Denoising Autoencoderã¨ãã®ä¸€èˆ¬åŒ–Machine Learning Advenc Calendar 2013ã®23日目担当ã®å¾—å±…ã§ã™ã€‚ æ ªå¼ä¼šç¤¾Preferred Infrastructureã§Jubatusを作ã£ãŸã‚Šã—ã¦ã„ã¾ã™ã€‚ 今日ã¯æ·±å±¤å¦ç¿’(deep learning)ã®è©±ã§ã™ã€‚ 深層å¦ç¿’ã¯ã“ã®2å¹´ã»ã©ã§å°‚門外ã®äººã«ã‚‚知れ渡るã»ã©ã«å¤§æµè¡Œã—ã¾ã—ãŸã€‚ データã•ãˆå¤§é‡ã«ã‚ã‚Œã°ãƒ†ã‚¯ãƒ‹ãƒƒã‚¯æ¬¡ç¬¬ã§ä»–ã®æ‰‹æ³•ã‚’圧倒ã™ã‚‹æ€§èƒ½ã‚’é”æˆã§ãã‚‹ã“ã¨ã‹ã‚‰ã€ç‰¹ã«å¤§é‡ã®ãƒ‡ãƒ¼ã‚¿ã‚’æŒã¤å¤§ä¼æ¥ã«ãŠã„ã¦å¤§ã€…çš„ãªå‚å…¥ãŒç›¸æ¬¡ãŽã¾ã—ãŸã€‚ 主ã«æµè¡Œã£ã¦ã„ã‚‹ã®ã¯æ•™å¸«ã‚ã‚Šå¦ç¿’ã§ã™ã€‚ 補助ã¨ã—ã¦æ•™å¸«ãªã—å¦ç¿’ã«ã‚ˆã‚‹äº‹å‰å¦ç¿’(pretraining)ã‚‚ã€ç‰¹ã«éŸ³å£°èªè˜ã®ã‚¿ã‚¹ã‚¯ã«ãŠã„ã¦ã¯è¡Œã‚ã‚Œã¦ã„るよã†ã§ã™ãŒã€ç”»åƒèªè˜ã‚’ä¸å¿ƒã«äº‹å‰å¦ç¿’ãªã—ã§ã‚‚テクニック次第ã§å¦ç¿’ã§ãã‚‹ã¨ã„ã†è¦‹æ–¹ãŒå¼·ã¾ã£ã¦ã„ã¾ã™ã€‚ 一方ã§æ•™å¸«ãªã—データã‹ã‚‰ã®å¦ç¿’ã¯ãƒ–レイクスルー待ã¡ã¨ã„ã£ãŸé›°å›²æ°—ã§ã™ã€‚ Deep
Path-Normalized Optimization of Recurrent Neural Networks with ReLU Activations
zero-shot learning - RupyWiki
WELQ
h2o-training-book/anomaly_detection.md at master · h2oai/h2o-training-book · GitHub
異常検知ã®ä¸–ç•Œã¸ã‚ˆã†ã“ã - Preferred Networks Research & Development
生æˆãƒ¢ãƒ‡ãƒ«ã® Deep Learning
{{#tags}}- {{label}}
{{/tags}}