 66, NO. 6 (Nov., 2001) (Received August 10, 2001; accepted November 9, 2001) Alternative approach using the Monte Carlo simulation to evaluate
More information178 New Horizon English Course 28 : NH 3 1. NH 1 p ALT HP NH 2 Unit 2 p. 18 : Hi, Deepa. What are your plans for the holidays? I m going to visi
: 中学校の英語教科書を批判的に見る : なぜ学びが深まらないのか 渡部友子 0. 15 1 2017 Q&A Q&A 29 178 New Horizon English Course 28 : NH 3 1. NH 1 p. 11 1 ALT HP NH 2 Unit 2 p. 18 : Hi, Deepa. What are your plans for the holidays? I m going
More informationkubostat7f p GLM! logistic regression as usual? N? GLM GLM doesn t work! GLM!! probabilit distribution binomial distribution : : β + β x i link functi
kubostat7f p statistaical models appeared in the class 7 (f) kubo@eeshokudaiacjp https://googl/z9cjy 7 : 7 : The development of linear models Hierarchical Baesian Model Be more flexible Generalized Linear
More information2015 8 65 87. J. Osaka Aoyama University. 2015, vol. 8, 65-87. 20 * Recollections of the Pacific War in the eyes of a school kid Hisao NAGAOKA Osaka Aoyama Gakuen Summary Seventy years have passed since
More information09‘o’–
Gerald Graff s Method of Teaching Writing to First-Year College Students: Toward an Argument Culture IZUMI, Junji Abstract It is not easy to teach today s college students how to argue. Building on over
More information., White-Box, White-Box. White-Box.,, White-Box., Maple [11], 2. 1, QE, QE, 1 Redlog [7], QEPCAD [9], SyNRAC [8] 3 QE., 2 Brown White-Box. 3 White-Box
White-Box Takayuki Kunihiro Graduate School of Pure and Applied Sciences, University of Tsukuba Hidenao Iwane ( ) / Fujitsu Laboratories Ltd. / National Institute of Informatics. Yumi Wada Graduate School
More information149 (Newell [5]) Newell [5], [1], [1], [11] Li,Ryu, and Song [2], [11] Li,Ryu, and Song [2], [1] 1) 2) ( ) ( ) 3) T : 2 a : 3 a 1 :
Transactions of the Operations Research Society of Japan Vol. 58, 215, pp. 148 165 c ( 215 1 2 ; 215 9 3 ) 1) 2) :,,,,, 1. [9] 3 12 Darroch,Newell, and Morris [1] Mcneil [3] Miller [4] Newell [5, 6], [1]
More information教育実践上の諸問題
I go school by bus. I ll give this book Mary. () () Please tell me the way the station. ( ) : Oh. : Uh, is MISUIKAN your favorite onsen? : O.K. Why? : You said to eat ice cream after onsen. What kind
More informationInput image Initialize variables Loop for period of oscillation Update height map Make shade image Change property of image Output image Change time L
1,a) 1,b) 1/f β Generation Method of Animation from Pictures with Natural Flicker Abstract: Some methods to create animation automatically from one picture have been proposed. There is a method that gives
More informationuntitled
c 645 2 1. GM 1959 Lindsey [1] 1960 Howard [2] Howard 1 25 (Markov Decision Process) 3 3 2 3 +1=25 9 Bellman [3] 1 Bellman 1 k 980 8576 27 1 015 0055 84 4 1977 D Esopo and Lefkowitz [4] 1 (SI) Cover and
More informationOverview (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高2SL高1HL 文法後期後半_テキスト-0108.indd
