Introduction

What is Association?

Type of learning where new response becomes linked to particular thing or event that evokes a specific
functional reaction in an organ or tissue.

Or

A connection or cooperative link between similar groups of function which are made to achieve same
goal.

Example:

Human identify objects based on their characteristics like shape, color, weight which they have seen
earlier. When we say an apple, we imagine red color spherical object with shape we can hold in our
hand. But how machine will memorize an apple? Here is explanation from table.

Diameter Weight Red Green Blue Name

2.8 10.1 172 85 3 Grape
3.9 6.2 166 78 1 Grape
11.2 108.5 157 98 2 Apple
10.2 120.3 161 78 2 Orange

For any object, machine will have their characteristics saved with reference to their size, weight, color
combination. As per table, if weight, diameter and color combinations are matched then that object is
what mentioned in last column.

Associative Learning

Associative learning is the process by which a person or animal learns an association between two
stimuli or events.In classical conditioning, a previously neutral stimulus is repeatedly paired with a
reflex-eliciting stimulus until eventually the neutral stimulus elicits a response on its own. In operant
conditioning, a behavior that is reinforced or punished in the presence of a stimulus becomes more or
less likely to occur in the presence of that stimulus.

Hopfield Network
Artificial Neural Network is designed in such a way that it will observe the ability of brain to recognize
and discriminate. Millions of signals continuously harmonize and oscillate; these networks are functional
architecture which underlie in our stress, happiness, dreams, hopes and so on. Integration of these
networks in human brain make it difficult to study where, when, how or even if computation in the
forms we are familiar with occurs. We’ll begin with conceptual background, then move to
implementation.

Hopfield networks are form of Recurrent Artificial Neural Networks (RNNs), first described by John
Hopfield in his 1982.

Hopfield’s unique network architecture was based on a physics model that explains the emergent
behavior of the magnetic fields produced by ferromagnetic materials.

Two types of HN:

Discrete Hopfield Network: It is a fully interconnected neural network where each unit is connected to
every other unit. It behaves in a discrete manner, i.e. it gives finite distinct output, generally of two
types:

Binary (0/1)
Bipolar (-1/1)

Continuous Hopfield Network: Unlike the discrete hopfield networks, here the time parameter is
treated as a continuous variable. So, instead of getting binary/bipolar outputs, we can obtain values that
lie between 0 and 1

where,

vi = g(ui)

vi = output from the continuous hopfield network

ui = internal activity of a node in continuous hopfield network.

Each neuron in the network has three qualities to consider:

connections to other neurons — each neuron in the network is connected to all other neurons, and
each connection has a unique strength, or weight, analogous to the strength of a synapse. These
connection strengths are stored inside a matrix of weights.

activation — this is computed via the net input from other neurons and their respective connection
weights, loosely analogous to the membrane potential of a neuron. The activation takes on a single
scalar value.

bipolar state — this is the output of the neuron, computed using the neuron’s activation and a
thresholding function, analogous to a neuron’s ‘firing state.’ In this case, -1 and +1.
Structure & Architecture

 Each neuron has an inverting and a non-inverting output.

 Being fully connected, the output of each neuron is an input to all other neurons but not self.

[ x1 , x2 , ... , xn ] -> Input to the n given neurons.

[ y1 , y2 , ... , yn ] -> Output obtained from the n given neurons
Wij -> weight associated with the connection between the ith and the jth neuron.

Weights assigned can be of two types, binary (0 and 1) or bipolar (weights have no self connection)

Step 1 - Initialize weights (wij to store patterns (using training algorithm).

Step 2 - For each input vector yi, perform steps 3-7.

Step 3 - Make initial activators of the network equal to the external input vector x.

Yi=Xi : (for i= 1 to N)
Step 4 - For each vector yi, perform steps 5-7.

Step 5 - Calculate the total input of the network yin using the equation given below.

Yinput (i) = Xi + ∑ j [ YjWji]

Step 6 - Apply activation over the total input to calculate the output as per the equation given below:

For Yi = 1 if Yin > Ɵi,

For Yi = Yi if Yin = Ɵi,
For Yi = 0 if Yin < Ɵi
(where θi (threshold) and is normally taken as 0)

Step 7 - Now feedback the obtained output yi to all other units. Thus, the activation vectors are updated.

