[B! Keras] sugyanã®ãƒ–ãƒƒã‚¯ãƒžãƒ¼ã‚¯

sugyan id:sugyan

Kerasã«é–¢ã™ã‚‹sugyanã®ãƒ–ãƒƒã‚¯ãƒžãƒ¼ã‚¯ (9)

${{author_name}}$

{{{comment_expanded}}}

{{label}}

{{#is_bookmark}}ãƒªã‚¹ãƒˆ{{/is_bookmark}}{{^is_bookmark}}ãƒªãƒ³ã‚¯{{/is_bookmark}}

${{author_name}}$
{{author_name}}{{created}}
{{ #comment }}{{ comment }}{{ /comment }}
- {{ label }}

${{author_name}}$

{{{comment_expanded}}}

{{label}}

{{#is_bookmark}}ãƒªã‚¹ãƒˆ{{/is_bookmark}}{{^is_bookmark}}ãƒªãƒ³ã‚¯{{/is_bookmark}}

ãƒ‡ã‚£ãƒ¼ãƒ—ãƒ©ãƒ¼ãƒ‹ãƒ³ã‚°ã§å‹•ç”»ã«è‡ªå‹•ã§ãƒ¢ã‚¶ã‚¤ã‚¯ã‚’ã‹ã‘ã‚‹ã€Œãƒ‡ã‚£ãƒ¼ãƒ—ãƒ¢ã‚¶ã‚¤ã‚¯ã€ä½œã£ã¦ã¿ã¾ã—ãŸ - karaage. [ã‹ã‚‰ã‚ã’]
ãƒ‡ã‚£ãƒ¼ãƒ—ãƒ©ãƒ¼ãƒ‹ãƒ³ã‚°ã§ã®ç‰©ä½“æ¤œå‡º ä»¥å‰ãƒ‡ã‚£ãƒ¼ãƒ—ãƒ©ãƒ¼ãƒ‹ãƒ³ã‚°ã‚’ä½¿ã£ãŸç‰©ä½“æ¤œå‡ºã‚’è©¦ã—ã¦ã¿ã¾ã—ãŸã€‚ å‰å›žã®æˆæžœç‰©ã¯ä»¥ä¸‹ã§ã™ã€‚ ãŸã ã€ã“ã®ã¾ã¾ã§ã¯ã¡ã‚‡ã£ã¨é¢ç™½ã„ã ã‘ã§ä½•ã®å½¹ã«ã‚‚ç«‹ã¡ã¾ã›ã‚“ã€‚æŠ˜è§’ãªã®ã§ä½•ã‹è‡ªå‰ã®ãƒ‡ãƒ¼ã‚¿ã®å¦ç¿’ã‚’ã•ã›ã¦ã€ä¸–ã®ãŸã‚äººã®ãŸã‚ã«ãªã‚‹ã‚‚ã®ã‚’ä½œã£ã¦ã¿ãŸã„ã§ã™ãï¼ˆç„¡è¡¨æƒ…ï¼‰ã€‚ ãµã¨æ€ã„å‡ºã—ãŸã®ã¯ã€ã‚¢ãƒ€ãƒ«ãƒ‰ãƒ“ãƒ‡ã‚ªã®ãƒ¢ã‚¶ã‚¤ã‚¯å…¥ã‚Œã§ã™ã€‚ã‚ã®å‡¦ç†ã£ã¦ã€å®Ÿã¯è‡ªå‹•åŒ–ã•ã‚Œã¦ã„ãªãã¦ã€è·äººã•ã‚“ãŒ1ãƒ•ãƒ¬ãƒ¼ãƒ æ¯Žã«ãƒ¢ã‚¶ã‚¤ã‚¯ã‚’å…¥ã‚Œã¦ã„ã‚‹ã‚‰ã—ã„ã®ã§ã™ã€‚ã“ã®ä½œæ¥ã‚’è‡ªå‹•åŒ–ã™ã‚Œã°ã€ãã£ã¨ã‚‚ã£ã¨ä¸–ã®ä¸ã«ç´ æ•µãªå‹•ç”»ãŒæº¢ã‚Œã¦ã¿ã‚“ãªãŒå¹¸ã›ã«ãªã‚‹ã¯ãšï¼ ã¨ã„ã†ã‚ã‘ã§ã€ãƒ¢ã‚¶ã‚¤ã‚¯ã‚’è‡ªå‹•ã§å…¥ã‚Œã‚‹ã‚½ãƒ•ãƒˆã‚’ä½œã‚Œãªã„ã‹è©¦ã—ã¦ã¿ã‚‹ã“ã¨ã«ã—ã¾ã—ãŸã€‚åä»˜ã‘ã¦ã€Œãƒ‡ã‚£ãƒ¼ãƒ—ãƒ¢ã‚¶ã‚¤ã‚¯ã€ã§ã™ã€‚é€”ä¸å®‡å®™èªžãŒç¶šãã®ã§ã€çµæžœãŒæ°—ã«ãªã‚‹æ–¹ã¯æœ€å¾Œã ã‘è¦‹ã¦ä¸‹ã•ã„ã€‚ ãƒ‡ã‚£ãƒ¼ãƒ—ãƒ©ãƒ¼ãƒ‹ãƒ³ã‚°ï¼ˆSSDï¼‰ã‚’ä½¿ã£ã¦ãƒ¢ã‚¶ã‚¤ã‚¯ã‚’ã‹ã‘ã‚‹æ–¹æ³• è¿½è¨˜ï¼š ã“ã®è¨˜äº‹ã§ã¯ã€VoTTã®å¤ã„ãƒãƒ¼ã‚¸ãƒ§ãƒ³ï¼ˆ
sugyan 2018/06/05
SSD

