17 lines
519 B
Plaintext
17 lines
519 B
Plaintext
The idea is always the same:
|
|
<l2st>
|
|
Define complicated model to learn (often millions of parameters)
|
|
Define loss function that this model should minimize (example: $\sum_i (y_i-f(x_i))^2$)
|
|
Find parameters that minimize the loss (->Backpropagation)
|
|
</l2st>
|
|
Usually Neural Networks:
|
|
<l2st>
|
|
$f(x)=f_n(x)=activation(A_n\cdot f_{n-1}(x)+b_n)$
|
|
$f_0(x)=x$
|
|
</l2st>
|
|
Powerful, as you can show that when there are 3 Layers+ (and infinitely sized matrices), you can approximate any function
|
|
->So a model becomes a loss function
|
|
|
|
|
|
|