摘要
本文提出一种基于Σ-Δ调制的比特流Tangent-Sigmoid(Tan-Sig)函数的数字电路实现方法.主要是通过修改反馈系数构建Σ-Δ限幅放大调制器,并以此组合实现对Tan-Sig函数的逼近.根据上述方法,在现场可编程门阵列上设计了具有非线性激活函数的比特流神经元以及复杂的比特流前馈型人工神经网络.最后利用该网络实现了在笛卡尔坐标系下的3-D空间判别.
A digital hardware implementation of Tangent Sigrnoid （Tan-Sig） activation function based on sigma-delta modu- lated bit-streams is proposed. Through the change of feedback coefficient in the framework of traditional sigma-delta modulator, a new limiting amplifier modulator （LAM） is fabricated, and the function approximation to Tan-Sig was achieved by the combination of several LAMs with different coefficients. Meanwhile, the bit-stream neurons with Tan-Sig activation function and the whole feed- forward artificial neural networks are implemented on field programmable gate array （FPGA）. And the 3-D space classification problem in Cartesian coordinate system was solved by the neural networks presented.
出处
《电子学报》
EI
CAS
CSCD
北大核心
2015年第5期862-869,共8页
Acta Electronica Sinica
基金
国家自然科学基金(No.60576028)