Step 8 - Test the network for convergence.

Simulated Annealing

Simulated annealing is a technique used in AI to find solutions to optimization problems. In other words,
simulated annealing can be used to find solutions to optimization problems by slowly changing the
values of the variables in the problem until a solution is found. It allows for small changes to be made to
the solution, which means that it can escape from local minima and find the global optimum. Simulated
annealing is not a guaranteed method of finding the best solution to an optimization problem, but it is a
powerful tool that can be used to find good solutions in many cases. Simulated annealing works by
starting with a random solution and then slowly improving it over time. The key is to not get stuck in a
local optimum, which can happen if the search moves too slowly.

How does it work?

We need to provide an initial solution so the algorithm knows where to start. This can be done in two
ways:
(1) using prior knowledge about the problem to input a good starting point and
(2) generating a random solution.

1. Move all points 0 or 1 units in a random direction

2. Shift input elements randomly
3. Swap random elements in input sequence
4. Permute input sequence
5. Partition input sequence into a random number of segments and permute segments

Benefits of using Simulated Annealing?

1. The ability to find global optima.

2. The ability to escape from local optima.
3. The ability to handle constraints.
4. The ability to handle noisy data.
5. The ability to handle discontinuities.
6. The ability to find solutions in a fraction of the time required by other methods.
7. The ability to find solutions to problems that are difficult or impossible to solve using other methods.

Drawbacks of using Simulated Annealing?

It can be slow and may not always find the best solution. Additionally, it can be difficult to tune the
parameters of the algorithm, which can lead to sub-optimal results.

Applications of Simulated Annealing

travelling salesman problem, the knapsack problem, and the satisfiability problem. It has also been used
in image recognition

Boltzmann machine and Boltzmann learning

What is a Boltzmann Machine?

A Boltzmann machine is a neural network of symmetrically connected nodes that make their own
decisions whether to activate. It uses a straightforward stochastic learning algorithm to discover
“interesting” features that represent complex patterns in the database.
This is categorized under “unsupervised deep learning”.

Note: stochastic means having a random probability distribution or pattern that may be analyzed
statistically but may not be predicted precisely.

The main purpose of Boltzmann Machine is to optimize the solution of a problem.

Types of Boltzmann Machines:

1. Restricted Boltzmann Machines (RBMs)

2. Deep Belief Networks (DBNs)
3. Deep Boltzmann Machines (DBMs)

There are two types of nodes in the Boltzmann Machine

Visible nodes: those nodes which we can represent and do measure
Hidden nodes: those nodes which we cannot represent or do not measure.
Irrespective of these different types, Boltzmann machine consider them same as machine work on single
system.
Best example we can give is recommendations of YouTube videos.
Considering architecture of ANN, we are aware about input layer where input nodes are known. Here
Visible nodes are nodes can be considered as recommendation seen on screen. For example; first video
is related to sports, second can be news, third can be a video of song, a next can be any interview (these
can be consider as Visible nodes). Here in this scenario, we keep scrolling until we find something
interesting. Once we get it, let’s say video of song from your favorite artist. Then now onwards, machine
will recommend you other videos of same artist. In this situation video from your favorite can be
considered as hidden node.

Boltzmann learning

Boltzmann learning is similar to error-correction learning and is used during supervised training.
In this algorithm, the state of each individual neuron, in addition to the system output, are taken into
account. In this respect, the Boltzmann learning rule is significantly slower than the error-correction
learning rule.

Note: Neural networks that use Boltzmann learning are called Boltzmann machines.

Boltzmann learning is similar to an error-correction learning rule, in that an error signal is used to train
the system in each iteration. However, instead of a direct difference between the result value and the
desired value, we take the difference between the probability distributions of the system.

Boltzmann machine has a set of units Ui and Uj and has bi-directional connections on them.

 We are considering the fixed weight say wij.

 wij ≠ 0 if Ui and Uj are connected.
 There also exists a symmetry in weighted interconnection, i.e. wij = wji.
 wii also exists, i.e. there would be the self-connection between units.
 For any unit Ui, its state ui would be either 1 or 0.

