Dense

• Weighted LinearRegression
• Forward
  • $z=W\cdot x+b$ , $y=g(z)$
• Backward
  • $\delta =g^{\prime }(z)\circ {\nabla }_{y}E$
  • ${\nabla }_{W}E=\delta \cdot x^T$ , ${\nabla }_{b}E=\delta$
  • ${\nabla }_{x}E=W^T\cdot \delta$