Data Festã§ã®ä»¥åã®è¬çŸ©ã¯ãæ°ããçš®é¡ã®æ€çŽ¢ãæ§ç¯ããããã«å¿
èŠãªã¢ã«ãŽãªãºã ã«å°å¿µããŸãã ã ä»æ¥ã®ã¬ããŒãããããæå³ã§ãç°ãªãã¢ã«ãŽãªãºã ã«é¢ãããã®ã§ãããããããããã®å€ãã®åºç€ãšãªãæ°åŠã«é¢ãããã®ã§ãã ç§åŠå士ã§ãããSkoltechã®èšç®ææ³ã°ã«ãŒãã®è²¬ä»»è
ã§ããIvan Oseledetsã¯ãè¡åå解ã«ã€ããŠèŽè¡ã«èªããŸããã
ã«ããã®äž-ãã³ãŒããšã»ãšãã©ã®ã¹ã©ã€ãã
ç·åœ¢ä»£æ°ãæ±ãå°ããª-æ¯èŒçæ©æ¢°åŠç¿-ã®ã³ãã¥ããã£ããããŸãã ç·åœ¢ä»£æ°ãããããã®äžã«å€éç·åœ¢ä»£æ°ããã³ãœã«å解ãæ±ã人ã
ãããŸãã ç§ã¯åœŒãã«é¢ä¿ããŠããŸãã ããã§ã¯ããªãå€ãã®èå³æ·±ãçµæãåŸãããŸããããã³ãã¥ããã£ãèªåèªèº«ãçµç¹åããŠå€ã«åºãããšã¯ã§ããªãããããããã®çµæã®äžéšã¯äžæã§ãã ãšããã§ãããã¯ç·åœ¢ä»£æ°ã«ãåœãŠã¯ãŸããŸãã NIPSãICMLãªã©ã®äž»èŠãªäŒè°ã®èšäºãèŠããšãç·åœ¢ä»£æ°ã®äºçŽ°ãªäºå®ã«é¢é£ããå€ãã®èšäºãèŠãããšãã§ããŸãã ããã¯èµ·ãããŸãããããã§ãèµ·ãããŸãã
ãã³ãœã«å±éã«ã€ããŠèª¬æããŸãããæåã«ãããªãã¯ã¹å±éã«ã€ããŠèª¬æããå¿
èŠããããŸãã
è¡åã ãã ã®2次å
ããŒãã«ã è¡åå解ã¯ãããŒã«çŽ æ°ã®ç©ã®åœ¢åŒã§ã®è¡åã®è¡šçŸã§ãã å®éããããã¯ã©ãã§ã䜿çšããã誰ããæºåž¯é»è©±ãæã¡ã座ã£ãŠããéãç·åœ¢ã·ã¹ãã ã解決ããããŒã¿ãéä¿¡ããè¡åãå解ããŸãã 確ãã«ããããã®ãããªãã¯ã¹ã¯éåžžã«å°ããã4x4ã8x8ã§ãã ããããå®éã«ã¯ãããã®è¡åã¯éåžžã«éèŠã§ãã
LUå解ãã¬ãŠã¹å解ãQRå解ããããšèšããããåŸãŸããããäž»ã«ç¹ç°å解ã«ã€ããŠã話ããŸããããã¯ãäœããã®çç±ã§ãããšãã°ãç§ãã¡ã®å€§åŠã®ç·åœ¢ä»£æ°ã³ãŒã¹ã§ã¯ãå¿ããããã¡ã§ãã 圌ãã¯ããžã§ãŒãã³åœ¢åŒã®ãããªèšç®ç§åŠã®èŠ³ç¹ãããã®ãããªãã³ã»ã³ã¹ãèªã¿ãŸãããããŠãããã¯ã©ãã§ã䜿ãããŸããã ç¹ç°å解ã¯éåžžãæçµçã«ã©ããã«ããããªãã·ã§ã³ãªã©ã§ããããã¯ééã£ãŠããŸãã Golubãšvan Lounã® ãMatrix Calculationsããšããæ¬ãåãäžãããšãLU-ãšQR-ã®ã©ã¡ãã§ããªããç¹ç°å解ã§å§ãŸããŸãã ç©ã®åœ¢åŒã®ãããªãã¯ã¹ã®ãã®è¡šçŸã¯ããŠãã¿ãªã察è§ããŠãã¿ãªã§ãã
ãããªãã¯ã¹å解ã«ã€ããŠãè¯ãããšã¯ããããã«ã€ããŠå€ããæ¬åœã«ç¥ãããŠããããšã§ãã åºæ¬çãªå解ã«ã¯ãå¹æçãªã¢ã«ãŽãªãºã ããããå®å®ããŠããŸããMATLABã§åŒã³åºãããšãã§ãããœãããŠã§ã¢ããããŸããPythonã§ã¯ãFortranã§èšè¿°ãããŠããŸãã ãããã«ãŠã³ãå¯èœãªåäžã®æäœãšæ£ããèŠãªããšãæ£ç¢ºãªçµæãåŸãããããšã確å®ã«ãªããŸããåŸé
