模式识别 - 南京大学计算机科学与技术系.pdfVIP

妯″紡璇嗗埆 鎺ㄧ悊鍜屽喅绛?Inference and Decision) 瀵嗗害浼拌(Density Estimation) 鍚村缓閼 鍗椾含澶у璁＄畻鏈虹郴锛?017 1 鐩爣 飪 鐞嗚В骞舵帉鎻″悇鍚嶈瘝鐨勫惈涔夛紝鏈€浣庤姹傛槸浣跨敤杩欎簺鍚 璇嶇殑鏃跺€欎笉鑳芥贩娣 飪 鎺屾彙鍩烘湰鐨勭偣浼拌鏂规硶锛岃兘鍦ㄤ笉鏌ヨ〃锛堢畝鍗曟儏鍐碉級 鎴栨煡琛紙澶嶆潅鎯呭喌锛変笅鑷富鎺ㄥ浼拌缁撴灉 飪 鎺屾彙Bayesian鐩稿叧鏂规硶鐨勬蹇 飪 鐞嗚В銆佹帉鎻￠潪鍙傛暟浼拌鐨勬柟娉 飪 鎻愰珮鐩爣 飩 杩涗竴姝ヨ兘閫氳繃鐙珛闃呰銆佷簡瑙ayesian鏂规硶 飩 杩涗竴姝ヨ兘閫氳繃鐙珛闃呰銆佷簡瑙ｉ潪鍙傛暟浼拌锛堝瑙ｅ喅璁 绠楁晥鐜囥€佹牳鐨勯€夋嫨绛夛級 2 鎺ㄧ悊鍜屽喅绛 Inference and Decision 3 浠庣粺璁″statistics鐨勮鐐圭湅 飪 鐩殑鏄緱鍒版槧灏勶細聿 鈫泊 飩 鏁版嵁鍒嗗竷饊仢(聿? 飩 鍏堥獙鍒嗗竷 prior distribution 饊仢(聿? 飦 a priori : Knowable without appeal to particular experience 飦 a priori distribution: special meaning, do not misuse 飩 鑱斿悎joint鍒嗗竷饊仢(聿?聿? 飩 绫绘潯浠跺垎甯凁€仢(聿硘肀 =肀? 飩 鍚庨獙鍒嗗竷posterior distribution 饊仢(肀 =肀東饊倷) 4 濡備綍琛ㄧず/浼拌姒傜巼瀵嗗害 飪糚arametric: 鍋囪PDF鏈嶄粠鏌愮鍑芥暟褰㈠紡 functional form 飩 濡傞珮鏂垎甯冪殑鍑芥暟褰㈠紡锛屽叾鍖呭惈鑻ュ共鍙傛暟 飩 褰撴寚瀹氬叾鎵€鏈夊弬鏁板€间箣鍚庯紝PDF灏卞畬鍏ㄧ‘瀹 飩 涓嶅悓鐨勬鐜囧垎甯冪敱涓嶅悓鐨勫弬鏁板€煎喅瀹 飩 浼拌PDF灏辨槸浼拌鍙傛暟parameter estimation 锛 飩 鎵€浠ュ彨鍙傛暟浼拌 5 闈炲弬鏁颁及璁 飪 Non-parametric: 涓嶅亣璁綪DF鏄换浣曞凡鐭ュ舰寮忕殑鍑芥暟 飩 浠庣洿瑙備笂鏇村悎鐞 飩 閭ｄ箞锛屽浣曚及璁★紵 飦 浣跨敤璁粌鏁版嵁鐩存帴浼拌绌洪棿涓换鎰忕偣鐨勫瘑搴 飦 饊仢(饊倷|饊€? 鈥撳悗闈㈠叿浣撹 飪 闈炲弬鏁颁笉浠ｈ〃鏃犲弬鏁帮紒 飩 瀹為檯涓婃槸鍙互鍏佽鏃犵┓澶氱殑鍙傛暟 飩 鑰屽弬鏁颁及璁＄殑鍙傛暟涓暟鏄湁闄愮殑 飪 鍙傛暟浼拌涓庨潪鍙傛暟浼拌 飩 Parametric and non-parametric estimate 飩 鍙兘鍙傛暟鍖栦及璁′笌闈炲弬鏁板寲浼拌鏇村垏鍚堝叾鍚箟 6 鎺ㄧ悊涓庡喅绛杋nference decision 飪 鐢熸垚妯″瀷鍜屽垽鍒ā鍨 飩 Generative (probabilistic) models: 浼拌 饊仢 饊倷 肀 = 肀 鍜岎€仢(饊倷) 飦 鐒跺悗鐢ㄨ礉鍙舵柉瀹氱悊姹傪€仢 肀 = 肀 饊倷 飩 Discriminative (probabilistic) m