The main objective of Boltzmann Machine is to maximize the Consensus Function CF which can be given
by the following relation;

CF = ∑I∑j≤I WijUiUj

State transition diagram and false minima problem

What is State Transition?

A State can be consider as current condition or situation of system; this can be active, inactive, in
progress, waiting, stopped and so many states.

Transition is the process or a period of changing state from one condition to another.

A State Transition diagram is a type of diagram used to describe the behavior of systems. State diagrams
require that the system described is composed of a finite number of states.

Four Parts of State Transition Diagram

1) States that the software might get

1st Try

2) Transition from one state to another state

3) Events that origin from / to transition

Incorrect Pin

4) Action can be result from Transition

Access Granted
State Transition Table
In state transition diagram the states are shown in boxed texts, and the transition is represented by
arrows. It is also called State Chart or Graph.

In state transition table all the states are listed on the left side, and the events are described on the top.
Each cell in the table represents the state of the system after the event has occurred. It is also called
State Table.

Example of State Transition Diagram

Let’s consider an ATM system function where if the user enters the invalid password three times the
account will be locked.
Incorrect Pin Incorrect Pin Incorrect Pin

sd
Start 1st Try 2nd Try 3rd Try

Correct Pin Incorrect Pin

Access Granted Account Blocked

In this system, if the user enters a valid password in any of the first three attempts the user will be
logged in successfully. If the user enters the invalid password in the first or second try, the user will be
asked to re-enter the password. And finally, if the user enters incorrect password 3rd time, the account
will be blocked.

In the diagram whenever the user enters the correct PIN he is moved to Access granted state, and if he
enters the wrong password he is moved to next try and if he does the same for the 3rd time the account
blocked state is reached.

State Transition Table

States Correct PIN Incorrect PIN

S1) Start S5 S2
S2) 1st attempt S5 S3
S3) 2nd attempt S5 S4
S4) 3rd attempt S5 S6
S5) Access Granted – –
S6) Account blocked – –
False Minima problem

Minima is the minimum value of function within the given set of ranges.

Note: Maxima and minima are the peaks and valleys in the curve of a function respectively. Maxima will
be the highest point on the curve within the given range and minima would be the lowest point on the
curve.

In the image given below, we can see various peaks and valleys in the graph.
At x = a and at x = 0, we get maximum values of the function,
and at x = b and x = c, we get minimum values of the function.
All the peaks are the maxima and the valleys are the minima.

What is False Minima Problem?

False minima, also known as local minima, is a common problem that can occur in optimization
algorithms used in artificial intelligence. In optimization, the goal is to find the minimum or maximum
value of a function. However, some functions can have multiple local minima, which are points where
the function has a lower value than all of its neighboring points, but are not the absolute minimum of
the function.

The false minima problem occurs when an optimization algorithm gets stuck in one of these local
minima instead of finding the global minimum. This can happen when the algorithm follows a path that
leads to a local minimum, but is unable to escape and find the global minimum. This can result in
suboptimal solutions and reduce the effectiveness of the algorithm.

There are several techniques used in artificial intelligence to address the false minima problem,
including:
1. Initialization: The initial values of the parameters can have a significant impact on the optimization
process. By selecting initial values that are more likely to lead to the global minimum, the chances of
getting stuck in a local minimum can be reduced.
2. Exploration vs. exploitation: Some optimization algorithms balance exploration, which involves
searching for new solutions, and exploitation, which involves refining known solutions. By balancing
exploration and exploitation, the algorithm can be more effective at finding the global minimum.
3. Stochastic optimization: Stochastic optimization involves introducing randomness into the
optimization process. By adding randomness, the algorithm can explore different regions of the
search space and reduce the chances of getting stuck in a local minimum.

Stochastic Update

Stochastic refers to a variable process where the outcome involves some randomness and has some
uncertainty.

Stochastic update is a technique used in machine learning and artificial intelligence to update the
parameters of a model based on a random subset of the training data, rather than the entire dataset.
Stochastic update is a form of stochastic gradient descent, which is a widely used optimization
technique in machine learning.

In stochastic update, the model parameters are updated based on the error or loss computed on a
randomly selected subset of the training data, also known as a mini-batch. The mini-batch size is
typically much smaller than the full training set, and the random selection of data points helps to
introduce randomness and prevent the optimization process from getting stuck in local minima.