éäžãè¡ãå¿
èŠã¯ãããŸããã å®éãããã¯æäœã®1ã€ã®ã¯ã©ã³ã¿ã ã§ãã åé¡ããããªãã¯ã¹å±éã®èšç®ã«æžãããå Žåãå®éã«ããã解決ããŸããã ããã¯ãããšãã°ãããŸããŸãªã¹ãã¯ãã«æ³ã®åºç€ãšãªããŸãã ããšãã°ãEMã¢ã«ãŽãªãºã ãããå Žåãå埩ã¢ã«ãŽãªãºã ã§ãããåæãé
ããªããŸãã ãããŠãã¹ãã¯ãã«ã¡ãœãããååŸã§ããå Žåãããªãã¯çŽ æŽããããã¯ã³ã¹ãããã®ãœãªã¥ãŒã·ã§ã³ãåŸãããŸãã
çŽæ¥åãç¹ç°å解ã¯åé¡ã解決ããŸããããäžè¬çãªåé¡ã解決ããããã®äžçš®ã®ãã«ãã£ã³ã°ãããã¯ã§ãã æºåž¯é»è©±ã«æ»ããŸãã ãã®äŸã¯ãæãããªçç±ã§ç§ãæ©ãŸããŸããç§ãã¡ã¯æ¬åœã«ãããå¿
èŠãšããéä¿¡äŒç€ŸãšååããŠããŸãã
ç¹ç°å解ãšã¯äœããèŠãŠããããã€ã³ããã¯ã¹åœ¢åŒã§æžããšã次ã®åŒãåŸãããŸãã ããã¯å€å
žçãªå€ãã¢ã€ãã¢ã«ã€ãªãããŸã-倧åŠã®æåã®ã³ãŒã¹ã«æ»ããšãå€æ°ã®åé¢ã«ã€ãªãããŸãã xãtããå解ããããã§Iãjããåé¢ãã1ã€ã®é¢æ£ã€ã³ããã¯ã¹ãå¥ã®ã€ã³ããã¯ã¹ããåé¢ããŸããã€ãŸãã1ã€ã®å€æ°ããã®é¢æ°ã®ç©ã®åœ¢ã§2ã€ã®å€æ°ããé¢æ°ãè¡šããŸãã ãã®åœ¢åŒã®é¢æ°ãã»ãšãã©ãªãããšã¯æããã§ããããã®ãããªé¢æ°ã®åèšãæžããšãããªãåºãã¯ã©ã¹ã«ãªããŸãã
å®éãç¹ç°å解ã®äž»ãªçšéã¯ããã®ãããªäœã©ã³ã¯ã®è¿äŒŒã®æ§ç¯ã§ãã ããã¯ã©ã³ã¯rã®è¡åã§ãããåœç¶ãã©ã³ã¯ãè¡åã®åèšãµã€ãºããã倧å¹
ã«å°ããå Žåã«é¢å¿ããããŸãã
ããã§ãã¹ãŠãé 調ã§ãã ã©ã³ã¯ãä¿®æ£ããæé©ãªè¿äŒŒå€ãèŠã€ããã¿ã¹ã¯ãèšå®ãããšãç¹ç°å解ã®è§£ãäžããããŸãã å®éãã¯ããã«å€ãã®ããšãç¥ãããŠãããç¹ç°å解ã®å©ããåããå Žåãããéãè¡ãããšãã§ããŸããè¡åã®ãã¹ãŠã®èŠçŽ ãèšç®ããããšãªãããªã©ã§ãã倧ããªçŸããçè«ããããŸã-äœã©ã³ã¯è¡åã®çè«ã¯ãçºå±ãç¶ããŠããããŸã éããããŠããŸããã
ãã ããéããããŠããŸãããã3ã€ä»¥äžã®ã€ã³ããã¯ã¹ãäœæããŸãã
ãã®ãããã¯ã«å¯Ÿããç§ã®æåã®é¢å¿ã¯çŽç²ã«çè«çãªãã®ã§ããã äžéšã®ã¢ããªã±ãŒã·ã§ã³ã§ã¯ããããªãã¯ã¹ã䜿çšããŠäœããå®è¡ããå解ããè¡šçŸããããšãã§ããŸãããããŸã§ã®ãšãããã©ã®ã¢ããªã±ãŒã·ã§ã³ã§ãåé¡ãããŸããã æåŸã«èšããŸãã
ãããŠãAïŒiãjïŒã®ä»£ããã«AïŒiãjkïŒãèšè¿°ããå€æ°ãåé¢ããããšãããšã©ããªããŸããã ãããã£ãŠããã³ãœã«å±éã®åé¡ã¯å®å
