18CSE479T - Statistical Machine Learning UNIT 4 & 5 (4 MARKS & MCQs)
4M:
Forward propagation:
Feedforward propagation is a fundamental step in training neural networks and other machine learning models. It involves computing the output of the model given an input example.
Input layer:
Receives the features of your data
Hidden layers:
Apply activation function to their weighted inputs
Weights and biases:
Each connection between neurons has an associated weight
Each neuron has a bias
Output layer:
Produces the final predictions
Forward propagation steps:
Compute weighted sum of inputs
Apply an activation function
Pass the values to next layer
Repeat the process for all hidden layers
Finally, compute the output
Feedforward Neural Networks:
Architecture:
Consists of three layers: input, hidden and output
No feedback loops; information flows forward
Purpose:
Approximates functions by mapping inputs to predictions
Learn weights and biases during training
Neurons and weights:
Layers contain neurons
Weights define connections between neurons
Applications:
Search engines
Machine translation
Object detection
FOR THIS QUESTION, YOU CAN WRITE THE SAME ANSWER FOR FORWARD PROPAGATION MINUS THE FP STEPS
Importance of dimensionality reduction:
Dimensionality reduction:
Process of reducing the number of features in a dataset while retaining as much important information as possible
Transforms high dimensional data into a lower dimensional space
Importance of dimensionality reduction:
Curse of dimensionality
Avoid Overfitting
Reducing Computational cost
Visualization
Removing irrelevant features
Approaches to dimensionality reduction:
Feature selection
Feature extraction
Advantages and disadvantages of PCA reduction
PCA:
Principal Component Analysis is a popular technique used for dimensionality reduction
It aims to reduce the number of features in a dataset while retaining essential information
Advantages:
Dimensionality reduction
Interpretability
Noise reduction
Multicollinearity removal
Disadvantages:
Data loss
Linear correlation (may not capture complex relationships)
How does forward propagation contribute to the training process of neural network?
Forward propagation:
Initial step in training a neural network
Involves applying a series of weights and biases to input data and passing the result through an activation function
Process:
Input data flow
Activation function
Contribution:
Forward propagation helps compute the predicted output based on the current weights and biases
The predicted output is then compared to the actual output to calculate the prediction error
This error is used during back propagation to adjust the weights and improve the model
Elbow method:
Technique used to determine the optimal number of clusters in a K-means clustering algorithm
Helps find the right balance between model complexity and data representation
We plot a graph of K versus WCSS where the optimal K value is where the graph starts to look like a straight line
Beyond the elbow point, adding more clusters doesn’t significantly improve the model
WCSS represents the sum of squared distances between data points and their cluster centroids (within cluster sum of squares)
K-Means:
K means clustering is an unsupervised learning technique that groups similar data points into clusters
K-means aims to partition n observations into k clusters
Each observation belongs to a cluster with the nearest cluster centroid
Process:
Assign each data point to the closest centroid
Calculate the variance and place a new centroid for each cluster
Repeat steps until no reassignment occurs
Same advantages and disadvantages as PCA
MCQs:
1. Identify algorithm that helps in feature reduction?
PCA (Prinicipal Component Analysis)
t-SNE
LLE
SVM
2. Which of the following is not a disadvantage of Support Vector Machines?
Overfitting
Handling non-linear data
High computational cost
Sensitive to outliers
3. What is the objective of Support Vector Machines?
To find the hyperplane that maximally separates the classes
To find the hyperplane that minimally separates the classes
To find the hyperplane that equally separates the classes
4. A 3-input neuron is trained to output a zero when the input is 110 and a one when the input is 111. After generalization, the output will be zero when and only when the input is? [any with less than 3 ones]
000
100
010
110
5. When using the basic SVM, which of the following types of classification can be used?
Binary classification
Multi-class classification
Regression
6. A 4-input neuron has weights 1, 2, 3 and 4. The transfer function is linear with the constant of proportionality being equal to 2. The inputs are 4, 10, 5 and 20 respectively. What will be the output?
108
238
432
544
7. Which of the following technique works only for square matrices?
Eigenvalue decomposition
Singular value decomposition
PCA
LLE
8. The Eigenvalues of a matrix are
Answer: Scalars that represent how much the matrix stretches or compresses a direction
Vectors that represent the direction of the matrix
Scalars that represent how much the matrix stretches or compresses a direction
Matrices that represent the inverse of the original matrix
9. (Duplicate question)
10. (Duplicate question)
11. How is the initial cluster centroid chosen in k-means clustering algorithm?
Randomly chosen from the data points
Chosen as the mean of the entire data
Chosen as the median of the entire data
12. (Duplicate question)
13. What is the objective of k-means clustering algorithm?
To group similar data points into clusters
To separate dissimilar data points
To find the hyperplane that maximally separates the classes
14. What is the significance of eigenvalues in PCA?
They represent the amount of variance explained by each principal component
They represent the amount of covariance between the components
They represent the amount of correlation between the components
Comments
Post a Comment