Why deeper is great
越深,可以得到越复杂的function;
再浅层网络中,两个neurons产生一种线段;但再深层中
if K is width,H is depth,
we can have at least $K^H$ piesces.
越靠近input的layer越重要
(上图表现各层对noise的敏感度,越往后noise影响越小;
越深,可以得到越复杂的function;
再浅层网络中,两个neurons产生一种线段;但再深层中
if K is width,H is depth,
we can have at least $K^H$ piesces.
越靠近input的layer越重要
(上图表现各层对noise的敏感度,越往后noise影响越小;