Keras

VoTT
ãƒªãƒ³ã‚¯
[iOS 11] Core MLã§ç„¼ãé³¥ã‚’æ©Ÿæ¢°å¦ç¿’ã•ã›ã¦ã¿ãŸ
Core ML Core MLã¯ã€å¦ç¿’ãƒ¢ãƒ‡ãƒ«ç‰ã‚’iOS / macOSä¸Šã§åˆ©ç”¨ã™ã‚‹éš›ã«ã€é–‹ç™ºè€…ãŒå°‚é–€çš„ãªçŸ¥è˜ã‚’å¿…è¦ã¨ã›ãšã«æ‰±ãˆã‚‹ã‚ˆã†ã«è£œåŠ©ã™ã‚‹ãƒ•ãƒ¬ãƒ¼ãƒ ãƒ¯ãƒ¼ã‚¯ã§ã™ã€‚WWDC2017ã§ç™ºè¡¨ã•ã‚ŒãŸéš›ã¯ã€ARKitã«ä¸¦ã‚“ã§ç¾åœ°ã§åéŸ¿ã®å¤§ãã‹ã£ãŸç™ºè¡¨ã§ã—ãŸã€‚ iOSã§ã¯ã€Core MLç™ºè¡¨ä»¥å‰ã‹ã‚‰æ©Ÿæ¢°å¦ç¿’ã‚’å–ã‚Šå…¥ã‚Œã‚ˆã†ã¨ã¯ã—ã¦ã„ã¾ã—ãŸã€‚ãŸã ã€ãã®ãŸã‚ã«ã¯CIImageã‚’åˆ©ç”¨ã—ãŸç”»åƒå‡¦ç†ã‚„ã€å¦ç¿’ãƒ¢ãƒ‡ãƒ«ã¸ã®å…¥å‡ºåŠ›ãªã©ã®ã‚³ãƒ¼ãƒ‡ã‚£ãƒ³ã‚°ã¨ã„ã£ãŸå°‚é–€çš„ãªçŸ¥è˜ãŒå¿…è¦ã§ã—ãŸã€‚ Core MLã®ç™»å ´ã§å¦ç¿’ãƒ¢ãƒ‡ãƒ«ã‚’ç¢ºä¿ã§ãã‚Œã°ã€ä»¥ä¸‹ã®æ©Ÿèƒ½ã‚’ä½¿ã£ã¦iOS / macOSã‚¢ãƒ—ãƒªã§æ©Ÿæ¢°å¦ç¿’ã—ã‚„ã™ããªã‚Šã¾ã™ã€‚ å¦ç¿’ãƒ¢ãƒ‡ãƒ«ã‚’iOS / macOSã‚¢ãƒ—ãƒªã§æ‰±ãˆã‚‹å½¢å¼ã«å¤‰æ›ã™ã‚‹ï¼ˆCore ML modelï¼‰ Core ML modelã¸ã®å…¥åŠ›ã¨å‡ºåŠ›ã‚’è£œåŠ©ã™ã‚‹ãƒ•ãƒ¬ãƒ¼ãƒ ãƒ¯ãƒ¼ã‚¯ ã‚¢ãƒ—ãƒªå†…ã«çµ„ã¿è¾¼ã¿ã€å¦ç¿’çµæžœã‚’å—ã‘å–ã‚‹ã“ã¨ã«ç‰¹åŒ–ã—ãŸCore
sugyan 2018/04/05
CoreML