šã«èªç¶ãªæ¹æ³ã§çºçããŸãã ãã³ãœã«ããããŸã-ãã³ãœã«ãšã¯ãå€æ¬¡å
é
åãåçŽãª2次å
ããŒãã«ãæå³ããŸã-ãããã®å€æ¬¡å
é
åãå§çž®ããŠã倧ãŸãã«èšã£ãŠãäœãã©ã¡ãŒã¿ãŒè¡šçŸãæ§ç¯ããéåžžã¯çŸããèšèã§èšããšã次å
ã®åªããå
æããããšæããŸãã å®éããã®çšèª-ããã¯ããã€ãã®ããã«ã1961幎ã®ãã«ãã³ã®äœåã§ããŸã£ããç°ãªãã¿ã¹ã¯ãšãŸã£ããç°ãªãé¢ä¿ã§ç»å ŽããŸããã ããããçŸåšãdã®nã¯éåžžã«æªããšããæå³ã§äœ¿çšãããŠããŸãã dãå¢ããããå Žåãææ°é¢æ°çãªæé·ãªã©ããããŸãã
æãäžè¬çãªæå³ã§ã®å æ°å解ïŒã¢ããªã±ãŒã·ã§ã³ã§çºçããç¹å®ã®ã¯ã©ã¹ã®ãã³ãœã«ãéžæãããããããã³ãœã«ããã°ããå埩ã§ãããããªäœãã©ã¡ãŒã¿ãŒè¡šçŸãæ§ç¯ããããšããŠããŸãã ãŸãããã©ã¡ãŒã¿ãŒã¯ã»ãšãã©ãããŸããã ååãšããŠããã¥ãŒã©ã«ãããã¯ãŒã¯ã¯ãããæå³ã§ãã³ãœã«ãè¿äŒŒããåé¡ã解決ãããšããããšãã§ããŸãããªããªãããããã¯ãã©ããå¥ã®ãã©ã¡ãŒã¿ç©ºéãããããã³ã°ãæ§ç¯ããããšããŠããããã§ãã
å æ°å解ã®åé¡ãã€ãŸãåçŽãªãªããžã§ã¯ãã®ç©ã®åœ¢åŒã§ã®è¡šçŸã«ããã³ãœã«ãæã€è¡åãšè¡åã®ç©ãããå Žåããã³ãœã«ã®æ±ºå®ã¯ããè€éãªã¿ã¹ã¯ã§ãã ããããäœã©ã³ã¯è¿äŒŒã®äžå®å
šãªç¹ç°å解ã®ã€ã³ããã¯ã¹ã¬ã³ãŒããèŠããšãåœç¶ãåæ§ã®åŒãåŸãããŸãã 圌女ãããã 1ã€ã®é¢æ£å€æ°ã®é¢æ°ã®ç©ã§ãããªããžã§ã¯ãã§ãã³ãœã«ãè¿äŒŒããé¢æ°t1ã«é¢æ°t2ãæããããšããŠããŸãã ã¢ã€ãã¢ã¯å®å
šã«èªç¶ã§ãã çšèªã1ã€ãããªãå Žåã¯ãäœãè¿ã¥ããŸããã ãããŠãããªããããªãå°æ°ã®ãã©ã¡ãŒã¿ãŒãåãããšãèš±å¯ããå Žåããã³ãœã«ã®éåžžã«èå³æ·±ãã¯ã©ã¹ãçºçããŸããããã¯ãã¡ãœããã®å€ãã®åé¡ã§æ瀺çãŸãã¯æé»çã«äœ¿çšãããŸãã ã»ãŒãã®åœ¢åŒã¯ãéåååŠã§æ³¢åé¢æ°ãè¡šãããã«äœ¿çšãããŠãããšèšããŸãã å察称æ§ãããã補åã®ä»£ããã«æ±ºå®èŠå ãçŸããŸã-ããããç§åŠã®å·šå€§ãªåéå
šäœã¯ãå®éããã®ãããªåé¢å¯èœãªè¿äŒŒã«åºã¥ããŠããŸãã
ãããã£ãŠããã¡ãããããããã£ã調ã¹ãããããã§äœãèµ·ãã£ãŠããããç解ãããããããšã¯èå³æ·±ãããšã§ãã
ãã®ãšã³ãã£ãã£-1ã€ã®å€æ°ã®é¢æ°ã®ç©ã®åèšãšããŠã®è¡šçŸ-ã¯æšæºå解ãšåŒã°ããŸãã æåã«ææ¡ãããã®ã¯1927幎ã§ãã ãããããæ°åŠããé ãé¢ããŠäœ¿çšãããŸããã çµ±èšåŠãã±ã¢ã¡ããªãã¯ã¹ãPsychomatrixãªã©ã®éèªã§ã人ã