Stochastic update has several advantages over batch update, which involves updating the model
parameters based on the full training set. These advantages include:

1. Faster convergence: Stochastic update can converge faster than batch update, especially when the
dataset is large, as the model parameters are updated more frequently.
2. More robustness: Stochastic update can be more robust to noisy or irrelevant data points, as they
are less likely to have a significant impact on the model parameters.
3. Reduced memory requirements: Stochastic update requires less memory than batch update, as it
only needs to store the mini-batch instead of the entire training set.

However, stochastic update can also have some disadvantages, including:

1. Higher variance: Stochastic update can introduce higher variance in the optimization process, as the
model parameters are updated based on a random subset of the training data.
2. Slower convergence at the end: At the end of the training process, stochastic update can converge
slower than batch update, as the updates become less frequent and more noisy.

Popular examples of stochastic optimization algorithms are: Simulated Annealing, Genetic Algorithm,
Particle Swarm Optimization
Basic functional units of ANN for pattern recognition tasks:

Pattern association

Pattern association is the process of memorizing input-output patterns in a heteroassociative network

architecture, or input patterns only in anautoassociative network, in order to recall the patterns when a
new input pattern is presented. It is not required that the new input pattern be exactly the same as one
that is memorized. It can be different, but similar to it. Three exemplar patterns are shown in figure.

After memorizing them in a system, a new pattern is presented, a corrupted variant of the pattern 3. An
associative memory system should associate this pattern with one of the memorized patterns. This is a
task for an autoassociative network architecture.

Pattern classification and pattern mapping tasks

Before searching for a pattern there are some certain steps and the first one is to collect the data from
the real world.
The collected data needs to be filtered and pre-processed so that system can extract the features from
the data. This data can be divided into general category like audio, number, image, text which can be in
the form of traffic signal images, human faces, handwritten letters, and the DNA sequences. These need
to be converted to machine understandable language like 0 and 1.

In classification, the algorithm assigns labels to data based on the predefined features. This is an
example of supervised learning.

What is Pattern Classification?

Pattern classification is a fundamental problem in artificial intelligence that involves assigning objects or
data points to predefined categories or classes. The goal of pattern classification is to develop an
algorithm or model that can accurately predict the class of a new data point based on its features or
attributes.
In pattern classification, the algorithm or model is typically trained on a labeled dataset, where each
data point is associated with a known class label. The model learns to recognize patterns in the data that
are characteristic of each class and uses these patterns to classify new, unseen data points.

There are several techniques used in artificial intelligence for pattern classification, including:

1. Supervised learning: This involves training a model on labeled data, where the class labels are
known. The model learns to recognize patterns in the data that are characteristic of each class and
uses these patterns to classify new, unseen data points.

2. Unsupervised learning: This involves training a model on unlabeled data, where the class labels are
not known. The model learns to identify patterns in the data without any prior knowledge of the
class labels.

3. Semi-supervised learning: This involves training a model on a combination of labeled and unlabeled
data. The model learns to identify patterns in the labeled data and uses this knowledge to classify
the unlabeled data.
Pattern mapping tasks

Pattern mapping is a common task in artificial intelligence (AI) and involves identifying and mapping
patterns within data. There are several techniques used in AI to perform pattern mapping tasks,
including:

1. Machine learning: This involves training an AI model on a large dataset and then using the model to
identify patterns in new data.

2. Deep learning: This is a subset of machine learning that uses neural networks to identify patterns in
data. Deep learning is particularly useful for tasks that involve image recognition and natural
language processing.

3. Clustering: This involves grouping data points based on their similarities. Clustering algorithms can
be used to identify patterns in large datasets.

4. Association rule mining: This involves identifying relationships between items in a dataset.
Association rule mining is commonly used in market basket analysis to identify products that are
frequently purchased together.

5. Decision trees: This involves breaking down a dataset into smaller subsets based on specific criteria.
Decision trees are commonly used in classification tasks to identify patterns that distinguish
between different classes of data.
Overall, pattern mapping tasks in AI involve using algorithms and techniques to identify and extract
meaningful patterns from large datasets. These patterns can then be used to make predictions or inform
decision-making processes in various domains.