iOS

coremltools

Keras
ãƒªãƒ³ã‚¯
Pythonã¨Kerasã‚’ä½¿ã£ã¦AlphaZero AIã‚’è‡ªä½œã™ã‚‹ | POSTD
è‡ªå·±å¯¾æˆ¦ã¨æ·±å±¤å¦ç¿’ã§ãƒžã‚·ãƒ³ã«ã‚³ãƒã‚¯ãƒˆãƒ•ã‚©ãƒ¼ï¼ˆConnect4ï¼šå››ç›®ä¸¦ã¹ï¼‰ã®æˆ¦ç•¥ã‚’å¦ç¿’ã•ã›ã¾ã—ã‚‡ã†ã€‚ ã“ã®è¨˜äº‹ã§ã¯æ¬¡ã®3ã¤ã®è©±ã‚’ã—ã¾ã™ã€‚ AlphaZeroãŒäººå·¥çŸ¥èƒ½ï¼ˆAIï¼‰ã¸ã®å¤§ããªã‚¹ãƒ†ãƒƒãƒ—ã§ã‚ã‚‹2ã¤ã®ç†ç”± AlphaZeroã®æ–¹æ³•è«–ã®ãƒ¬ãƒ—ãƒªã‚«ã‚’ ä½œã£ã¦ ã‚³ãƒã‚¯ãƒˆ4ã®ã‚²ãƒ¼ãƒ ã‚’ãƒ—ãƒ¬ã‚¤ã•ã›ã‚‹æ–¹æ³• ãã®ãƒ¬ãƒ—ãƒªã‚«ã‚’æ”¹è‰¯ã—ã¦ä»–ã®ã‚²ãƒ¼ãƒ ã‚’ãƒ—ãƒ©ã‚°ã‚¤ãƒ³ã™ã‚‹æ–¹æ³• Alpha Goâ†’Alpha Go Zeroâ†’AlphaZero 2016å¹´3æœˆã€Deepmindã®Alpha Goï¼ˆã‚¢ãƒ«ãƒ•ã‚¡ç¢ï¼‰ãŒã€å›²ç¢ã®18å›žã®ä¸–ç•ŒçŽ‹è€…ã€æŽä¸–ä¹ï¼ˆã‚¤ãƒ¼ãƒ»ã‚»ãƒ‰ãƒ«ï¼‰ã¨ã®äº”ç•ªå‹è² ã§ã€2å„„äººã®è¦‹å®ˆã‚‹ä¸ã€4-1ã§å‹åˆ©ã—ã¾ã—ãŸã€‚æ©Ÿæ¢°ãŒè¶…äººçš„ãªå›²ç¢ã®æŠ€ã‚’å¦ç¿’ã—ãŸã®ã§ã™ã€‚ä¸å¯èƒ½ã ã¨ã‹ã€å°‘ãªãã¨ã‚‚10å¹´é–“ã¯é”æˆã§ããªã„ã¨æ€ã‚ã‚Œã¦ã„ãŸå‰æ¥ã§ã™ã€‚ Alpha Goã€€å¯¾ã€€æŽä¸–ä¹ã®ç¬¬3å±€ ã“ã®ã“ã¨ã ã‘ã§ã‚‚é©šãã¹ãåŠŸç¸¾ã§ã™ãŒã€DeepMindã¯ã€2017å¹´10æœˆã€
sugyan 2018/03/30
AlphaZero

