[1] Legaspi Inductive Logic Programming; ILP Genetic Algorithm; GA [2]

Transcription

1 a) []Legaspi Inductive Logic Programming; ILP Genetic Algorithm; GA [2] [3] [4] [5] 2 2 GA (Harmony Search; HS) Faculty of Environmental and Information Studies Tokyo City University 2 The Institute of Scientific and Industrial Research Osaka University a) otani@tcu.ac.jp HS c 203 Information Processing Society of Japan

2 Table Features of a Frame Structure. pops C D D E E F F G G A A B C D D E E F G G A A B major minor 4/4 larghetto andante moderato allegretto allegro vivace presto square lead bass and lead nylon guitar chiff lead sawtooth lead flute pan flute alto sax strings vibraphone harmonica bassoon recorder soprano sax tenor sax clarinet steel guitar charang lead calliope lead english horn lead guitar flute sax vibraphone harmonica bassoon recorder clarinet strings english horn clean guitar steel guitar nylon guitar distortion guitar overdriven guitar clavi synth brass bright piano grand piano electric grand piano electric piano rock organ synth voice strings halo pad synth brass guitar organ piano clavi synth voice strings pad 2 Table 2 Features of a Chord. I I II II III III IV IV V V VI VI VI VII VII M 7 M7 6 aug aug7 m m7 mm7 m6 m7(b5) dim7 sus4 7sus4 add9 9th 9th 9th th th 3th 3th () (2) (3) (4) (5) (6) (7) GA [3] [4] [5] GA HS 3 Table 3 Positive and Negative Examples t t t 35 2 t 3 25 t t 3 45 (2) (6) (2) (6) (2) (6) 2.3 SD [6] t t 3 3 t t ILP FOIL[7] c 203 Information Processing Society of Japan 2

3 %" $" &&&&&% $ ) ) # )! ( " ) ) #" ( " ) )!" ) ) ( ) " ()nullnullnullnullnullnullnullnull beat-beat-nullnullbeat- (2) nullnullnullnullbeat-beat- (3) nullnullnullnullbeat-beat- (4)--beat nullnullnullnull Fig. An Example of Background Knowledge frame(brighta) :- tempo(aandante) chord_category(apiano). motif(brighta) :- motif(abar((ivmajor)-_-)bar(_---)). chords(brighta) :- has_chord(abc)root(cvi)type(cminor) next_to(abd_) has_chord(ade)root(evi). rhythm(brighta) :- has_rhythm(abc)beat2(cnull) next_to(abd_) has_rhythm(ade)beat(ebeat)beat4(e-). Fig. 2 2 Examples of Personal Sensibility Model /4 null beat /4 - ()(4) ()(4) 2 frame motif IV major chords VI minor2 VI 2 rhythm /4 /4 3. ( ) 2 ( 2 ) HM ; worst := HM ; worstfit := f(worst); for i := to G { r := ; if(r < Rc) { new := HM ; r2 := ; if(r2 < Ra) { new := new; else { new := ; newfit := f(new); if(newfit > worstfit) { worst new ; worst := HM ; worstfit := f(worst); Fig. 3 3 HS Pseudo Code of a Harmony Search Algorithm ( 3 ) HS HS 3. HS HS [8] ( ) ( 2 ) ( 3 ) HS (Harmony Memory; HM) 3 ( ) HM ( 2 ) HM ( 3 ) HS HM HS 3 G R c HM R a f(x) c 203 Information Processing Society of Japan 3

4 N... N Fig HS N Example of a Harmony N M 4MN N /4 0 / R f(r) () 3 r(m i) f(r) = c(m i j) n(r M i j) i= j= () r(m i ) c(m i j) n(r M i j) (2i ) j= M i t i M i t i c(m i j) M i j n(r M i j) R M i j r(m i ) M i HM HM 4 Table 4 HS Parameters in HS G HM R c 0.85 HM R a 0.30 HM [9] Affective Value Scale of Music; AVSM nomodel nomodel t t 2 HS nomodel nomodel nomodel 6 7 c 203 Information Processing Society of Japan 4

5 2 rhythm(darka) :- has_rhythm(a_b)beat4(bbeat). rhythm(darka) :- has_rhythm(abc)beat(cbeat)next_to(abd_) has_rhythm(ade)beat2(ebeat)beat3(ebeat).!!!!!!!!!!!$!!!!!!!!!!!#!!!!!!!!!$!!!!!!!!!#!!!!!!!!!!!$!!!!!!!!!!!#!!"!!#!"!#!!!"!!!#!!!"!!!#!!!"!!!#!!"!!#!!!!"!!!!#!!"!!#!!"!!#!!!!!!!!!!$!!!!!!!!!!#!!"!!#!!"!!# * & $% " $ # % " % $ % % $ % $ $ % % $ ( &!!!!!!!! * ( / ( / /. / * & ) & + /. /)0)23 (// / / (/ /.!!!!!!!!!!!$!!!!!!!!!!!#!!!!!!!!!!!!$!!!!!!!!!!!!#!!!!!!!!!!!!$!!!!!!!!!!!!#!!!!!!!!!!!!!$!!!!!!!!!!!!!#!!!!"!!!!#!!!!"!!!!#!!!"!!!#!!!"!!!#!!!"!!!#!!!"!!!#!"!#!!!!!!"!!!!!!#!!!"!!!#!!!"!!!#!!!!"!!!!#!!!"!!!#!!!"!!!#!!"!!# * ( ( ( % $ $ % $ " $ # % $ *!! " ( / (! ( /. ( - * (// / -. /( ) --- //// / + Fig. 5 5 A Personal Sensibility Model and a Composed dark Musical Piece t % nomodel 3 t % nomodel nomodel 8 t [5] 0. nomodel c 203 Information Processing Society of Japan 5

6 t Fig. 6 Evaluation of the Impression of Composed Musical Pieces by the Proposed Method t nomodel Fig. 7 Evaluation of the Impression of Composed Musical Pieces by nomodel Method 5 nomodel Fig. 8 Evaluation of the Quality of Composed Musical Pieces 5. HS (B) nomodel nomodel [] Numao M. Takagi S. and Nakamura K.: Constructive Adaptive User Interfaces - Composing Music Based on Human Feelings Proc. 8th National Conference on Artificial Intelligence AAAI Press pp (2002). [2] Legaspi R. Hashimoto Y. Moriyama K. Kurihara S. and Numao M.: Music Compositional Intelligence with an Affective Flavor Proc. of ACM International Conference on Intelligent User Interfaces pp (2007). [3] 23 F- (2009). [4] 23 3H3-2 (2009). [5] Otani N. Tadokoro K. Kurihara S. and Numao M.: Generation of Chord Progression Using Harmony Search Algorithm for a Constructive Adaptive User Interface Proc. of 2th Pacific Rim International Conference on Artificial Intelligence LNAI 7458 pp (202). [6] Osgood C. Suci G. and Tannenbaum P.: The Measurement of Meaning University of Illinois Press (957). [7] Quinlan J.: Learning Logical Definitions from Relations Machine Learning Vol. 5 pp (990). [8] Geem Z. Kim J. and Loganathan G.: A New Heuristic Optimization Algorithm: Harmony Search Simulation Vol. 76 pp (200). [9] Vol. 65 pp (995). c 203 Information Processing Society of Japan 6