ãããŒã¿ãã¥ãŒããåãåã£ãäœåããããŸããã å
åããããŸã-äœããèµ·ãããŸããã ãããã£ãŠã圌ãã¯å€èŠçŽ ã¢ãã«ãæ§ç¯ããŸããã ãããã®èŠå ã®ãããªãã¯ã¹ã¯ãããã€ãã®èŠå ãšããŠè§£éã§ããŸãã ãããã®ãã¡ãããã€ãã®çµè«ãå°ãåºãããšãã§ããŸãã
ãã©ã¡ãŒã¿ãŒã®æ°ãæ°ãããšãdnrãã©ã¡ãŒã¿ãŒãååŸãããŸãã rãå°ããå Žåããã¹ãŠãæ£åžžã§ãã ããããç§ã解決ããããšããŠããåé¡ã¯ãååãšããŠãããšããã®ãããªè¿äŒŒãååšããããšãããã£ãŠããŠãããããèšç®ããããšã¯å°é£ãªäœæ¥ã§ããããšã§ãã ã©ã³ã¯ãèšç®ããã¿ã¹ã¯ã®ããã«ãäžè¬çãªå Žåã¯NPè€éã§ãããšèšãããšããã§ããŸãã ãããªãã¯ã¹ã®å Žåãããã¯ããã§ã¯ãããŸããã è¡åã®ã©ã³ã¯ãèšç®ã§ããŸã-ããšãã°ãã¬ãŠã¹äŸå€ã䜿çšãããšãããã¯å€é
åŒã®æŒç®æ°ã§å®è¡ã§ããŸãã ããã§ã¯ãã¹ãŠãééã£ãŠããŸãã
ãµã€ãºã9x9x9ã®ãã³ãœã«ã®äŸããããŸããããã®æå°é
æ°ã®æ£ç¢ºãªå€ã¯ãŸã äžæã§ãã ããã¯23以äžã21以äžã§ããããšãç¥ãããŠããŸããã¹ãŒããŒã³ã³ãã¥ãŒã¿ãŒãåçŽãªå€é
åŒïŒããã¯ãå€é
æ¹çšåŒç³»ã®å¯è§£æ§ã®åé¡ã§ãã ãããããã®ãããªã¿ã¹ã¯ã¯éåžžã«é£ããå ŽåããããŸãã ãããŠããã³ãœã«ã¯å®éçã«éèŠã§ãã ãã®ãã³ãœã«ã®ã©ã³ã¯ã¯ãè¡åä¹ç®ã¢ã«ãŽãªãºã ã®è€é床ã€ã³ããã¯ã¹ã«é¢é£ããŠããŸãã Strassenã®å¯Ÿæ°ã¯7ããã®ãã€ããªã§ãã ãããã£ãŠãå®éã«ã¯ãææ°ãèšç®ããã¿ã¹ã¯ã¯ãäžéšã®ãã³ãœã«ã®æšæºã©ã³ã¯ãèšç®ããããšã«ãªããŸãã
å€ãã®ã¢ããªã±ãŒã·ã§ã³ã§ã¯ãç¹å®ã®ãã³ãœã«ã®ãã®è¡šçŸã¯ããŸãæ©èœããŸãã ããããäžè¬çã«ããã®ã¿ã¹ã¯ã¯æªãã§ãã
æè¿ãéåžžã«å°æ¬ãããŠãã人ã
ã®ä»äºãçŸããŸãããããã§åœŒãã¯ãã®èãããã£ãŒããã¥ãŒã©ã«ãããã¯ãŒã¯ã§åé¢å€æ°ãšçµã³ä»ããããã€ãã®çµæã蚌æããããšããŠããŸãã å®éããããã解èªãããšããããã¯ãã³ãœã«å解ã«é¢ããéåžžã«èå³æ·±ã代æ°ççµæã§ãã çµæããã³ãœã«å±éã®èšèªã«ç¿»èš³ãããšãæšæºçãªãã³ãœã«å±éãå¥ã®ãã³ãœã«å±éãããåªããŠããããšãããããŸãã ã©ã¡ããè¯ããã¯ãŸã 瀺ããŠããŸããããæšæºçãªãã®ã§ãã ãã®æå³ã§Canonicalã¯æªãã§ãã ãã®çµè«ã¯ãæšæºçãªã¢ãŒããã¯ãã£ãæµ
ããæµ
ãããæ·±ããããåºç¯ã§ãªããã°ãªããªããªã©ã®äºå®ã«ãã£ãŠæ£åœåãããŸãã