Keras
ãƒªãƒ³ã‚¯
ãƒ¢ãƒ‡ãƒ«åœ§ç¸® - nico-opendata
nico-opendataniconicoã§ã¯ã€å¦è¡“åˆ†é‡Žã«ãŠã‘ã‚‹æŠ€è¡“ç™ºå±•ã¸ã®å¯„ä¸Žã‚’ç›®çš„ã¨ã—ã¦ã€ ç ”ç©¶è€…ã®æ–¹ã‚’å¯¾è±¡ã«å„ç¨®ã‚µãƒ¼ãƒ“ã‚¹ã®ãƒ‡ãƒ¼ã‚¿ã‚’å…¬é–‹ã—ã¦ã„ã¾ã™ã€‚ ãƒ‹ã‚³ãƒ‹ã‚³å‹•ç”»ã‚³ãƒ¡ãƒ³ãƒˆç‰ãƒ‡ãƒ¼ã‚¿ã‚»ãƒƒãƒˆ(æ ª)ãƒ‰ãƒ¯ãƒ³ã‚´åŠã³(æœ‰)æœªæ¥æ¤œç´¢ãƒ–ãƒ©ã‚¸ãƒ«ã¨å›½ç«‹æƒ…å ±å¦ç ”ç©¶æ‰€ãŒå”åŠ›ã—ã¦ç ”ç©¶è€…ã«æä¾›ã—ã¦ã„ã‚‹ãƒ‡ãƒ¼ã‚¿ã‚»ãƒƒãƒˆã§ã™ã€‚ ãƒ‹ã‚³ãƒ‹ã‚³å‹•ç”»ã‚³ãƒ¡ãƒ³ãƒˆç‰ã®ãƒ‡ãƒ¼ã‚¿ãŒåˆ©ç”¨å¯èƒ½ã§ã™ã€‚ åˆ©ç”¨ç”³è«‹ãƒ•ã‚©ãƒ¼ãƒ â€»å›½ç«‹æƒ…å ±å¦ç ”ç©¶æ‰€ã¸ãƒªãƒ³ã‚¯ã—ã¾ã™ ãƒ‹ã‚³ãƒ‹ã‚³å¤§ç™¾ç§‘ãƒ‡ãƒ¼ã‚¿(æ ª)ãƒ‰ãƒ¯ãƒ³ã‚´åŠã³(æœ‰)æœªæ¥æ¤œç´¢ãƒ–ãƒ©ã‚¸ãƒ«ã¨å›½ç«‹æƒ…å ±å¦ç ”ç©¶æ‰€ãŒå”åŠ›ã—ã¦ç ”ç©¶è€…ã«æä¾›ã—ã¦ã„ã‚‹ãƒ‡ãƒ¼ã‚¿ã‚»ãƒƒãƒˆã§ã™ã€‚ ãƒ‹ã‚³ãƒ‹ã‚³å¤§ç™¾ç§‘ã®ãƒ‡ãƒ¼ã‚¿ãŒåˆ©ç”¨å¯èƒ½ã§ã™ã€‚ åˆ©ç”¨ç”³è«‹ãƒ•ã‚©ãƒ¼ãƒ â€»å›½ç«‹æƒ…å ±å¦ç ”ç©¶æ‰€ã¸ãƒªãƒ³ã‚¯ã—ã¾ã™ Nico-Illustãƒ‡ãƒ¼ã‚¿ã‚»ãƒƒãƒˆComicolorization: Semi-Automatic Manga ColorizationChie Furusawa*ã€Kazuyuki Hi
sugyan 2017/05/19
ãƒ¢ãƒ‡ãƒ«åœ§ç¸®

