Scribd
Upload
13 views

Tutorial Associative Memory

Local feedback occurs within the hidden layer and produces a simpler, more stable training process than global feedback, which connects the network output to the input. Autoassociative memory finds the stored pattern most similar to the current input pattern, with the output being the same as the input, as in Hopfield networks. Heteroassociative memory retrieves a different pattern than the input, as in Bidirectional Associative Memory (BAM) networks, which have two neuron layers and can store associations up to the size of the smaller layer. BAM was developed to store patterns bidirectionally between its two layers. Recurrent neural networks involve three stages - storage, testing, and retrieval through repetition until stability. The maximum memories stored differs

Uploaded by

ilyakhad08
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
13 views

Tutorial Associative Memory

Local feedback occurs within the hidden layer and produces a simpler, more stable training process than global feedback, which connects the network output to the input. Autoassociative memory finds the stored pattern most similar to the current input pattern, with the output being the same as the input, as in Hopfield networks. Heteroassociative memory retrieves a different pattern than the input, as in Bidirectional Associative Memory (BAM) networks, which have two neuron layers and can store associations up to the size of the smaller layer. BAM was developed to store patterns bidirectionally between its two layers. Recurrent neural networks involve three stages - storage, testing, and retrieval through repetition until stability. The maximum memories stored differs

Uploaded by

ilyakhad08
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

Ilya Khadija Binti Rosli 2022780489 CS2424A

Question 1

Local feedback is when there is an introduction of feedback within the hidden layer.
The training process is simpler and more stable compared to global feedback. On the
other hand, global feedback can be produced by connecting the network output to the
network input. This feedback has a sensitive dependence and unstable training
process.

Question 2

Autoassociative memory is used to find a previously stored pattern that resembles the
current pattern the most, s = t. One example of autoassociative memory is the Hopfield
Network. The output pattern is the same as the input pattern because Hopfield network
is a single-layer network. One feature of Hopfield network is training patterns will
become stable due to weight updates that minimize energy.

Meanwhile, heteroassociative memory is when the retrieved pattern is not the same
as the previously stored patter, s /= t. Bidirectional Associative Memory (BAM) is an
example of heteroassociative memory. BAM consists of two layers of neurons and is
unconditionally stable, so there is no risk of instability when learning any set of
associations. However, the maximum number of stored associations in the BAM must
not be more than the number of neurons in the smaller layer.

Question 3

Bidirectional associative memory (BAM) is an example of heteroassociative memory.


It consists of two layers of neurons that are connected by directional weighted
connection paths. BAM was developed by Kosko in 1988 for the purpose of storing a
set of patterns. BAM network works by passing signals back and forth between the
two layers and repeats the process until all neurons reached equilibrium. BAM can
respond to the inputs in either layer since the weights associated with the network are
bidirectional.
Question 4

There are 3 main stages in the Recurrent Neural Network. The first step is storage
where a set of M fundamental memories must be stored in the n-neuron Hopfield
network and calculate the weight matrix. The second step is testing. We need to test
if the Hopfield network can recall any fundamental memory when presented with it as
an input. The last step is retrieval where an unknown n-dimensional vector or also
known as probe is presented to the network. This step is repeated until a stable state
is achieved.

Question 5

The maximum number of fundamental memories that can be stored in Hopfield


network is limited by Mmax = 0.15n. However, for BAM, the maximum number of
associations stored must not exceed the number of neurons in the smaller layer.

You might also like