ãããã£ãŠãããã«ã¹ã ãŒãºã«ç§»è¡ããŠã¿ãŸãããã æ£èŠå解ã®å Žåãæé©åã¢ã«ãŽãªãºã ã¯ãã£ããåæããŸããã€ãŸãã沌å°ãªã©ããããŸãããã®åŒãèŠããšãç·åœ¢ä»£æ°ã«ç²ŸéããŠãããã¹ãŠã®äººãã䜿çšã§ããã¢ã«ãŽãªãºã ãããã«ç解ã§ããŸãã åŸé
éäžæ³ã䜿çšã§ããŸã-1ã€ãé€ããã¹ãŠã®å åãä¿®æ£ããç·åœ¢æå°äºä¹åé¡ãååŸããŸãã ãããããã®æ¹æ³ã¯éåžžã«é
ããªããŸãã
ãã®ãããªã¢ãã«ã«ã¯1ã€ã®åªããç¹æ§ããããŸããããã§ã解決çãèŠã€ãã£ãå Žåãããã¯äºçŽ°ãªããšãŸã§å¯äžã®ãã®ã§ãã ããšãã°ãsãs-1ãæ¿å
¥ã§ããå Žåãè¡åå解ã®åé¡ã¯ãããŸããã æ£èŠå解ãèšç®ãããå Žåããããå¯äžã®ãã®ã§ãããããã¯è¯ãããšã§ãã ããããæ®å¿µãªãããèšç®ãé£ããå ŽåããããŸãã
Tuckerã®å¥ã®å解ããããŸãã ããã¯ããçš®ã®å¿ç枬å®ã§ããã ã¢ã€ãã¢ã¯ããã«ããçš®ã®æ¥çèŠçŽ ãå°å
¥ããããšã§ã-ãã®ãããããããã®èŠçŽ ã®éã«ã¯ã€ãªããããããŸããã ãã¹ãŠãè¯å¥œã§ããã®å解ã¯å®å®ããåžžã«æè¯ã®è¿äŒŒãååšããŸãã ããããããšãã°d = 10ã®å Žåãããã䜿çšããããšãããšãæ ŒçŽããå¿
èŠãããè£å©çãª2次å
é
åãååšãã匱äœåããã圢åŒã§ã¯ãããã次å
ã®åªããæ®ããŸãã
ã¯ããããšãã°ã3次å
ãã4次å
ã®åé¡ã®å Žåãã¿ãã«ãŒå解ã¯éåžžã«å¹æçã§ãã ãããé©çšãããäœåã¯ããªãå€æ°ãããŸãã ããã«ãããããããç§ãã¡ã®ç©¶æ¥µã®ç®æšã¯ããã§ã¯ãããŸããã
ããã©ã«ãã§ã¯ææ°é¢æ°çãªæ°ã®ãã©ã¡ãŒã¿ãŒã¯ãªããããã¹ãŠãèšç®ã§ããå ŽæãååŸãããã æåã®ã¢ã€ãã¢ã¯éåžžã«åçŽã§ããããã¹ãŠããããªãã¯ã¹ã«é©ããŠããå Žåããã³ãœã«ãå€æ¬¡å
é
åãããããªãã¯ã¹ãã¹ã«ã«ããããŸãããã ã©ããã£ãŠããã®ïŒ ç°¡åã§ãã dåã®ã€ã³ããã¯ã¹ããããŸããããããã°ã«ãŒãã«åãããããªãã¯ã¹ãäœæããå解ãã«ãŠã³ãããŸãããã
å£ããŸãã ã€ã³ããã¯ã¹ã®äžéšãå°æåã®ã€ã³ããã¯ã¹ãšããŠå®£èšããä»ã®éšåãåã€ã³ããã¯ã¹ãšããŠå®£èšããŸãã ã©ããããããåé
眮ã MATLABããã³Pythonã§ã¯ãããã¯ãã¹ãŠreshapeããã³transposeã³ãã³ãã«ãã£ãŠè¡ãããŸããäœã©ã³ã¯ã®è¿äŒŒãèæ
®ããŸãã ãã1ã€ã®è³ªåã¯ããã®æ¹æ³ã§ãã
åœç¶ãçºçããå°ããªé
åãååž°çã«äœæããããšããããšãã§ããŸãã åçŽãªæ¹æ³ã§ããããäœæãããšã次ã®è€éããçããããšãããããŸã ãr log d ã ãã¯ãææ°é¢æ°çã§ã¯ãããŸããããrã¯å€§å¹
ã«å¢å ããŸãã
ãããæ
éã«è¡ããšã1ã€ã®ãã¬ãŒãºãèšããŸãããã€ã³ããã¯ã¹ã1ã€è¿œå ãããŸãã æ°ããã€ã³ããã¯ã¹ãšèŠãªãããå¿
èŠããããŸãã 9次å