Keras
ãƒªãƒ³ã‚¯
ç•³ã¿è¾¼ã¿ãƒ‹ãƒ¥ãƒ¼ãƒ©ãƒ«ãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯ã®å¯è¦–åŒ– - äººå·¥çŸ¥èƒ½ã«é–¢ã™ã‚‹æ–å‰µéŒ²
Deep Learningã®å¦ç¿’çµæžœï¼ˆé‡ã¿ï¼‰ã¯ãƒ–ãƒ©ãƒƒã‚¯ãƒœãƒƒã‚¯ã‚¹ã§ã€éš ã‚Œå±¤ã®ãƒ¦ãƒ‹ãƒƒãƒˆï¼ˆç‰¹ã«æ·±ã„å±¤ã®ï¼ï¼‰ãŒä¸€ä½“ä½•ã‚’å¦ç¿’ã—ãŸã®ã‹ãŒã‚ˆãã‚ã‹ã‚‰ãªã„ã¨é•·å¹´è¨€ã‚ã‚Œã¦ããŸã€‚ã—ã‹ã—ã€ä»Šå›žç´¹ä»‹ã™ã‚‹æ–¹æ³•ã‚’ä½¿ã†ã¨ãƒ‹ãƒ¥ãƒ¼ãƒ©ãƒ«ãƒãƒƒãƒˆãŒä½•ã‚’å¦ç¿’ã—ãŸã®ã‹ç›®ã§è¦‹ãˆã‚‹å½¢ã§è¡¨ç¾ã§ãã‚‹ã€‚ ç•³ã¿è¾¼ã¿ãƒ‹ãƒ¥ãƒ¼ãƒ©ãƒ«ãƒãƒƒãƒˆã§å¦ç¿’ã—ãŸãƒ•ã‚£ãƒ«ã‚¿ã®å¯è¦–åŒ–ã¨ã„ã†ã¨ä»¥å‰ã‚„ã£ãŸã‚ˆã†ã«å¦ç¿’ã—ãŸç¬¬1å±¤ã®ãƒ•ã‚£ãƒ«ã‚¿ã®é‡ã¿ã‚’ç›´æŽ¥ç”»åƒã¨ã—ã¦å¯è¦–åŒ–ã™ã‚‹æ–¹æ³•ãŒã‚ã‚‹ã€‚ ã—ã‹ã—ã€ç•³ã¿è¾¼ã¿ãƒ•ã‚£ãƒ«ã‚¿ã®ã‚µã‚¤ã‚ºã¯åŸºæœ¬çš„ã«æ•°ãƒ”ã‚¯ã‚»ãƒ«ï¼ˆMNISTã®ä¾‹ã§ã¯5x5ãƒ”ã‚¯ã‚»ãƒ«ç¨‹åº¦ï¼‰ã®ã¨ã¦ã‚‚å°ã•ãªç”»åƒãªã®ã§ã“ã‚Œã‚’ç›´æŽ¥å¯è¦–åŒ–ã—ã¦ã‚‚ä½•ãŒå¦ç¿’ã•ã‚ŒãŸã‹æŠŠæ¡ã™ã‚‹ã®ã¯ã¨ã¦ã‚‚é›£ã—ã„ã€‚ãŸã¨ãˆã°ã€MNISTã‚’å¦ç¿’ã—ãŸç•³ã¿è¾¼ã¿ãƒ‹ãƒ¥ãƒ¼ãƒ©ãƒ«ãƒãƒƒãƒˆï¼ˆ2016/11/20ï¼‰ã®ãƒ•ã‚£ãƒ«ã‚¿ã‚’å¯è¦–åŒ–ã—ã¦ã‚‚å„ãƒ•ã‚£ãƒ«ã‚¿ãŒã©ã®æ–¹å‘ã«åå¿œã—ã‚„ã™ã„ã‹ãŒã‚ã‹ã‚‹ç¨‹åº¦ã ã€‚ å„ãƒ•ã‚£ãƒ«ã‚¿ãŒä½•ã‚’å¦ç¿’ã—ãŸã‹ã‚’å¯è¦–åŒ–ã™ã‚‹åˆ¥ã®ã‚¢ãƒ—ãƒãƒ¼ãƒã¨ã—ã¦å„ãƒ•ã‚£ãƒ«ã‚¿
sugyan 2017/02/17
é¢ç™½ã„ã€‚ã‚¢ã‚¤ãƒ‰ãƒ«é¡”è˜åˆ¥å™¨ã§é¨™ã™ã®ã¯è©¦ã—ãŸã‘ã©ã‚ãƒ¬ã‚¤ãªå…¥åŠ›ã¤ãã‚‹ã®å‡ºæ¥ãªã‹ã£ãŸã‹ã‚‰ã¾ãŸã‚„ã£ã¦ã¿ã‚‹ã‹â€¦

