Convnd

• $A\ast B$
• Connect a neighbor only to spatial neighborhood → spatial order
  • Some rotation and illumination invariance
• Slide over → Same weights independant of location → less weights
• Subsample after conv
• multiple 2d feature maps
• Similar to Gabor filters after learning
• Some might collapse to 0
• $out(x,y)={\Sigma }_i{\Sigma }_{j}input(x+i,y+j)kernel(i,j)$
• $Z^j={\Sigma }_{i=0}^{l-1}{W}^{ji}\ast X^i+b^{ij}$
• $Y^j=g(Z^j)$
• Output shape : $\frac{({)}_i-f+2p}{s}$
  • If $p=\frac{f-1}{2}$ and $s=1$ then dimensions maintained
• One operation repeated over and over starting with raw
• 
• Padded Conv
• Strided
• Depthwise Separable
• Causal 1D Conv
• Causal Dilated Conv