ã®é
åããããã€ã³ããã¯ã¹ã5 + 4ã«åå²ããå解ããè¿œå ã®ã€ã³ããã¯ã¹ã1ã€ã6次å
ãš5次å
ã®ãã³ãœã«ã1ã€ååŸããããããå解ãç¶ãããšããŸãã ãã®åœ¢åŒã§ã¯ãææ°é¢æ°çãªè€éãã¯çºçããããŸãã«ãã®ãããªæ°ã®ãã©ã¡ãŒã¿ãŒãæã€è¡šçŸãåŸãããŸãã ãããŠã誰ããç§ãã¡ã«ãéè·¯ã§åŸããããã¹ãŠã®è¡åãäœãã©ã³ã¯ãæã£ãŠãããšèšã£ããšãããïŒ ãã®ãããªååž°æ§é ãå€æããŸããã ããã¯éåžžã«åä»ã§ãããã°ã©ã ããã®ã¯å«ã§ããç§ã¯ãããããã®ãé¢åã§ããã
ããæç¹ã§ããã®æ§é ã®æãåçŽãªåœ¢åŒãéžæã§ããããšã«æ°ä»ããŸããããããã¯éåžžã«åŒ·åã§ãã ããããã³ãœã«ãã¬ã€ã³ããã³ãœã«ãã¬ã€ã³ã«ãªããŸããã ç§ãã¡ã¯ãããææ¡ããŸãããååã¯åºãŸããŸããã ããããããã¡ãããç§ãã¡ã¯ãããæãã€ããæåã®äººã§ã¯ãªãããšãããããŸããã åºäœç©çåŠã§ã¯ãè¡åç©ç¶æ
ãšããŠç¥ãããŠããŸããã ããããè¯ãããšã¯å°ãªããšã2åçºèŠããªããã°ãªããªãããšããã¹ãŠã®å ±åã§ç¹°ãè¿ããŸãããããªããã°ããããè¯ããã©ããã¯ããŸãæ確ã§ã¯ãããŸããã å°ãªããšãããã®ããã©ãŒãã³ã¹ã¯å°ãªããšã2åãããã«ã¯2.5åããŸãã¯3åãéãããŸããã
ããã«ã¢ã€ãã¢ããããŸããç§ã¯ãã®6ã7幎ã®éãå解ãšç 究ã«åãçµãã§ãããæåŸãŸã§ç 究ããããšã¯èšããŸããã
ãã³ãœã«ã¯ã1ã€ã®ã€ã³ããã¯ã¹ã®ã¿ã«äŸåãããªããžã§ã¯ãã®ç©ãšããŠè¡šãããŸãã å®éãå€æ°ã®åé¢ã«ã€ããŠè©±ããŠããŸããã1ã€ã®å°ããªäŸå€ããããŸãããããã®å°ããªãã®ã¯è¡åã§ãã ãããã®è¡åãä¹ç®ããŠããã€ã³ããã³ãœã«ã®å€ãååŸããŸãã æåã®èŠçŽ ã¯è¡ã次ã«ãããªãã¯ã¹ããããªãã¯ã¹ãåã§ã-æ°å€ãååŸããŸãã ãŸãããããã®è¡åã¯ãã©ã¡ãŒã¿ãŒã«äŸåãã3次å
ãã³ãœã«ãšããŠä¿åã§ããŸãã ãããªãã¯ã¹ã®ã³ã¬ã¯ã·ã§ã³ããããæåã«ã¹ãããã®ã³ã¬ã¯ã·ã§ã³ããããŸãã ç§ãã¡ã¯ããããããŒãã©ã€ã³ããããã5çªç®ã®ãããªãã¯ã¹ããããã3çªç®ã®ãããªãã¯ã¹ãå¿
èŠã§ãããšåçŽã«ä¹ç®ããŸãã æããã«ãèŠçŽ ã®èšç®ã«ã¯dr 2ã®æäœãå¿
èŠã§ãã ã€ã³ããã¯ã¹åœ¢åŒã§ãã ã³ã³ãã¯ããªå€èŠ³ã
ã»ãšãã©äœãå€ãããªãããã«èŠããŸãããå®éã«ã¯ããã®ãããªè¡šçŸã¯ç¹ç°å解ã®ãã¹ãŠã®ç¹æ§ãä¿æããŸãã èšç®ãçŽäº€åãããã³æé©ãªè¿äŒŒãèŠã€ããããšãã§ããŸãã ããã¯ããã¹ãŠã®ãã³ãœã«ã®ã»ããã®äœãã©ã¡ãŒã¿å€æ§äœã§ããã幞éã§ããã解決çãããã«ããå Žåã¯ãç°¡åã«èŠã€ããããšãã§ããŸãã
