Topic 1
Topic 1
Topic 1
1.ROSENBLATT PERCEPTRON:
topic-1 1
topic-1 2
topic-1 3
topic-1 4
Linear separability is a property of a dataset in which the samples can be separated
by a single line or hyperplane in a multi-dimensional space.
This means that given a set of data points, it is possible to draw a straight line
or a hyperplane that separates the points into two distinct classes or categories.
If the data is linearly separable, then a single layer feedforward neural network
with a linear activation function can be used to solve the problem.
However, if the data is not linearly separable, a multi-layer neural network with
non-linear activation functions must be used to learn a more complex boundary
that can separate the classes.
topic-1 5
In summary, linear separability is an important concept in the design of neural
networks, as it determines the complexity of the architecture that must be used
to solve a classification problem.
The Euclidean norm, also known as the L2 norm, is a measure of the length or
magnitude of a vector in a Euclidean space.
In the context of machine learning and neural networks, the Euclidean norm is
often used as a regularization term in optimization problems.
topic-1 6
This helps to prevent overfitting and improve the generalization performance of
the model by adding a penalty for large weights.
The Euclidean norm is also used to measure the similarity between two vectors,
as the Euclidean distance between two vectors is equal to the Euclidean norm
of the difference between the vectors.
It states that for any two vectors u and v in a Euclidean space, the dot product
of the vectors is always less than or equal to the product of the Euclidean
norms of the vectors. The inequality can be expressed as follows:
Where "." represents the dot product, and "|| ||" represents the Euclidean norm
of a vector.
topic-1 7
topic-1 8
topic-1 9
topic-1 10
topic-1 11
topic-1 12
The covariance matrix C is nondiagonal, which means that the samples drawn from
classes and are correlated. It is assumed that C is nonsingular, so that its inverse
topic-1 13
matrix C1 exists.
If a matrix has an inverse, then it means that it can be used to solve linear
equations and represents a one-to-one mapping from R^n to R^n.
A singular matrix, on the other hand, has a determinant of zero and does not
have an inverse, meaning it cannot be used to solve linear equations and
represents a many-to-one mapping from R^n to R^n.
topic-1 14
where y is the class label, x is the data, P(y | x) is the probability of y given x, P(x |
y) is the probability of x given y, P(y) is the prior probability of y, and P(x) is the
probability of x
Bayes' classifier uses this formula to calculate the probability of each class given
the data, and then chooses the class with the highest probability.
topic-1 15
The perceptron, on the other hand, does not assume any particular functional form
for the decision boundary, but rather learns it directly from the training data by
adjusting the weights of the input features.
The number of parameters in the perceptron is not fixed in advance, but instead
increases or decreases as the number of input features changes.
Once the parameters are estimated, the Bayes classifier uses Bayes' theorem
to compute the posterior probability of each class given the input data, and
selects the class with the highest posterior probability.
The perceptron is a simple and adaptive algorithm for binary classification that
learns a linear decision boundary that separates the positive and negative examples in
the input feature space.
The perceptron update and convergence procedure is simple and adaptive in
several ways:
topic-1 16
1. Update rule: The perceptron algorithm updates the weight vector based on
whether the prediction made by the current weights is correct or incorrect.
Specifically, if the perceptron predicts the correct label for a training example, it
does not update the weights. If the prediction is incorrect, it updates the weights
in the direction of the misclassified example. This simple update rule is efficient
and easy to understand, making the perceptron algorithm easy to implement
and apply.
3. Adaptive: The perceptron algorithm is adaptive because it can learn from new
examples as they arrive. Unlike batch algorithms that require all training
examples to be seen at once, the perceptron updates the weight vector
incrementally as each example is presented. This makes the perceptron well-
suited for online learning tasks, where new examples arrive over time and the
model needs to adapt to changing data distributions.
topic-1 17
topic-1 18