第 20 講 関係詞 3 ポイント 1 -ever 2 3 ポイント 1 複合関係詞 (-ever) ever whoever whatever whichever whenever wherever You may take whoever wants to go. Whenever she comes, she brings us presents. = no matter whoever =
More informationkut-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 informationCain & Abel
Cain & Abel: False Religion vs. The Gospel Now Adam knew Eve his wife, and she conceived and bore Cain, saying, I have gotten a man with the help of the LORD. And again, she bore his brother Abel. Now
More informationはじめに
IT 1 NPO (IPEC) 55.7 29.5 Web TOEIC Nice to meet you. How are you doing? 1 type (2002 5 )66 15 1 IT Java (IZUMA, Tsuyuki) James Robinson James James James Oh, YOU are Tsuyuki! Finally, huh? What's going
More informationN cos s s cos ψ e e e e 3 3 e e 3 e 3 e
3 3 5 5 5 3 3 7 5 33 5 33 9 5 8 > e > f U f U u u > u ue u e u ue u ue u e u e u u e u u e u N cos s s cos ψ e e e e 3 3 e e 3 e 3 e 3 > A A > A E A f A A f A [ ] f A A e > > A e[ ] > f A E A < < f ; >
More informationキャリアワークショップ教師用
iii v vi vii viii ix x xi xii 1 2 3 4 1.1 CYCLE OF SELF-RELIANCE GOALS SUCCESS INTERACTION RESOURCES 5 6 7 8 9 10 11 12 13 14 15 16 17 18 2.1 MY RESOURCES FOR THE EARTH IS FULL, AND THERE IS ENOUGH AND
More information平成○○年度知能システム科学専攻修士論文
A Realization of Robust Agents in an Agent-based Virtual Market Makio Yamashige 3 7 A Realization of Robust Agents in an Agent-based Virtual Market Makio Yamashige Abstract There are many people who try
More informationQ [4] 2. [3] [5] ϵ- Q Q CO CO [4] Q Q [1] i = X ln n i + C (1) n i i n n i i i n i = n X i i C exploration exploitation [4] Q Q Q ϵ 1 ϵ 3. [3] [5] [4]
1,a) 2,3,b) Q ϵ- 3 4 Q greedy 3 ϵ- 4 ϵ- Comparation of Methods for Choosing Actions in Werewolf Game Agents Tianhe Wang 1,a) Tomoyuki Kaneko 2,3,b) Abstract: Werewolf, also known as Mafia, is a kind of
More informationThere are so many teachers in the world, and all of them are different. Some teachers are quiet and dont like to talk to students. Other teachers like
17 章 関 係 代 名 詞 ( 目 的 格 ) わからないときは サポート のココ! E3G 9 章 1,2,3 解 答 時 間 のめやす 45 分 アウ The girl looks very pretty is Mary. What is the book you bought yesterday? This is a book makes me happy. アwhichイthatウwho
More information生研ニュースNo.132
No.132 2011.10 REPORTS TOPICS Last year, the Public Relations Committee, General Affairs Section and Professor Tomoki Machida created the IIS introduction video in Japanese. As per the request from Director
More information(MIRU2008) HOG Histograms of Oriented Gradients (HOG)
(MIRU2008) 2008 7 HOG - - E-mail: katsu0920@me.cs.scitec.kobe-u.ac.jp, {takigu,ariki}@kobe-u.ac.jp Histograms of Oriented Gradients (HOG) HOG Shape Contexts HOG 5.5 Histograms of Oriented Gradients D Human
More informationFig. 2 28th Ryuou Tournament, Match 5, 59th move. The last move is Black s Rx5f. 1 Tic-Tac-Toe Fig. 1 AsearchtreeofTic-Tac-Toe. [2] [3], [4]
1,a) 2 3 2017 4 6, 2017 9 5 Predicting Moves in Comments for Shogi Commentary Generation Hirotaka Kameko 1,a) Shinsuke Mori 2 Yoshimasa Tsuruoka 3 Received: April 6, 2017, Accepted: September 5, 2017 Abstract:
