www-ai.cs.tu-dortmund.de/de/LEHRE/FACHPROJEKT/WS1718/folien4.pdf
DeepLearning on FPGAs - Introduction to FPGAs
153 11
∗ −0.5 1
1 −0.5 =
180 · 1− 80 · 0.5− 20 · 0.5 + 120 · 1 = 25010 · 1− 120 · 0.5− 45 · 0.5 + 140 · 1 = 67170 · 1− 20 · 0.5− 122 · 0.5 + 39 · 1 = 138153 · 1− 11 · 0.5− 70 · 0.5 + 200 · 1 = 255
250 67 [...] “clusters” c1, . . . , cK in data x1, x2, . . . , xN ∈ R Mathematically: argminc1,...,cK
∑K k=1
∑N i=1(xi − ck)2
Idea Iterative algorithm
1: for i = 1, 2, ...,K do 2: c[i] = random without replacement(x1, ... [...] different NN structure (2)
Recall
δ (l−1) j =
∂h(y (l−1) i )
∂y (l−1) i
M(l)∑
k=1
δ (l) k w
(l) j,k
For CPUs Use hash function h
δ (l−1) j =
∂h(y (l−1) i )
∂y (l−1) i
M(l)∑
k=1
δ (l) k w
(l) φ(j,k)
For FPGAs
Compute …