www-ai.cs.tu-dortmund.de/LEHRE/FACHPROJEKT/WS1617/folien3.pdf
DeepLearning on FPGAs - Artificial Neural Networks: Backpropagation and more
− α · ∇ŵ`(D, ŵ)
Loss function:
`(D, ŵ) =
√√√√ 1
N
N∑
i=1
( yi − f̂(~xi)
)2
DeepLearning on FPGAs 6
MLP: Learning
Obviously: We need to learn the weights w (l) i,j and bias b
(l) j
So far: We intuitively [...] − α · ∇ŵ`(D, ŵ)
Loss function:
`(D, ŵ) =
√√√√ 1
N
N∑
i=1
( yi − f̂(~xi)
)2
DeepLearning on FPGAs 6
MLP: Learning
Obviously: We need to learn the weights w (l) i,j and bias b
(l) j
So far: We intuitively [...] − α · ∇ŵ`(D, ŵ)
Loss function:
`(D, ŵ) =
√√√√ 1
N
N∑
i=1
( yi − f̂(~xi)
)2
DeepLearning on FPGAs 6
MLP: Learning (2)
`(D, ŵ) =
√√√√ 1
N
N∑
i=1
( yi − f̂(~xi)
)2
Observation: We need to take the derivative …