More information2) TA Hercules CAA 5 [6], [7] CAA BOSS [8] 2. C II C. ( 1 ) C. ( 2 ). ( 3 ) 100. ( 4 ) () HTML NFS Hercules ( )
1,a) 2 4 WC C WC C Grading Student programs for visualizing progress in classroom Naito Hiroshi 1,a) Saito Takashi 2 Abstract: To grade student programs in Computer-Aided Assessment system, we propose
More informationIPSJ SIG Technical Report Vol.2010-GN-74 No /1/ , 3 Disaster Training Supporting System Based on Electronic Triage HIROAKI KOJIMA, 1 KU
1 2 2 1, 3 Disaster Training Supporting System Based on Electronic Triage HIROAKI KOJIMA, 1 KUNIAKI SUSEKI, 2 KENTARO NAGAHASHI 2 and KEN-ICHI OKADA 1, 3 When there are a lot of injured people at a large-scale
More information(3.6 ) (4.6 ) 2. [3], [6], [12] [7] [2], [5], [11] [14] [9] [8] [10] (1) Voodoo 3 : 3 Voodoo[1] 3 ( 3D ) (2) : Voodoo 3D (3) : 3D (Welc
1,a) 1,b) Obstacle Detection from Monocular On-Vehicle Camera in units of Delaunay Triangles Abstract: An algorithm to detect obstacles by using a monocular on-vehicle video camera is developed. Since
More informationVol. 43 No. 2 Feb. 2002,, MIDI A Probabilistic-model-based Quantization Method for Estimating the Position of Onset Time in a Score Masatoshi Hamanaka
Vol. 43 No. 2 Feb. 2002,, MIDI A Probabilistic-model-based Quantization Method for Estimating the Position of Onset Time in a Score Masatoshi Hamanaka, Masataka Goto,, Hideki Asoh and Nobuyuki Otsu, This
More informationa) Extraction of Similarities and Differences in Human Behavior Using Singular Value Decomposition Kenichi MISHIMA, Sayaka KANATA, Hiroaki NAKANISHI a
a) Extraction of Similarities and Differences in Human Behavior Using Singular Value Decomposition Kenichi MISHIMA, Sayaka KANATA, Hiroaki NAKANISHI a), Tetsuo SAWARAGI, and Yukio HORIGUCHI 1. Johansson
More informationIPSJ SIG Technical Report Vol.2009-DPS-141 No.20 Vol.2009-GN-73 No.20 Vol.2009-EIP-46 No /11/27 1. MIERUKEN 1 2 MIERUKEN MIERUKEN MIERUKEN: Spe
1. MIERUKEN 1 2 MIERUKEN MIERUKEN MIERUKEN: Speech Visualization System Based on Augmented Reality Yuichiro Nagano 1 and Takashi Yoshino 2 As the spread of the Augmented Reality(AR) technology and service,
More informationJFE.dvi
,, Department of Civil Engineering, Chuo University Kasuga 1-13-27, Bunkyo-ku, Tokyo 112 8551, JAPAN E-mail : atsu1005@kc.chuo-u.ac.jp E-mail : kawa@civil.chuo-u.ac.jp SATO KOGYO CO., LTD. 12-20, Nihonbashi-Honcho
More information<30375F97E996D88E812E696E6464>
Abstract: This study is intended as an investigation of the transition of Lady Windermere s Fan on stage in the Republic of china. Oscar Wild s Lady Windermere s Fan was adapted for the Chinese stage by
More informationWhat s your name? Help me carry the baggage, please. politeness What s your name? Help me carry the baggage, please. iii
What s your name? Help me carry the baggage, please. politeness What s your name? Help me carry the baggage, please. iii p. vi 2 50 2 2016 7 14 London, Russell Square iv iii vi Part 1 1 Part 2 13 Unit