Keras

CNN

visualization
ãƒªãƒ³ã‚¯
æ‰‹æ›¸ãã²ã‚‰ãŒãªã®èªè˜ã§99.78%ã®ç²¾åº¦ã‚’ãƒ‡ã‚£ãƒ¼ãƒ—ãƒ©ãƒ¼ãƒ‹ãƒ³ã‚°ã§ - Qiita
Deleted articles cannot be recovered. Draft of this article would be also deleted. Are you sure you want to delete this article? æ‰‹æ›¸ãã²ã‚‰ãŒãªã®èªè˜ã€æ•™ç§‘æ›¸é€šã‚Šã®ç•³ã¿è¾¼ã¿ãƒ‹ãƒ¥ãƒ¼ãƒ©ãƒ«ãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯ï¼ˆãƒ‡ã‚£ãƒ¼ãƒ—ãƒ©ãƒ¼ãƒ‹ãƒ³ã‚°ï¼‰ã§ã€99.78%ã®ç²¾åº¦ãŒå‡ºã¾ã—ãŸã€‚æ•™ç§‘æ›¸é€šã‚Šã§ã‚ã‚‹äº‹ãŒï¼ˆç‹¬è‡ªæ€§ãŒãªã„äº‹ãŒï¼‰é€†ã«èªã‚€æ–¹ãƒ»æ›¸ãæ–¹ã«ãƒ¡ãƒªãƒƒãƒˆãŒã‚ã‚Šãã†ãªã®ã§ã€Qiitaã§è¨˜äº‹ã«ã—ã¾ã™ã€‚ ã‚½ãƒ¼ã‚¹ã‚³ãƒ¼ãƒ‰ ã‚½ãƒ¼ã‚¹ã‚³ãƒ¼ãƒ‰ã¯ https://github.com/yukoba/Cnn JapaneseCharacter ã§ã™ã€‚ ã„ãã•ã¤ ã„ãã•ã¤ã¯ã€å‹äººã¨æ‰‹æ›¸ãã²ã‚‰ãŒãªã®èªè˜ã®é›‘è«‡ã‚’Facebookã§ã—ã¦ã„ã¦ã€ã‚°ã‚°ã£ãŸã‚‰ã€ã“ã®2ã¤ãŒè¦‹ã¤ã‹ã‚Šã¾ã—ãŸã€‚ ã€ŒTensorflowã‚’ï¼’ãƒ¶æœˆè§¦ã£ãŸã®ã§"æ‰‹æ›¸ãã²ã‚‰
sugyan 2016/10/04
DeepLearning