ããã€ãã®ããããã£ããããŸãã çŸåšãè€æ°ã®ã©ã³ã¯ããããŸãã ãããªãã¯ã¹ãåããŸãã¯è¡ã«ã¯1ã€ã®ã©ã³ã¯ãããããŸãããããããªãã¯ã¹ã®ã©ã³ã¯ã¯åãã§ããããã幞éã§ãã ãã³ãœã«ã«ã¯å€ãã®ã©ã³ã¯ããããŸãã å
é ã®æ°årã¯æšæºã©ã³ã¯ã§ãã ããã«ã©ã³ã¯d-1ããããŸãã dããšã«ç¬èªã®ã©ã³ã¯ããããå€ãã®ã©ã³ã¯ããããŸãã ããã«ããããããããããã®è€éãã決å®ãããã¹ãŠã®ã©ã³ã¯ãšããã®ãããªè¡šçŸãæ ŒçŽããããã®ã¡ã¢ãªã¯ãå°ãªããšãçè«çã«ã¯ãããã€ãã®ãããªãã¯ã¹ã®ã©ã³ã¯ãããããã¹ã€ãŒããšããŠèšç®ã§ããŸãã å€æ¬¡å
é
åãååŸãããããè¡åã«å€æããã©ã³ã¯ãèæ
®ããŸã-ãã®ãããªã©ã³ã¯ã«ã¯å解ããããŸãã
ãªãããããã©ãŒããããšåŒã°ããã®ã§ããïŒ ãªããžã§ã¯ãããã®åœ¢åŒã§ä¿åãããŠããå Žåãå ç®ãä¹ç®ããã«ã ã®èšç®ã®ä»£æ°æŒç®ã¯ã圢åŒãé¢ããããšãªãçŽæ¥å®è¡ã§ããŸãã ãã®å Žåããšã«ããäœããèµ·ãããŸãã ããšãã°ã2ã€ã®ãã³ãœã«ãåèšãããšãçµæã®ã©ã³ã¯ã¯ã©ã³ã¯ã®åèšãè¶
ããªããªããŸãã å埩ããã»ã¹ã§ãããäœåºŠãå®è¡ãããšãdã®äŸåæ§ãrã§ç·åœ¢ããã³2次ã§ããå Žåã§ããã©ã³ã¯ã¯éåžžã«å€§ãããªããŸãã ã©ã³ã¯100ã200ã§å®å
šã«äœæ¥ã§ããŸãã ã¯ã©ã¹ã¿ãŒã®ç©çåŠè
ã¯ã©ã³ã¯4000ã5000ã§äœæ¥ã§ããŸãããååã«å€§ããªå¯åºŠã®ãããªãã¯ã¹ãä¿åããå¿
èŠãããããã䜿çšã§ããªããªããŸããã
ããããã©ã³ã¯ãäžããããšãã§ããéåžžã«ã·ã³ãã«ã§çŸãã代æ°ã¢ã«ãŽãªãºã ããããŸãã ãã©ãããŠã¹ã®ãã«ã ã«ã¯èš±å®¹ã§ãã粟床ãããã粟床ãäžããæå°ã©ã³ã¯ãèŠã€ããããšæããŸãã ããããç§ã¯åæã«ã©ã³ã¯ãäžããŠããŸãã äžžãã æµ®åå°æ°ç¹æ°ã䜿çšããå Žåã100æ¡ãã¹ãŠãä¿åããããã§ã¯ãããŸããã 16æ¡ã ãæ®ããŸãã ãããã®è¡åã®ãµã€ãºãªã©ã蚱容ã§ãã粟床ãäžãããã©ã¡ãŒã¿ãŒãèŠã€ããŸãã ãããè¡ãä¿¡é Œæ§ã®é«ãå
ç¢ãªã¢ã«ãŽãªãºã ããããŸã-MATLABãŸãã¯Pythonã§20ã30è¡ããããŸãã
ãã1ã€çŸããããšããããŸãã ç©çåŠè
ã¯åœŒå¥³ã確å®ã«ç¥ããŸããã§ããã ç§ã¯ãã®çµæãèªãã«æããŸããããã¯ç§ã®åœæã®ããŒãã§ãããšãã²ããŒã»ãšãã²ããšãŽã£ããã»ã¿ã«ãã£ã·ã¥ãã³ããšç§ã®ãã®ã§ãã äžãããããã³ãœã«ããã®ãããªæ§é ãæã£ãŠããããšãããã£ãŠããå Žåãdnr 2èŠçŽ ãã確å®ã«åŸ©å
ã§ããŸãã ã€ãŸãããã¹ãŠã®èŠçŽ ãèšç®ããã«ãå°æ°ã®èŠçŽ ã調ã¹ãŠãåºåºã䜿çšããã«ãã³ãœã«å
šäœã埩å
