Why KDAC? A general activation function for knowledge discovery

Deep learning oriented named entity recognition (DNER) has gradually become the paradigm of knowledge discovery, which greatly promotes domain intelligence. However, activation function DNER fails to treat gradient vanishing, no negative output or non-differentiable existence, may impede exploration due omission and incomplete representation latent semantic. To break through dilemma, we present a novel termed KDAC. Detailly, KDAC is an aggregation with multiple conversion modes. The backbone interaction between exponent linearity, both ends are extended adaptive linear divergence, can surmount vanishing output. Crucially, points be alerted eliminated by approximate smoothing algorithm. series brilliant properties, such as nonlinear, stable near-linear transformation derivative, well dynamic style, etc. We perform experiments based on BERT-BiLSTM-CNN-CRF model six benchmark datasets containing different knowledge, Weibo, Clinical, E-commerce, Resume, HAZOP People's daily. evaluation results show that advanced effective, provide more generalized stimulate performance DNER. hope exploited promising devote itself construction knowledge.