Keras

Python
ãƒªãƒ³ã‚¯
Kerasã§å¦ã¶Autoencoder
Kerasã®å…¬å¼ãƒ–ãƒã‚°ã«Autoencoderï¼ˆè‡ªå·±ç¬¦å·åŒ–å™¨ï¼‰ã«é–¢ã™ã‚‹è¨˜äº‹ãŒã‚ã‚Šã¾ã™ã€‚ä»Šå›žã¯ã“ã®è¨˜äº‹ã®æµã‚Œã«æ²¿ã£ã¦å®Ÿè£…ã—ã¤ã¤ã€Autoencoderã®è§£èª¬ã‚’ã—ã¦ã„ããŸã„ã¨æ€ã„ã¾ã™ã€‚é–“é•ã„ãŒã‚ã‚Œã°æŒ‡æ‘˜ã—ã¦ä¸‹ã•ã„ã€‚ã¾ãŸã€Kerasã®å…¬å¼ãƒ–ãƒã‚°ã¯Kerasã§ã®å®Ÿè£…ã«é–¢ã—ã¦ã ã‘ã§ãªãã€æ©Ÿæ¢°å¦ç¿’è‡ªä½“ã«ã¤ã„ã¦ã‚‚å‹‰å¼·ã«ãªã‚‹ã“ã¨ãŒå¤šãã€éžå¸¸ã«ãŠã™ã™ã‚ã§ã™ã€‚ ä»Šå›žã®è¨˜äº‹ã§ã¯æ§˜ã€…ãªã‚¿ã‚¤ãƒ—ã®Autoencoderã‚’ç´¹ä»‹ã—ã¾ã™ãŒã€æ—¥æœ¬èªžã§ã¯ã‚ã¾ã‚Šèžãæ…£ã‚Œãªã„ã‚‚ã®ã‚‚ã‚ã‚‹ã‹ã¨æ€ã„ã¾ã™ã®ã§ã€ä»Šå›žã¯åç§°ã‚’è‹±èªžã§çµ±ä¸€ã—ãŸã„ã¨æ€ã„ã¾ã™ã€‚ ç›®æ¬¡ ã‚¤ãƒ³ãƒˆãƒãƒ€ã‚¯ã‚·ãƒ§ãƒ³ Undercomplete Autoencoder Sparse Autoencoder Deep Autoencoder Convolutional Autoencoder Denoising Autoencoder ã¾ã¨ã‚ ã‚¤ãƒ³ãƒˆãƒãƒ€ã‚¯ã‚·ãƒ§ãƒ³ Autoencod
sugyan 2016/07/20
AutoEncoder

Keras
ãƒªãƒ³ã‚¯
DQNã‚’Kerasã¨TensorFlowã¨OpenAI Gymã§å®Ÿè£…ã™ã‚‹
ã¯ã˜ã‚ã« å°‘ã—æ™‚ä»£é…ã‚Œã‹ã‚‚ã—ã‚Œã¾ã›ã‚“ãŒã€å¼·åŒ–å¦ç¿’ã®æ‰‹æ³•ã®ã²ã¨ã¤ã§ã‚ã‚‹DQNã‚’DeepMindã®è«–æ–‡Mnih et al., 2015, Human-level control through deep reinforcement learningã‚’å‚è€ƒã«ã—ãªãŒã‚‰ã€Kerasã¨TensorFlowã¨OpenAI Gymã‚’ä½¿ã£ã¦å®Ÿè£…ã—ã¾ã™ã€‚ å‰åŠã§ã¯è»½ãDQNã®ãŠã•ã‚‰ã„ã‚’ã—ã¾ã™ãŒã€å°‘ã—ã®å¼·åŒ–å¦ç¿’ã®çŸ¥è˜ã‚’æŒã£ã¦ã„ã‚‹ã“ã¨ã‚’å‰æã«ã—ã¦ã„ã¾ã™ã€‚ ã™ã§ã«ã„ãã¤ã‹è‰¯è¨˜äº‹ãŒå‡ºã¦ã„ã‚‹ã®ã§ç´¹ä»‹ã—ãŸã„ã¨æ€ã„ã¾ã™ã€‚åˆã‚ã›ã¦èªã‚€ã¨ç†è§£ã®åŠ©ã‘ã«ãªã‚‹ã¨æ€ã†ã®ã§ã€æ˜¯éžå‚è€ƒã«ã—ã¦ã¿ã¦ãã ã•ã„ã€‚ DQNã®ç”Ÿã„ç«‹ã¡ã€€ï¼‹ã€€Deep Q-Networkã‚’Chainerã§æ›¸ã„ãŸ DQNãŒç”Ÿã¾ã‚ŒãŸèƒŒæ™¯ã«ã¤ã„ã¦èª¬æ˜Žã—ã¦ãã‚Œã¦ã„ã¾ã™ã€‚Chainerã§ã®å®Ÿè£…ã‚‚ã‚ã‚‹ãã†ã§ã™ã€‚ ã‚¼ãƒã‹ã‚‰Deepã¾ã§å¦ã¶å¼·åŒ–å¦ç¿’ ã‚¿ã‚¤ãƒˆãƒ«ã®é€šã‚Šã€ã‚¼ãƒã‹ã‚‰Deepã¾
sugyan 2016/06/30
DQN