More information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 informationABSTRACT The "After War Phenomena" of the Japanese Literature after the War: Has It Really Come to an End? When we consider past theses concerning criticism and arguments about the theme of "Japanese Literature
More informationModal Phrase MP because but 2 IP Inflection Phrase IP as long as if IP 3 VP Verb Phrase VP while before [ MP MP [ IP IP [ VP VP ]]] [ MP [ IP [ VP ]]]
30 4 2016 3 pp.195-209. 2014 N=23 (S)AdvOV (S)OAdvV 2 N=17 (S)OAdvV 2014 3, 2008 Koizumi 1993 3 MP IP VP 1 MP 2006 2002 195 Modal Phrase MP because but 2 IP Inflection Phrase IP as long as if IP 3 VP Verb
More informationディープラーニングとオープンサイエンス ~研究の爆速化が引き起こす摩擦なき情報流通へのシフト~
KITAMOTO Asanobu http://researchmap.jp/kitamoto/ KitamotoAsanob u 1 2 3 4 5 1. 2. 3. 6 Lawrence Lessig (Founder of Creative Commons), Code: And Other Laws of Cyber Space (first edition 1999) 7 NSF Data
More information鹿大広報149号
No.149 Feb/1999 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Learned From Japanese Life and Experiences in Kagoshima When I first came to Japan I was really surprised by almost everything, the weather,
More informationIPSJ SIG Technical Report Vol.2012-CG-148 No /8/29 3DCG 1,a) On rigid body animation taking into account the 3D computer graphics came
3DCG 1,a) 2 2 2 2 3 On rigid body animation taking into account the 3D computer graphics camera viewpoint Abstract: In using computer graphics for making games or motion pictures, physics simulation is
More information& Vol.5 No (Oct. 2015) TV 1,2,a) , Augmented TV TV AR Augmented Reality 3DCG TV Estimation of TV Screen Position and Ro
TV 1,2,a) 1 2 2015 1 26, 2015 5 21 Augmented TV TV AR Augmented Reality 3DCG TV Estimation of TV Screen Position and Rotation Using Mobile Device Hiroyuki Kawakita 1,2,a) Toshio Nakagawa 1 Makoto Sato
More information情報処理学会研究報告 IPSJ SIG Technical Report Vol.2013-MPS-93 No /5/23 統計的文法獲得モデルのための部分木ブロック化サンプリング法 進藤裕之 1,a) 松本裕治 2 永田昌明 1 概要 : 自然言語処理分野における統計的文法獲得では,
統計的文法獲得モデルのための部分木ブロック化サンプリング法 進藤裕之 1,a) 松本裕治 2 永田昌明 1 概要 : 自然言語処理分野における統計的文法獲得では, 確率文法モデルの学習に Gibbs サンプリング法が広く用いられている. しかしながら, 木構造データを扱う場合には,Gibbs サンプリング法のように変数の値を一つずつ順番に更新していく方法では局所解に留まりやすく, 十分に尤度の高い解を得られないという問題がある.
More information( )
NAIST-IS-MT0851100 2010 2 4 ( ) CR CR CR 1980 90 CR Kerberos SSH CR CR CR CR CR CR,,, ID, NAIST-IS- MT0851100, 2010 2 4. i On the Key Management Policy of Challenge Response Authentication Schemes Toshiya
More informationIPSJ SIG Technical Report Vol.2014-DBS-159 No.6 Vol.2014-IFAT-115 No /8/1 1,a) 1 1 1,, 1. ([1]) ([2], [3]) A B 1 ([4]) 1 Graduate School of Info
1,a) 1 1 1,, 1. ([1]) ([2], [3]) A B 1 ([4]) 1 Graduate School of Information Science and Technology, Osaka University a) kawasumi.ryo@ist.osaka-u.ac.jp 1 1 Bucket R*-tree[5] [4] 2 3 4 5 6 2. 2.1 2.2 2.3
More informationCOE-RES Discussion Paper Series Center of Excellence Project The Normative Evaluation and Social Choice of Contemporary Economic Systems Graduate Scho
COE-RES Discussion Paper Series Center of Excellence Project The Normative Evaluation and Social Choice of Contemporary Economic Systems Graduate School of Economics and Institute of Economic Research
More information