Differences

This shows you the differences between two versions of the page.

--- data_mining:neural_network:types [2017/01/15 18:15] – angelegt phreazer
+++ data_mining:neural_network:types [2019/10/26 12:09] (current) – ↷ Links adapted because of a move operation phreazer
@@ Line 1: / Line 1: @@
 ====== Types of NNs ======
+===== Perceptron =====
+See [[data_mining:neural_network:perceptron|Perceptron]]
 ===== Feed-forward NN =====
 First layer is input, last layer output, hidden layers inbetween.
@@ Line 8: / Line 9: @@
 Activity of neurons in each layer are non-linear function of the activities in the layer below.
-Recurrent NN
+===== Recurrent NN =====
-- Directed cycles (you can get back to the neurons where you start).
-- Harder to train
+  * Directed cycles (you can get back to the neurons where you start).
+  * Harder to train
+Natural for modeling sequential data:
+  * Equivalent to very deep nets with one hidden layer per time slice, except that they use same weights at every time sclice and get input at very time slice.
+  * Can rember info in the hidden state for a long time.
+See [[data_mining:neural_network:sequences:sequence_learning|Sequence learning]]
+===== Symmetrically connected NN =====
+Like RNN, but connections between units are symmetrical (same weigths in both directions).
+  * Easier to analyze
+  * Restricted: Cannot model cycles
+  * "Hopfield nets" if they have no hidden layer.
-Natural for modelling sequential data:
+===== Convolutional NN =====
-- Equivalent to very deep nets with one hidden layer per time slice, except that they use same weights at every time sclice and get input at very time slice.
+See [[data_mining:neural_network:cnn:cnn|Convolutional neural network]]
-- Can rember info in the hidden state for a long time.