Keras

TensorFlow

ReinforcementLearning
ãƒªãƒ³ã‚¯
Kerasã§å¦ã¶è»¢ç§»å¦ç¿’
å‰å›žè¨˜äº‹ã§ã¯ã€Kaggleã®Facial Keypoints Detectionã‚’é¡Œæã«ã—ã¦ã€å˜ç´”ãªãƒ‹ãƒ¥ãƒ¼ãƒ©ãƒ«ãƒãƒƒãƒˆãƒ¯ãƒ¼ã‚¯ã‹ã‚‰è»¢ç§»å¦ç¿’ã¾ã§è§£èª¬ã—ã¾ã—ãŸã€‚ äº‹å‰ã«å¦ç¿’ã—ãŸé‡ã¿ã‚’èªã¿è¾¼ã‚“ã å¾Œã€å…¨ã¦ã®å±¤ã§å¦ç¿’ã™ã‚‹ã®ã§ã¯ãªãã€ä¸€éƒ¨ã®å±¤ã‚’ãƒ•ãƒªãƒ¼ã‚ºã•ã›ã‚‹ã“ã¨ã‚‚ã§ãã‚‹ã¨ã„ã†è©±ã‚’æœ€å¾Œã«å°‘ã—ã ã‘ã—ã¾ã—ãŸã€‚ã¡ã‚‡ã†ã©ãã®å¾Œã€è»¢ç§»å¦ç¿’ã«ã¤ã„ã¦è©³ç´°ã«è§£èª¬ã—ã¦ã„ã‚‹Kerasã®å…¬å¼ãƒ–ãƒã‚°è¨˜äº‹ãŒå…¬é–‹ã•ã‚Œã¾ã—ãŸã“ã¨ã‚‚ã‚ã‚Šã€ä»Šå›žã¯ã“ã®è¨˜äº‹ã‚’å‚è€ƒã«ã—ã¤ã¤ã€è»¢ç§»å¦ç¿’ã‚’ãƒ¡ã‚¤ãƒ³ã«è§£èª¬ã—ã¦ã„ãã¾ã™ã€‚é–“é•ã„ãŒã‚ã‚Œã°æŒ‡æ‘˜ã—ã¦ãã ã•ã„ã€‚ä»Šå›žã‚‚Facial Keypoints Detectionã®ãƒ‡ãƒ¼ã‚¿ã‚’ä½¿ã£ã¦è§£èª¬ã—ã¦ã„ãã®ã§ã€å‰å›žè¨˜äº‹ã‚‚æ˜¯éžåˆã‚ã›ã¦èªã‚“ã§ã¿ã¦ãã ã•ã„ã€‚ ã¾ãŸã€Keras 1.0.4ãŒå…¬é–‹ã•ã‚ŒãŸã®ã§ã¾ã ã®äººã¯ã‚¢ãƒƒãƒ—ãƒ‡ãƒ¼ãƒˆã—ã¦ãŠãã¨è‰¯ã„ã‹ã¨æ€ã„ã¾ã™ã€‚ ç›®æ¬¡ è»¢ç§»å¦ç¿’ å¯è¦–åŒ– å…¨çµåˆå±¤ã®ã¿å¦ç¿’ï¼ˆå‰å›žãƒ¢ãƒ‡ãƒ«ï¼‰ å…¨çµåˆå±¤+ä¸€éƒ¨ã®ç•³ã¿è¾¼
sugyan 2016/06/23
ã‚³ãƒ¼ãƒ‰ã¯Kerasã ã‘ã©èª¬æ˜ŽãŒã¨ã¦ã‚‚ä¸å¯§ã ã—å‰å›žã®ã¨åˆã‚ã›ã¦é¢ç™½ã„

Keras

DeepLearning

CNN
ãƒªãƒ³ã‚¯
1