Flashcard Set Preview
| Side A | Side B | ||
| 1 |
Name two neural network topologies
|
Feed forward topology (data flow is in the forward direction only; no feedback) and Recurrent...
|
|
| 2 |
Examples of feed forward networks
|
The perceptron and adaline
|
|
| 3 |
What is supervised learning?
|
Training by providing an input/output set and using the actual output of the network to calculate...
|
|
| 4 |
What is unsupervised learning?
|
a kind of learning where the system is trained to discover statistically salient features of...
|
|
| 5 |
what is adaptive resonance theory?
|
no idea...
|
|
| 6 |
What is neural computing?
|
Neural computing is the study of networks of adaptable units which, through a process oflearning...
|
|
| 7 |
formal definition of the associative memory problem
|
store a set of k patterns such that when presented with a new pattern Pi the network responds...
|
|
| 8 |
Stability criterion for associative networks
|
The network enters a state where for all j the sign of wjxij = the sign of the pattern imposed... |
|
| 9 |
What property relating to errors do we require of neural networks
|
Insensitivity to small errors in the input pattern.
|
|
| 10 |
What are attractors?
|
These are stored patterns in the configuration space and they attract the network to one of...
|
|
| 11 |
State Oja's rule for unsupervised learning
|
It's based on Hebbian learning, which states that the weights are modified by a term proportional... |
|
| 12 |
In unsupervised, competitive learning what determines the winning unit? *hint: grandmother...
|
The unit with the largest net input is the winner
|
|
| 13 |
In Oja's rule what's the limiting term and what does it do?
|
The limiting term is rV2wj. This ensures that |wj| is <= 1.
|
|
| 14 |
state the competitive learning rule
|
change in weight = r(input - weight of winning input)
|
|
| 15 |
state the Kohonen learning rule for neural networks
|
change in weight = r * NF(i, i*)(xj - wij ) for all ijNF = neighbourhood functionNF(i, i*)...
|
|
| 16 |
State the differences between the Kohonen rule and the competitive learning rule
|
With the competitive rule only the inputs to the winning units are updated but with the...
|
|
| 17 |
Name the condition for the strengthning of synaptic connection between two cells
|
This occurs when both cells are simulstaneously activated.
|
|
| 18 |
what's the significance of the cost function?
|
it determines the proximity of the output of the network to the desired state.
|
|
| 19 |
what's the derivative of the sigmoid function?
|
image
|
|
| 20 |
what is the derivative of the tanh(beta*h) function
|
image
|
|
| 21 |
for MLPs and backprop and perhaps simple perceptrons how is the change in the weights determined?
|
determined by the partial differential of the cost function with respect to the weight vector.
|
|
| 22 |
for MCPs the new weight equals..?
|
weight_new = weight_old + change
|
No comments yet! Be the first to add a comment below!
Please login to post comments.
After login, we will forward you back to this flashcard.