ã§ããŸãã ããã¯èå³æ·±ãäºå®ã§ãããè¡åã®å Žåãç¥ãããŠããã»ã©ç¥ãããŠããŸããã ã©ã³ã¯rã®è¡åd = 2ããããŸãã råãšrè¡ã§åŸ©å
ã§ããŸãã
ãã®ãããªå€æ§æ§ãããå Žåãã©ã³ã¯ãå¶éããŠæ©èœãæå°åã§ããŸãã æšæºçãªç¶æ³ïŒèŠçŽ ã«ã¯ã¢ã¯ã»ã¹ã§ããŸããããæå°éã«æãããç¹å®ã®æ©èœããããŸãã ãã®ããã瀺ãããå¶éã§æé©åããæ¹æ³ã«ã€ããŠããŸã£ããå¥ã®ç 究ã©ã€ã³ãçºçããŸãã ããã¯åžéåã§ã¯ãããŸããããå¥åã®ç 究ãå¿
èŠãšããå€ãã®èå³æ·±ãããããžãŒç¹æ§ãæã£ãŠããŸãã
æè¿ãäž»ã«å埮åæ¹çšåŒããçããåé¡ã«ã€ããŠãã©ã³ã¯ãå¿
èŠãªç²ŸåºŠã«å¿ããŠå¯Ÿæ°çã«æ¯ãèããšããæšå®ãæçµçã«èšŒæããããšãã§ããŸãã ãã®ãããªè¡šçŸãæºæé©ã§ããåé¡ã®ã¯ã©ã¹å
šäœããããŸãã
TTã©ã³ã¯ã®ã«ãŠã³ãæ¹æ³ãèŠãŠãããŸãããã ãããªãã¯ã¹ãååŸããæåã®kåã®ã€ã³ããã¯ã¹ãè¡ã€ã³ããã¯ã¹ãæåŸã®dkãåã€ã³ããã¯ã¹ãšèŠãªããŸãã , . .
. , . . â d-1 , . , 10-6, .
d . â . , . 1010, 100100. , 1000010000, .
dnr 3 . , - .
. - 100 , 10 , .
. .
. , . . , , . . , .
r, r r , . , , , .
, , . , , . â , 5, 5 5 , .
, . , .
, , , . - . , . , , . . : , .
. , -, . , . , , . . , .
, , .
â , . , d-. d- , . . 2 d 2222. , . . , , , .
â , - , â , - .
. . , . , , , . â .
: ttpy Python TT-Toolbox MATLAB. , , , , . , , .
. , , , . , , - , . - . , : . 10 -6 10 -8 . â , â . â , . , , , . , , , - . èŠãŠã¿ãŸãããã
, - , . TensorFlow . , â , . - , . , , .
, . . 2 60 . . , n 3 , n = 2 d , 220, . 2222, 60- . .
, : 2 d, 60, , 50. , .
, . , , , , . .
, . TensorNet, , NIPS . , . , , .
. . , .
, . , , . , . , , : , .
. : , , , , . , , . , . , â , . , . , , . -, . 1500 â 10 5 . MATLAB . .
. 12 , 12- , 30 12 . .
Exponential Machines, x1, x2, , 2 d . 2222 : . åäœããŸãã , , .
. , , . , - â , . RecSys. Polara. , , . : . : , « », SVD « ». « » . . : « â » « â â » , . - . , , , - : , , , . , .
, â . . , deep learning. , arithmetic circuit. , .
, GitHub , . , .